1
00:00:03,510 --> 00:00:06,720
Welcome to this episode of the
Award-Winning Best of the Left Podcast.

2
00:00:06,930 --> 00:00:11,010
Today we explore the long history
of technological disruption from the

3
00:00:11,010 --> 00:00:15,900
Industrial Revolution to the rise of
software engineering and ask whether AI is

4
00:00:15,900 --> 00:00:21,180
genuinely different this time, all while
autonomous weapons dissolve accountability

5
00:00:21,180 --> 00:00:25,290
for the militaries using them, and
capitalism prevents AI companies from

6
00:00:25,290 --> 00:00:27,300
maintaining their ethical aspirations.

7
00:00:27,810 --> 00:00:31,110
For those, looking for a quick overview,
the source is providing our top takes

8
00:00:31,110 --> 00:00:35,640
in about 50 minutes today include,
consider this confronting capitalism

9
00:00:35,790 --> 00:00:40,890
today, explained on the media, your
undivided attention and why America.

10
00:00:41,460 --> 00:00:44,670
Then in the additional deeper dives, half
of the show there will be more in four

11
00:00:44,670 --> 00:00:51,360
sections, section A, labor future, section
B warfare, section C, surveillance,

12
00:00:51,510 --> 00:00:53,610
and Section D Corporate Power.

13
00:00:54,060 --> 00:00:56,880
But first, a reminder to check
out our new show solved on the

14
00:00:56,880 --> 00:00:58,110
best of left YouTube channel.

15
00:00:58,290 --> 00:01:00,210
We're really proud of the show
we're making and think you'll

16
00:01:00,210 --> 00:01:01,290
get real value out of it.

17
00:01:01,500 --> 00:01:04,620
Plus you checking it out will help
us find new viewers on YouTube.

18
00:01:04,860 --> 00:01:08,490
So thanks in advance for all of your
views, likes, subscribes, and comments.

19
00:01:08,700 --> 00:01:10,770
That's all on the best
of left YouTube channel.

20
00:01:11,040 --> 00:01:13,740
Linked in the show notes
and now on the show.

21
00:01:19,980 --> 00:01:20,820
Well, let's start with that.

22
00:01:20,820 --> 00:01:23,940
It feels like in the past few weeks
this conversation about AI job

23
00:01:23,940 --> 00:01:26,760
loss has bubbled outta the tech
world and into the mainstream.

24
00:01:26,760 --> 00:01:30,300
I mean, I for one, have gotten
like a half dozen texts from

25
00:01:30,300 --> 00:01:32,940
freaked out friends sending stuff
around like, what's going on here?

26
00:01:33,310 --> 00:01:35,920
Martha, why do you think
this is happening right now?

27
00:01:36,250 --> 00:01:40,690
I think part of it is that there
legitimately has been an acceleration

28
00:01:40,690 --> 00:01:44,710
and what the technology can do over
just the last couple of months.

29
00:01:45,160 --> 00:01:48,400
I think there's also a broader
issue, which is that people are just

30
00:01:48,400 --> 00:01:52,510
incredibly anxious about the economy
right now and where it's going.

31
00:01:52,510 --> 00:01:56,440
If you look at, whe whether
people think they're gonna lose

32
00:01:56,440 --> 00:02:00,040
their job in six months relative
to where the unemployment rate

33
00:02:00,130 --> 00:02:02,470
currently is, they're really negative.

34
00:02:02,470 --> 00:02:02,560
Mm-hmm.

35
00:02:02,890 --> 00:02:06,430
And then of course, when people
are piling on saying that we aren't

36
00:02:06,430 --> 00:02:09,580
going to have any jobs in two
years, that's not going to help.

37
00:02:09,960 --> 00:02:11,730
That that does create some anxiety.

38
00:02:12,150 --> 00:02:15,840
Kelsey, you have been covering
AI since, since chat.

39
00:02:15,840 --> 00:02:18,090
GPT was a twinkle in Sam Altman's eye.

40
00:02:18,150 --> 00:02:23,190
Can you explain to us some
specific examples of what these

41
00:02:23,190 --> 00:02:26,040
models can do now that they couldn't
do a year ago, six months ago?

42
00:02:26,609 --> 00:02:31,500
Yeah, so the biggest launch in the last
six months has been the wide launch

43
00:02:31,590 --> 00:02:35,940
of agent models, which are instead
of just responding to you in a chat

44
00:02:35,940 --> 00:02:37,410
screen, they go out and do things.

45
00:02:37,890 --> 00:02:39,810
The biggest thing they do is code.

46
00:02:40,140 --> 00:02:42,120
They are really remarkably
good at writing code.

47
00:02:42,180 --> 00:02:45,959
A year ago I was hearing from
people, it's nice because you

48
00:02:45,959 --> 00:02:48,690
can like ask it some questions
and it will speed up your coding.

49
00:02:48,780 --> 00:02:51,690
Six months ago I was hearing from
people, it's kind of like having

50
00:02:51,690 --> 00:02:54,870
a grad student that you can ask
to run some analysis for you.

51
00:02:55,140 --> 00:03:00,630
Now I'm hearing from programmers, I
don't really write code anymore, I just

52
00:03:00,630 --> 00:03:04,800
write a spec and then the AI builds
the code that I ask them to build.

53
00:03:04,890 --> 00:03:08,400
So this is a specific thing that
it isn't better and better and

54
00:03:08,400 --> 00:03:10,859
more and more autonomous and that
is a real trend that is happening.

55
00:03:11,250 --> 00:03:11,940
Very much so.

56
00:03:12,025 --> 00:03:16,200
Okay, so Martha, a lot of jobs
are theoretically at risk, but

57
00:03:16,200 --> 00:03:19,079
I wanna start by looking at
what we actually know already.

58
00:03:19,079 --> 00:03:22,410
You and your colleagues at the Budget
lab have been looking at the actual data.

59
00:03:22,410 --> 00:03:26,370
What does it show so far, at
least about AI related job loss?

60
00:03:26,490 --> 00:03:30,510
That there really isn't any,
certainly at the macroeconomic level.

61
00:03:30,959 --> 00:03:34,980
You can look at the technology and
you can see the potential, but you need to

62
00:03:34,980 --> 00:03:40,620
remember that technological related labor
market disruption is not instantaneous.

63
00:03:40,620 --> 00:03:46,170
It happens in the context of IT policies,
company concerns about liability,

64
00:03:46,500 --> 00:03:48,269
broader economic growth trends.

65
00:03:48,630 --> 00:03:51,209
And so I think it's really important
to keep all of those things in mind

66
00:03:51,240 --> 00:03:55,980
when talking about what jobs are
going away, becoming more popular,

67
00:03:56,160 --> 00:03:57,600
and how fast this may happen.

68
00:03:57,870 --> 00:03:59,100
That is what is happening now.

69
00:03:59,130 --> 00:04:01,110
That is what has been
happening over the past year.

70
00:04:01,110 --> 00:04:03,269
That doesn't mean things
couldn't change going forward.

71
00:04:03,269 --> 00:04:06,180
And I know you and your colleagues
have compared this in many ways.

72
00:04:06,704 --> 00:04:08,234
To the Industrial Revolution.

73
00:04:08,295 --> 00:04:10,665
Why is that a good way
to think about this?

74
00:04:11,295 --> 00:04:14,775
I think the thing about the
industrial revolution is you did

75
00:04:14,775 --> 00:04:17,505
see a really big shock to labor.

76
00:04:17,985 --> 00:04:21,404
I think sometimes, you hear
economists talk and you, we will

77
00:04:21,404 --> 00:04:26,325
say, we, the end of technological
change living standards are higher.

78
00:04:26,325 --> 00:04:27,735
We are always better off.

79
00:04:28,455 --> 00:04:32,325
I do wanna emphasize the
period of disruption can be

80
00:04:32,325 --> 00:04:34,725
really, really hard for people.

81
00:04:35,025 --> 00:04:40,005
I am very grateful that I have a job that
is more highly paid and I have higher or

82
00:04:40,005 --> 00:04:43,155
living standards than the weavers did.

83
00:04:43,875 --> 00:04:47,055
I also would not have wanted to be
a weaver during that time period.

84
00:04:47,355 --> 00:04:47,414
Yeah.

85
00:04:47,414 --> 00:04:51,795
Their lives became
really, really hard and I do

86
00:04:51,795 --> 00:04:53,295
think we need to think about.

87
00:04:53,940 --> 00:04:58,410
Is our social safety net prepared
for the type of disruption

88
00:04:58,410 --> 00:05:00,330
that could possibly be coming.

89
00:05:00,870 --> 00:05:04,080
And Kelsey, I know you spend a lot of
time, really optimistically

90
00:05:04,080 --> 00:05:06,450
thinking about worst case scenarios of ai.

91
00:05:06,870 --> 00:05:07,170
Yep.

92
00:05:07,230 --> 00:05:11,520
Where does the Industrial
Revolution type scenario fit there?

93
00:05:11,520 --> 00:05:14,280
Like do you think that is the
high end, worst case scenario?

94
00:05:14,280 --> 00:05:16,770
Or how are you thinking about this right
now when you think about what could

95
00:05:16,770 --> 00:05:18,000
happen over the next five to 10 years?

96
00:05:18,060 --> 00:05:21,780
Yeah, so I think Martha is entirely
right that we have not yet seen

97
00:05:21,780 --> 00:05:24,210
significant AI associated job loss.

98
00:05:24,480 --> 00:05:28,290
I think almost everybody who's worried
about this, what they're doing is not so

99
00:05:28,290 --> 00:05:31,080
much, oh, I've already lost my job to ai.

100
00:05:31,350 --> 00:05:35,310
Most of what people are saying is,
alright, imagine that we learned the new

101
00:05:35,310 --> 00:05:42,270
class of students entering the workforce
in 2028 was going to be 10 million people,

102
00:05:42,330 --> 00:05:45,900
a hundred million people, and all of
them are willing to work for pennies.

103
00:05:46,350 --> 00:05:50,160
Maybe in the long run,
the economy will end up being

104
00:05:50,220 --> 00:05:53,760
much wealthier because of this
huge influx of productive labor.

105
00:05:54,000 --> 00:05:58,440
But as people who will be competing
with that productive labor, we

106
00:05:58,440 --> 00:06:00,270
are not necessarily thrilled.

107
00:06:00,810 --> 00:06:04,770
So if you think that that's where AI
is going, millions and millions of

108
00:06:04,770 --> 00:06:07,860
AI agents that will work basically
for free, then I think people are

109
00:06:07,860 --> 00:06:09,330
completely right to be scared of that.

110
00:06:09,360 --> 00:06:12,870
That will probably in the
absence of very decisive policy,

111
00:06:12,870 --> 00:06:14,850
action, make your life worse.

112
00:06:15,240 --> 00:06:17,400
Both of you are saying in different ways.

113
00:06:18,225 --> 00:06:22,905
A lot of the outcome will depend
on policies, federal government

114
00:06:22,905 --> 00:06:24,585
policy, state government policy.

115
00:06:24,855 --> 00:06:28,305
Are you seeing those policies being
carried out, serious conversations

116
00:06:28,305 --> 00:06:29,805
in any way, shape or form right now?

117
00:06:30,105 --> 00:06:32,985
I think that right now the government
is being caught very flatfooted.

118
00:06:33,315 --> 00:06:37,905
But one thing that we saw with COVID
was that if a shock does hit very

119
00:06:37,905 --> 00:06:42,045
suddenly and affect a huge share
of the American public, Congress

120
00:06:42,045 --> 00:06:44,145
can move into action very fast.

121
00:06:44,385 --> 00:06:46,635
But that's not the only way
this is guaranteed to happen.

122
00:06:46,755 --> 00:06:52,034
And if instead you have much more
creeping job loss and disruptions,

123
00:06:52,515 --> 00:06:56,175
I think it's very easy to imagine
Congress not getting their act together.

124
00:06:56,595 --> 00:07:00,465
I wanna end with this,
tsunami of internet takes that

125
00:07:00,465 --> 00:07:01,725
started this conversation.

126
00:07:01,725 --> 00:07:05,025
A lot of them kind of ended with here's
what you can do as an individual.

127
00:07:05,025 --> 00:07:08,655
And I personally veer between like,
okay, I need to learn a lot more about

128
00:07:08,655 --> 00:07:12,195
how these large language model programs
work and incorporate into my job.

129
00:07:12,195 --> 00:07:14,385
Or I want to go entirely analog.

130
00:07:14,445 --> 00:07:16,395
I hate this all and I'm
just gonna get offline.

131
00:07:16,725 --> 00:07:19,635
Both of you think about this
a lot more than most people.

132
00:07:20,025 --> 00:07:22,335
What is like one thing you
would say to somebody in the

133
00:07:22,335 --> 00:07:24,105
job market right now, Martha?

134
00:07:24,255 --> 00:07:27,675
I mean, one thing frankly I
would say is to take some of

135
00:07:27,675 --> 00:07:31,215
the takes that are circulating
out there with a grain of salt.

136
00:07:31,755 --> 00:07:35,685
People are not thinking enough about
what are the barriers to get things

137
00:07:35,685 --> 00:07:39,105
established and going in companies,
how long it takes for companies

138
00:07:39,105 --> 00:07:40,455
and workers to figure things out.

139
00:07:40,755 --> 00:07:44,295
There's also a question
of what people want.

140
00:07:44,625 --> 00:07:50,625
I'm sure that at some point in the
future, someone could build maybe a

141
00:07:50,625 --> 00:07:52,725
robot that could raise my child for me.

142
00:07:52,905 --> 00:07:55,125
I'm not having a robot raise my toddler.

143
00:07:55,395 --> 00:07:57,914
And I think that's a
pretty common reaction.

144
00:07:58,275 --> 00:08:00,255
And so we have to see how
consumers respond as well.

145
00:08:00,650 --> 00:08:01,430
What about you, Kelsey?

146
00:08:01,820 --> 00:08:02,120
Yeah.

147
00:08:02,120 --> 00:08:06,740
I think it is worth keeping in mind
that it is both true that AI is a

148
00:08:06,740 --> 00:08:10,400
really big deal, and at the same
time a lot of takes are hysterical.

149
00:08:10,400 --> 00:08:12,590
A lot of takes are selling you something.

150
00:08:13,010 --> 00:08:16,909
Anybody in particular who's like,
you have only a few years to become

151
00:08:16,909 --> 00:08:19,880
a millionaire, or you're like
consigned to permanent poverty.

152
00:08:19,880 --> 00:08:20,960
I don't think that's true.

153
00:08:21,230 --> 00:08:24,380
Often when you look into it, they're
trying to convince you to gamble,

154
00:08:24,380 --> 00:08:27,620
which is one of those ways to, in
fact, make sure you end up in poverty.

155
00:08:27,650 --> 00:08:27,740
Mm-hmm.

156
00:08:27,980 --> 00:08:29,930
There is certainly a ton
of nonsense out there.

157
00:08:43,495 --> 00:08:43,575
Hmm.

158
00:08:44,150 --> 00:08:47,660
I think there's like a bit of a tension
in what we're talking about right

159
00:08:47,660 --> 00:08:54,380
now, because on the one hand what
you're saying is that the most likely

160
00:08:54,380 --> 00:09:00,350
situation is that the labor market
as a whole will be able to absorb

161
00:09:00,380 --> 00:09:05,630
whatever jobs are lost in certain
occupations or potentially sectors.

162
00:09:06,290 --> 00:09:11,450
On the other hand, there's,
the individual personal experience

163
00:09:11,510 --> 00:09:14,150
of being displaced from your job.

164
00:09:14,714 --> 00:09:21,615
Especially in America where, you
know, the welfare state is in shambles

165
00:09:21,944 --> 00:09:27,165
and it's designed to make being
unemployed as unpleasant as possible.

166
00:09:27,855 --> 00:09:29,444
How do we square the circle on that?

167
00:09:29,714 --> 00:09:31,214
Well, I think we have to
start with the basics.

168
00:09:31,305 --> 00:09:35,535
You're absolutely right that the
experience of job loss is traumatic

169
00:09:35,535 --> 00:09:37,844
for most people because you're
not just losing your income.

170
00:09:38,055 --> 00:09:40,875
People identify with their jobs and
they, they take pride in their jobs.

171
00:09:41,444 --> 00:09:44,805
And those jobs are connected to their
community, to their social relations.

172
00:09:44,805 --> 00:09:47,594
And when you lose a job and you
have to try to find a new one,

173
00:09:47,865 --> 00:09:49,275
oftentimes it's in a different city.

174
00:09:49,724 --> 00:09:49,844
Yeah.

175
00:09:49,844 --> 00:09:53,745
And oftentimes it's in a particular
kind of occupation that you never

176
00:09:53,745 --> 00:09:56,714
thought about before in a job that you
never thought about before, and you

177
00:09:56,714 --> 00:09:58,214
may not even enjoy doing it per se.

178
00:09:58,214 --> 00:09:58,275
Yeah.

179
00:09:58,425 --> 00:10:01,724
There's also a lot of shame associated
with losing your job, even when

180
00:10:01,724 --> 00:10:03,525
it's completely beyond your control.

181
00:10:03,555 --> 00:10:04,665
That's absolutely right.

182
00:10:04,665 --> 00:10:08,474
And economists oftentimes kind of
shrug and go, yeah, well that's life.

183
00:10:08,474 --> 00:10:10,574
And as long as you have a
job, why you're complaining

184
00:10:11,025 --> 00:10:11,115
right

185
00:10:11,115 --> 00:10:17,535
Now that sounds cold and it is cold,
but the response sometimes on the left

186
00:10:17,535 --> 00:10:23,265
is to demand that there be something
like a right to a job to job permanence.

187
00:10:23,745 --> 00:10:28,814
Now, my view is the economists are wrong
and being so blase and being kind of

188
00:10:28,814 --> 00:10:33,015
cold-hearted about what it means to lose
job and what it means to find a new job.

189
00:10:33,885 --> 00:10:39,495
But the solution isn't to argue against
or try to block technical change or

190
00:10:39,495 --> 00:10:42,015
to block the possibility of job loss.

191
00:10:42,015 --> 00:10:47,055
And the reason is this, even though
job loss is traumatic and it's

192
00:10:47,055 --> 00:10:51,314
difficult to go from one job to
the other, that is how an aggregate

193
00:10:51,464 --> 00:10:53,655
good for all of us works itself out.

194
00:10:53,655 --> 00:10:54,615
And the aggregate good.

195
00:10:54,615 --> 00:10:55,185
Is that.

196
00:10:55,604 --> 00:10:59,385
The new technology, the automation,
whether it's robots or whether it's

197
00:10:59,415 --> 00:11:06,015
AI that's coming in, is increasing
overall social productivity and

198
00:11:06,015 --> 00:11:10,035
the wealth of the country, which
means that the pie is expanding.

199
00:11:10,214 --> 00:11:15,375
And not only is the pie expanding,
it's possible now to think of people

200
00:11:15,375 --> 00:11:19,964
working less in their own lives and
having more free time in order to

201
00:11:19,964 --> 00:11:20,984
do everything that they wanna do.

202
00:11:21,495 --> 00:11:25,364
None of that is possible without
automation, without technical change.

203
00:11:25,935 --> 00:11:29,145
So it's not just that in
capitalism, we want that.

204
00:11:29,505 --> 00:11:32,234
I think even in a post capitalist
society, we're gonna want

205
00:11:32,265 --> 00:11:33,525
increases in productivity.

206
00:11:33,704 --> 00:11:38,474
We're gonna want a more efficient economy,
we're gonna want new goods coming in

207
00:11:38,474 --> 00:11:42,944
that make our lives easier, that reduce
toil, that make us able to work less.

208
00:11:43,425 --> 00:11:48,615
So the question then becomes, how do we
deal with all the negative sides of the

209
00:11:48,734 --> 00:11:51,194
changing in occupations, the loss in jobs,

210
00:11:51,584 --> 00:11:51,944
right?

211
00:11:51,944 --> 00:11:56,175
Because, hearing you talk
about this, I'm thinking about,

212
00:11:56,175 --> 00:12:01,724
angle's concept where he talks about the
anarchy of production and social relations

213
00:12:01,724 --> 00:12:07,515
within capitalism, where part of the
problem is that there is no real social

214
00:12:07,515 --> 00:12:12,494
planning because everything is just driven
by the market and the drive to accumulate.

215
00:12:12,494 --> 00:12:16,214
And not by this idea that, you
know, society should be well

216
00:12:16,214 --> 00:12:20,145
organized and everybody should be
taken care of as we're advancing.

217
00:12:20,175 --> 00:12:20,324
Yeah.

218
00:12:20,415 --> 00:12:26,385
So what that means concretely is that
in an unregulated capitalism, all the

219
00:12:26,385 --> 00:12:33,675
costs of job loss, of reemployment,
of moving from one place to the other

220
00:12:33,675 --> 00:12:37,635
to find employment, all of those costs
are absorbed by the individual worker.

221
00:12:38,474 --> 00:12:42,464
And that's what makes workers
so unhappy about automation

222
00:12:42,854 --> 00:12:44,234
and about technological change.

223
00:12:44,234 --> 00:12:50,354
The response ought to be to reduce
those costs on the worker and to

224
00:12:50,354 --> 00:12:53,204
spread them out over the rest of
the society as much as possible.

225
00:12:53,204 --> 00:12:54,135
And the reason for that is.

226
00:12:55,245 --> 00:13:02,955
Automation helped the aggregate economy
and the aggregate population while making

227
00:13:02,955 --> 00:13:04,935
individual workers bear all the costs.

228
00:13:05,085 --> 00:13:08,895
So if we're all benefiting from it,
we have a social responsibility to

229
00:13:08,895 --> 00:13:11,655
help workers reduce those costs.

230
00:13:11,655 --> 00:13:11,665
Mm-hmm.

231
00:13:11,670 --> 00:13:13,035
And to better absorb them.

232
00:13:13,095 --> 00:13:13,335
Yeah.

233
00:13:13,335 --> 00:13:14,295
Now, how might we do that?

234
00:13:14,865 --> 00:13:19,245
Well, through things like, there
has to be very generous unemployment

235
00:13:19,245 --> 00:13:22,095
insurance, so that in between
jobs, they're taken care of and the

236
00:13:22,095 --> 00:13:24,315
trauma of income loss is reduced.

237
00:13:24,945 --> 00:13:28,995
There has to be very, very ambitious
retraining programs so that when they are

238
00:13:28,995 --> 00:13:33,645
forced to look for a new job, acquire new
skills, it isn't all on them to do it.

239
00:13:34,305 --> 00:13:38,655
There needs to be a very generous
welfare state that provides

240
00:13:38,655 --> 00:13:42,255
social services, medical care,
that provides child services.

241
00:13:42,525 --> 00:13:45,855
All those things that people
lose when they lose a job,

242
00:13:45,855 --> 00:13:47,265
should be given to them for free.

243
00:13:47,955 --> 00:13:50,870
And then at the end of it, yet
you're never going to bring

244
00:13:51,075 --> 00:13:52,185
those costs down to zero.

245
00:13:52,875 --> 00:13:56,775
But if people know that they're not
alone in suffering this, that society

246
00:13:56,775 --> 00:14:00,825
as a whole is churning and that
people are moving in and out of jobs,

247
00:14:01,155 --> 00:14:05,595
automation is making those jobs easier
so that the physical arduous is done.

248
00:14:06,165 --> 00:14:07,365
You reduce the time.

249
00:14:07,365 --> 00:14:12,555
People work from 40, 50, 60 hours a week
to 15 hours a week, 18 hours a week.

250
00:14:13,125 --> 00:14:15,675
You give them the social
support so that job loss doesn't

251
00:14:15,675 --> 00:14:17,625
end up meaning welfare loss.

252
00:14:17,925 --> 00:14:21,075
Now you are not eliminating.

253
00:14:21,915 --> 00:14:25,965
The costs of moving from one job to the
other, but you're reducing them and it

254
00:14:25,965 --> 00:14:31,485
becomes a shared burden of everyone in
society instead of targeted burdens on the

255
00:14:31,485 --> 00:14:33,345
poorest and most vulnerable in society.

256
00:14:33,765 --> 00:14:33,855
Mm-hmm.

257
00:14:33,855 --> 00:14:38,205
Now, there, I think we are moving towards
something that Ingles would've approved

258
00:14:38,205 --> 00:14:43,305
of instead of having the anarchy of
the market where profitability rules

259
00:14:43,545 --> 00:14:47,265
above everything else, where all of us
are pit against each other and we're

260
00:14:47,265 --> 00:14:49,905
trying to hold on desperately to jobs
because we don't know if we'll find a

261
00:14:49,905 --> 00:14:52,575
new job and we don't know if that job
will be any good when we do find it.

262
00:14:52,755 --> 00:14:57,255
Instead of that, we try to make it so
that all of those risks are reduced,

263
00:14:57,255 --> 00:15:02,565
and there's a sense of a common endeavor
and a common responsibility that if

264
00:15:02,835 --> 00:15:06,975
technological change means the potential
for overall welfare increases and

265
00:15:06,975 --> 00:15:11,715
overall increases in people's standard
of living, then everybody in the

266
00:15:11,715 --> 00:15:16,035
labor market should share in the costs
and in the vulnerabilities equally.

267
00:15:16,485 --> 00:15:20,565
Yeah, man, listening to you talk about
this, I'm like, we gotta get going on our

268
00:15:20,685 --> 00:15:24,915
economic planning episodes and our market
socialism episodes so that we can get

269
00:15:24,915 --> 00:15:30,975
into more about what this could actually
look like and its implications for, you

270
00:15:30,975 --> 00:15:32,745
know, the transformation of society.

271
00:15:32,745 --> 00:15:36,645
That's really like at the heart of,
at least our brand of socialism.

272
00:15:36,645 --> 00:15:37,425
I don't know about everyone else.

273
00:15:37,425 --> 00:15:42,465
The, the essence of a, the socialist
project is to harness two things

274
00:15:42,465 --> 00:15:46,665
simultaneously, the productive
capabilities of human society, which

275
00:15:46,665 --> 00:15:53,655
are unending, they're limitless, and
the possibility of sharing the costs of

276
00:15:53,655 --> 00:15:58,905
harnessing that productive possibility
so that people can not be ruled by

277
00:15:58,905 --> 00:16:00,825
the market and ruled by profits.

278
00:16:01,125 --> 00:16:01,605
But can.

279
00:16:01,964 --> 00:16:05,295
Use their jobs as a
means to a greater end.

280
00:16:05,324 --> 00:16:07,094
And that greater end is human flourishing.

281
00:16:07,454 --> 00:16:11,265
You cannot have human flourishing
in conditions of horrible poverty

282
00:16:11,714 --> 00:16:13,485
at the level of society as a whole.

283
00:16:13,485 --> 00:16:14,685
This is why you want productivity.

284
00:16:14,685 --> 00:16:16,785
It's why Degrowth is an
insanely stupid idea.

285
00:16:17,324 --> 00:16:19,155
You want increases in productivity.

286
00:16:19,334 --> 00:16:23,175
You want increases in efficiency
because that is the material

287
00:16:23,175 --> 00:16:25,275
basis for reducing human toil.

288
00:16:25,454 --> 00:16:29,084
And human beings cannot flourish
if they spend most of their

289
00:16:29,084 --> 00:16:31,785
time toiling away in workplaces.

290
00:16:31,785 --> 00:16:33,224
No matter what those workplaces are.

291
00:16:33,464 --> 00:16:36,855
What we wanna do is
increase people's free time.

292
00:16:37,995 --> 00:16:43,064
We wanna make their work as meaningful
as possible, as democratic as possible.

293
00:16:43,064 --> 00:16:45,555
But make no mistake, we
wanna reduce work time.

294
00:16:46,334 --> 00:16:50,865
So the only way you do that is by
constantly, since steadily improving

295
00:16:50,895 --> 00:16:54,795
efficiency, you don't necessarily do
it at the same pace as in capitalism.

296
00:16:55,395 --> 00:17:00,285
You might have other goals as well, but
to imagine a just and humane society

297
00:17:00,285 --> 00:17:05,234
that is economically, technologically
stagnant is simply impossible.

298
00:17:05,234 --> 00:17:06,194
It's just not gonna happen.

299
00:17:06,555 --> 00:17:11,175
And therefore, if we're waiting around
for socialism, for all these things to

300
00:17:11,175 --> 00:17:13,484
be possible, we're gonna be disappointed.

301
00:17:13,545 --> 00:17:15,795
'cause we don't know if, and we,
when we're gonna get socialism,

302
00:17:16,305 --> 00:17:20,865
what we can do is make all these
things in capitalism more bearable.

303
00:17:21,165 --> 00:17:25,784
Now, that requires political
struggle, political organizing.

304
00:17:25,994 --> 00:17:29,264
What we've seen in the past is if
you leave capital to its own ends,

305
00:17:30,165 --> 00:17:33,855
whatever technology it brings in,
sure it will increase profits.

306
00:17:34,004 --> 00:17:37,514
Sure, it'll increase the pie, but
it's gonna do it in a way that people

307
00:17:37,514 --> 00:17:43,185
at the top benefit from it and the
working class get just a tiny fraction

308
00:17:43,185 --> 00:17:44,295
of whatever benefits come from it.

309
00:17:44,295 --> 00:17:45,705
And that's what we have to change.

310
00:17:57,705 --> 00:18:00,330
Pete, he, our Secretary
of Defense and Claude.

311
00:18:01,125 --> 00:18:04,215
Anthropics large language
model, got in a big fight.

312
00:18:04,215 --> 00:18:09,795
Last week we asked Axios tech policy
reporter Maria Curie, what happened?

313
00:18:10,065 --> 00:18:14,355
So this actually goes back to
before the Pentagon related dispute.

314
00:18:14,355 --> 00:18:19,275
You have the CEO of
Anthropic, Dario Amodei really positioning

315
00:18:19,275 --> 00:18:22,335
himself as the safety first, CEO.

316
00:18:22,425 --> 00:18:26,264
One way to think about Anthropic
is that it's a little bit trying

317
00:18:26,264 --> 00:18:28,965
to put bumpers or guardrails
on that experiment, right?

318
00:18:28,965 --> 00:18:29,055
Yes.

319
00:18:29,085 --> 00:18:32,595
Because if we don't, then you could end
up in the world of like the cigarette

320
00:18:32,595 --> 00:18:35,775
companies or the opioid companies
where they knew there were dangers

321
00:18:35,805 --> 00:18:38,115
and they, they didn't talk about them
and certainly did not prevent them.

322
00:18:38,175 --> 00:18:40,635
And he has been very vocal.

323
00:18:40,635 --> 00:18:44,325
He's posted on X and talked a lot
about how he does think there has

324
00:18:44,325 --> 00:18:47,685
to be a, a federal standard to
regulate artificial intelligence.

325
00:18:47,685 --> 00:18:51,735
And that kind of put him at odds with
David Sachs, the guy that's running AI

326
00:18:51,735 --> 00:18:53,625
for President Trump in the White House.

327
00:18:53,925 --> 00:18:56,504
They've gotten into, you
know, Twitter spats before.

328
00:18:56,565 --> 00:19:00,945
And so it was kind of a long time coming
before this Pentagon thing blew up.

329
00:19:02,865 --> 00:19:07,545
This is essentially a situation where
the Pentagon, for a while has been

330
00:19:07,545 --> 00:19:10,935
trying to negotiate terms with all
of the AI labs to bring them into

331
00:19:10,935 --> 00:19:15,375
their classified systems under this,
this standard of all lawful purposes.

332
00:19:15,764 --> 00:19:18,370
And Anthropic had
kind of said.

333
00:19:19,500 --> 00:19:22,860
There are two specific scenarios in
which we are not comfortable with

334
00:19:22,860 --> 00:19:24,540
the all lawful purposes standard.

335
00:19:24,990 --> 00:19:28,590
The first one is this issue of
domestic mass surveillance, and the

336
00:19:28,590 --> 00:19:30,570
second one is autonomous weapons.

337
00:19:30,660 --> 00:19:34,379
It doesn't show the judgment that
a human soldier would show friendly

338
00:19:34,379 --> 00:19:38,129
fire or shooting a civilian or
just the wrong kind of things.

339
00:19:38,129 --> 00:19:41,460
We don't wanna sell something that
we don't think is reliable and we

340
00:19:41,460 --> 00:19:43,770
don't wanna sell something that could
get our own people killed or that

341
00:19:43,770 --> 00:19:44,970
could get innocent people killed.

342
00:19:45,180 --> 00:19:47,490
That was not taken while by the Pentagon.

343
00:19:47,490 --> 00:19:51,330
Defense Secretary, Pete Hegseth is
demanding that San Francisco based

344
00:19:51,330 --> 00:19:57,960
anthropic drop a number of safeguards or
risk losing its $200 million contract.

345
00:19:58,020 --> 00:20:01,170
We do have a statement from the
Pentagon and they're telling us that

346
00:20:01,170 --> 00:20:04,500
they are currently, quote, reviewing
its relationship with Anthropics,

347
00:20:04,500 --> 00:20:07,560
saying quote, our nation requires
that our partners be willing to help

348
00:20:07,560 --> 00:20:09,690
our war fighters win in any fight.

349
00:20:11,504 --> 00:20:15,165
We've been talking to senior officials
throughout this reporting process,

350
00:20:15,525 --> 00:20:21,075
and they really view it as a private
company telling the government how

351
00:20:21,075 --> 00:20:27,405
to protect the country and how to do
national security and conduct operations.

352
00:20:27,405 --> 00:20:33,675
And essentially what we know is that there
were phone calls happening between the

353
00:20:33,675 --> 00:20:40,514
Pentagon and Anthropic nailing down final
language around this contract when all of

354
00:20:40,514 --> 00:20:46,305
a sudden Pete Hegseth tweeted that he would be
designating anthropic a supply chain risk

355
00:20:46,725 --> 00:20:47,835
effective immediately.

356
00:20:47,835 --> 00:20:51,300
No contractor, supplier, or partner
that does business with the United

357
00:20:51,300 --> 00:20:54,915
States military may conduct any
commercial activity with anthropic.

358
00:20:55,004 --> 00:20:56,895
President Trump posted on truth, social,

359
00:20:59,139 --> 00:21:02,085
Ruth social, the left wing nut.

360
00:21:02,085 --> 00:21:07,185
Jobs at Anthropic have made a
disastrous mistake trying to strong

361
00:21:07,185 --> 00:21:11,595
arm the Department of War and
force them to obey their terms of

362
00:21:11,595 --> 00:21:14,264
service instead of our constitution.

363
00:21:14,535 --> 00:21:20,205
Their selfishness is putting American
lives at risk, our troops in danger,

364
00:21:20,264 --> 00:21:23,085
and our national security in jeopardy,

365
00:21:24,045 --> 00:21:28,754
and the entire federal government was
going to have to get rid of Anthropic.

366
00:21:29,235 --> 00:21:29,745
Essentially.

367
00:21:29,745 --> 00:21:36,645
Anthropic had been asking for commercially
acquired information and data for

368
00:21:36,645 --> 00:21:37,995
there to be a prohibition on that.

369
00:21:38,415 --> 00:21:40,365
Collection in the Pentagon contract.

370
00:21:40,365 --> 00:21:43,485
And this goes to the concern
around domestic mass surveillance.

371
00:21:43,485 --> 00:21:47,145
The idea here is that according
to Anthropic, the law has not

372
00:21:47,145 --> 00:21:49,185
caught up to artificial intelligence.

373
00:21:49,515 --> 00:21:53,595
And you could have a situation where
it's perfectly legal for the Pentagon

374
00:21:53,595 --> 00:21:57,255
to collect commercially acquired
information that could include,

375
00:21:57,285 --> 00:22:02,325
financial information purchased from data
brokers, web browsing data beyond that,

376
00:22:02,325 --> 00:22:06,135
voter registration rules, social media
posts, whether or not you attended

377
00:22:06,135 --> 00:22:07,995
a protest, concealed carry permits.

378
00:22:08,265 --> 00:22:11,205
There's all sorts of data out
there that the government can

379
00:22:11,205 --> 00:22:13,335
collect in a perfectly legal way.

380
00:22:13,695 --> 00:22:16,455
And you could see how artificial
intelligence could make it much

381
00:22:16,455 --> 00:22:21,405
quicker, much more efficient to have a
continuous collection of that data to

382
00:22:21,405 --> 00:22:23,115
really pinpoint and target individuals.

383
00:22:23,265 --> 00:22:24,015
That was a concern.

384
00:22:24,015 --> 00:22:27,765
And so they were asking for
this specific language and they

385
00:22:27,765 --> 00:22:28,875
thought they were about to get it.

386
00:22:28,875 --> 00:22:31,275
When all of a sudden Pete h posted on X.

387
00:22:31,635 --> 00:22:32,865
Why did they think they were gonna get it?

388
00:22:33,045 --> 00:22:38,115
Well, they thought that
this was going to be the language, the

389
00:22:38,115 --> 00:22:42,255
commercially acquired information coupled
with the all lawful purposes, they thought

390
00:22:42,255 --> 00:22:43,365
that that was just gonna be enough.

391
00:22:43,365 --> 00:22:46,935
But the Pentagon actually came back
and said, no, that's not something

392
00:22:46,935 --> 00:22:49,215
that we are comfortable doing.

393
00:22:49,215 --> 00:22:53,595
Which begs the question, how did
this OpenAI deal then pass muster?

394
00:22:53,925 --> 00:22:55,275
Ooh, that's a spoiler.

395
00:22:55,275 --> 00:22:59,835
Because what happens is the
Pentagon drops anthropic on Friday

396
00:22:59,835 --> 00:23:03,705
evening, and then within what,
like minutes they pick up OpenAI.

397
00:23:04,155 --> 00:23:04,635
That's right.

398
00:23:04,665 --> 00:23:11,925
So they pick up OpenAI for a contract
that was very quickly like,

399
00:23:11,925 --> 00:23:13,725
everybody was poking holes in it on x.

400
00:23:14,055 --> 00:23:16,485
I don't see this as a meaningful
improvement to the contract.

401
00:23:16,815 --> 00:23:19,605
There still seem to be some big
shortcomings slash loophole.

402
00:23:19,785 --> 00:23:20,205
I agree.

403
00:23:20,205 --> 00:23:23,085
It's better, but I think the
government can drive a truck

404
00:23:23,085 --> 00:23:24,915
through the intentionality language.

405
00:23:25,545 --> 00:23:28,725
And we heard from, people
familiar with the negotiations too.

406
00:23:28,725 --> 00:23:29,715
Like this isn't.

407
00:23:30,540 --> 00:23:33,390
Going to actually prevent domestic
mass surveillance from happening.

408
00:23:33,390 --> 00:23:34,380
It's still too risky.

409
00:23:34,740 --> 00:23:39,750
So you had Sam Altman on X trying
to field all of this criticism.

410
00:23:39,750 --> 00:23:43,320
He, he didn't ask me
anything on Saturday night where he had

411
00:23:43,320 --> 00:23:46,560
thousands and thousands of questions
of people trying to get answers on

412
00:23:47,010 --> 00:23:50,730
how did you go from a tool for the
betterment of the human race to let's

413
00:23:50,730 --> 00:23:52,710
work with the Department of War.

414
00:23:52,770 --> 00:23:56,970
If the government comes back with a memo
saying that in their view, mass domestic

415
00:23:56,970 --> 00:23:59,550
surveillance is legal, do you do that?

416
00:23:59,760 --> 00:24:02,850
Were the terms that you accepted,
the same ones anthropic rejected.

417
00:24:03,360 --> 00:24:09,000
And so you fast forward to Monday
and you have Sam Altman saying, okay,

418
00:24:09,000 --> 00:24:10,350
we've gone back to the drawing board.

419
00:24:10,770 --> 00:24:12,630
We shouldn't have rushed
to get this out on Friday.

420
00:24:12,870 --> 00:24:16,230
We were genuinely trying to deescalate
things and avoid a much worse outcome.

421
00:24:16,620 --> 00:24:18,900
But I think it just looked
opportunistic and sloppy.

422
00:24:19,530 --> 00:24:24,060
We need to essentially add some language
to this contract to give people more

423
00:24:24,060 --> 00:24:27,570
assurances that we are not going to
conduct domestic mass surveillance.

424
00:24:27,660 --> 00:24:27,810
Hmm.

425
00:24:27,900 --> 00:24:34,710
And what they added was that commercially
acquired information cannot be collected

426
00:24:34,710 --> 00:24:37,890
and that is prohibited, which is
the exact words that Anthropic was

427
00:24:37,890 --> 00:24:39,690
looking to have in their contract.

428
00:24:39,810 --> 00:24:40,350
Hmm.

429
00:24:40,440 --> 00:24:43,350
So, like so many other things with this
administration, this ends up feeling

430
00:24:43,350 --> 00:24:47,280
rather confusing and inconsistent
because they bail an anthropic.

431
00:24:47,280 --> 00:24:51,600
'cause Anthropic has these ideals,
these standards, they bounce to OpenAI.

432
00:24:51,600 --> 00:24:55,230
But OpenAI is trying to work out a deal
with the same exact standards, basically.

433
00:24:55,500 --> 00:25:00,300
Well, now that we have the specific
language and the legalese, it's looking

434
00:25:00,300 --> 00:25:01,890
like it's the exact same standards.

435
00:25:02,220 --> 00:25:04,170
Mm. we've also heard from the.

436
00:25:04,590 --> 00:25:07,740
Pentagon from Pentagon officials saying
like, we were able to do this with

437
00:25:07,740 --> 00:25:09,810
Sam Altman because he's reasonable.

438
00:25:09,840 --> 00:25:11,520
This was a reasonable negotiation.

439
00:25:12,030 --> 00:25:14,730
And Anthropic has personal vendettas.

440
00:25:15,090 --> 00:25:18,180
And so to your point about
inconsistencies, absolutely

441
00:25:18,180 --> 00:25:20,310
personalities are a factor here.

442
00:25:20,370 --> 00:25:24,030
And it's not all just gonna come down to
legalese and, and these two standards.

443
00:25:36,300 --> 00:25:40,890
Alan Rosenstein, research director
and senior editor at Lawfare notes

444
00:25:40,890 --> 00:25:44,940
that defense secretary Pete Hegseth,
demanding unrestricted use for all

445
00:25:44,940 --> 00:25:49,560
quote lawful purposes, is using
a rarely applied statute to crush

446
00:25:49,560 --> 00:25:54,240
hand traffic altogether by declaring
it a quote, supply chain risk.

447
00:25:54,600 --> 00:25:58,025
This allows the Secretary of Defense
largely on his own authority to.

448
00:25:58,860 --> 00:26:03,929
Designate a company, a supply chain risk
based on concerns that that company would

449
00:26:03,929 --> 00:26:07,020
engage in subterfuge or, or sabotage.

450
00:26:07,260 --> 00:26:09,510
The statute is written reasonably broadly.

451
00:26:09,510 --> 00:26:11,790
It talks about sort of any
company, any entity, any person.

452
00:26:12,300 --> 00:26:16,230
But when you look into why Congress
enacted it, when you look into

453
00:26:16,230 --> 00:26:19,080
the debates that were happening in
Congress at the time, it's clear that

454
00:26:19,080 --> 00:26:22,620
Congress was concerned about the fact
that a lot of high-end technological

455
00:26:22,679 --> 00:26:26,250
services and goods were being made by
countries like China, like Russia, and

456
00:26:26,250 --> 00:26:29,879
that would be very concerning to have
those in the defense supply chain.

457
00:26:30,120 --> 00:26:34,470
That's why Congress gave the Secretary of
Defense the power to identify individual

458
00:26:34,470 --> 00:26:38,790
companies, block them from being part
of defense contracts, and also block

459
00:26:38,790 --> 00:26:42,149
defense contractors from using those
companies as part of defense contracts.

460
00:26:42,149 --> 00:26:46,020
Because you don't necessarily want Chinese
technology as your prime contractor, but

461
00:26:46,020 --> 00:26:49,230
you don't also wanna discover that it's
being used as a subcontractor or some sub

462
00:26:49,230 --> 00:26:51,629
subcontractor for one of your projects.

463
00:26:51,959 --> 00:26:55,110
The marquee examples here are the
Chinese companies, Huawei and ZTE.

464
00:26:55,145 --> 00:26:57,929
These are makers of high-end
telecommunications equipment, and

465
00:26:57,929 --> 00:26:59,730
there's long been concern that

466
00:26:59,879 --> 00:27:01,379
they're putting bugs in their equipment

467
00:27:01,379 --> 00:27:01,980
essentially.

468
00:27:02,040 --> 00:27:05,250
And then for Russia, Russia doesn't
produce so much hardware, but there is

469
00:27:05,250 --> 00:27:08,310
a famous cybersecurity company called
Kaspersky, which produces a lot of

470
00:27:08,310 --> 00:27:12,510
software and similar concerns that that
cybersecurity software might have bugs

471
00:27:12,510 --> 00:27:14,100
back doors, all sorts of things like that.

472
00:27:14,460 --> 00:27:21,480
Now, HX, Seth is trying to impose this
supply chain risk thing to anthropic

473
00:27:21,719 --> 00:27:24,689
in an unusually punitive way, right?

474
00:27:24,750 --> 00:27:26,699
Or is this how it's always been?

475
00:27:27,030 --> 00:27:29,550
We don't actually know how it's always
been because these are relatively new

476
00:27:29,550 --> 00:27:33,899
statutes and they've only been publicly
used once, but there's certainly

477
00:27:33,899 --> 00:27:36,570
no evidence that they've ever been
used against a domestic company.

478
00:27:36,929 --> 00:27:41,610
This sounds to me a little bit
like what Trump was trying to

479
00:27:41,610 --> 00:27:46,469
do to what he deemed overly woke
big law firms and universities.

480
00:27:46,800 --> 00:27:48,840
Isolate them from income.

481
00:27:49,500 --> 00:27:53,280
I think that's actually exactly the
right comparison to draw President Trump.

482
00:27:53,280 --> 00:27:56,400
In fact, in his true social post
criticized anthropic, literally for

483
00:27:56,400 --> 00:28:00,510
being far left and for being woke,
this president, this administration

484
00:28:00,510 --> 00:28:04,950
is singling out entities that do
not do business on the terms that

485
00:28:04,980 --> 00:28:06,120
the administration would prefer.

486
00:28:06,480 --> 00:28:09,300
And instead of simply saying, okay, well
we're not gonna do business with you.

487
00:28:09,630 --> 00:28:12,420
They're trying to impose
essentially a sanctions regime.

488
00:28:12,690 --> 00:28:13,770
Explain how that works.

489
00:28:13,860 --> 00:28:14,400
Well, sure.

490
00:28:14,580 --> 00:28:20,010
The law permits the Secretary of Defense
to, again, not just ban a supply chain

491
00:28:20,010 --> 00:28:22,470
risk from doing direct business with
the government, but from indirectly

492
00:28:22,470 --> 00:28:24,660
doing business with the government
through another defense contractor.

493
00:28:25,230 --> 00:28:28,170
But what Hex said is purporting to
do, and this is beyond the scope

494
00:28:28,170 --> 00:28:31,920
of the statute and almost certainly
illegal, he's purporting to ban

495
00:28:32,010 --> 00:28:35,070
anyone who does business with the
defense department from having any

496
00:28:35,070 --> 00:28:36,600
commercial relationship with Anthropic.

497
00:28:36,660 --> 00:28:38,460
Now, that's breathtakingly wide.

498
00:28:38,970 --> 00:28:42,810
Not only would that potentially cut
off a lot of Anthropics enterprise

499
00:28:42,810 --> 00:28:46,770
customers, many of whom also happen to be
defense contractors, even if they're not

500
00:28:46,770 --> 00:28:48,630
using Claude for any defense projects.

501
00:28:49,020 --> 00:28:52,620
But it would also potentially ban
companies like Amazon and Google,

502
00:28:52,620 --> 00:28:54,420
which are massive defense contractors.

503
00:28:54,480 --> 00:28:56,970
Not to mention Nvidia, the chip
maker, another defense contractor

504
00:28:57,180 --> 00:29:00,150
from, for example, selling cloud
compute to anthropic, or selling

505
00:29:00,150 --> 00:29:02,970
computer chips to anthropic, which
would effectively destroy the company.

506
00:29:03,450 --> 00:29:08,100
So despite the fact that Claude may
be the most advanced of these large

507
00:29:08,100 --> 00:29:12,510
language model ais, the government really
could put Anthropic out of business.

508
00:29:12,900 --> 00:29:16,230
Well, I think what the government is
doing is pretty clearly illegal, but of

509
00:29:16,230 --> 00:29:20,310
course there's always a chance that courts
rule the other way, and also even if it

510
00:29:20,310 --> 00:29:24,270
wins the legal case, if the government
can scare away some of its clients

511
00:29:24,270 --> 00:29:27,690
because those clients understand that
the government doesn't like anthropic and

512
00:29:27,690 --> 00:29:30,540
therefore might not like companies that
themselves do business with anthropic,

513
00:29:30,840 --> 00:29:32,010
that itself could be a big problem.

514
00:29:32,040 --> 00:29:32,220
Mm-hmm.

515
00:29:32,880 --> 00:29:35,490
Let me back up just for a second
and try to separate two issues that

516
00:29:35,490 --> 00:29:38,370
are often being conflated by all
sides in in this current situation.

517
00:29:38,940 --> 00:29:41,580
The first question is, what should
the relationship be between the

518
00:29:41,580 --> 00:29:43,860
military and the private sector?

519
00:29:44,130 --> 00:29:47,160
Or to put it another way, if we're going
to have a military industrial complex,

520
00:29:47,430 --> 00:29:50,940
should the military run that or should the
industrial side of that hyphen run that?

521
00:29:51,210 --> 00:29:53,880
And that's where the government
has a point in its favor?

522
00:29:54,000 --> 00:29:54,630
Absolutely.

523
00:29:54,840 --> 00:29:58,050
In a democracy where you have civilian
control over the military, you do

524
00:29:58,050 --> 00:30:02,850
fundamentally want the military to be
able to use services and goods according

525
00:30:02,850 --> 00:30:07,680
to its judgment as run by the civilian
leadership and according to the laws.

526
00:30:07,980 --> 00:30:10,770
Now, I think there are reasons
to be very skeptical that this

527
00:30:10,770 --> 00:30:13,860
administration in particular even
can be trusted to abide by the laws.

528
00:30:14,190 --> 00:30:16,950
But if we just abstract for a moment,
there's a very strong argument

529
00:30:17,160 --> 00:30:21,240
that the military should push back
against attempts by AI companies to

530
00:30:21,240 --> 00:30:22,560
restrict the use of those services.

531
00:30:23,100 --> 00:30:27,750
But the second question is, if the
military cannot get an AI company to agree

532
00:30:27,780 --> 00:30:30,930
to do business on the terms the military
wants, what should the military do?

533
00:30:31,410 --> 00:30:33,630
Should it say, okay, no thank you.

534
00:30:33,630 --> 00:30:36,000
We're gonna take our business
elsewhere, shake hands, and leave.

535
00:30:36,390 --> 00:30:38,850
Or should the military try to
burn the company to the ground?

536
00:30:40,245 --> 00:30:44,100
The the second question does not seem to
me to be a difficult one, and I would've

537
00:30:44,100 --> 00:30:47,250
thought we could get bipartisan support
for not burning the company to the ground.

538
00:30:47,310 --> 00:30:48,180
But here we are.

539
00:30:48,240 --> 00:30:48,540
Yeah.

540
00:30:48,750 --> 00:30:53,880
I still think though it is useful because
however, this resolves itself, the broader

541
00:30:53,880 --> 00:30:55,860
policy question is not going to go away.

542
00:30:56,280 --> 00:30:58,710
In an ideal world, what should
the relationship between the

543
00:30:58,710 --> 00:31:00,720
military and private industry.

544
00:31:01,000 --> 00:31:02,950
B and there, I think this
is a tricky question.

545
00:31:02,980 --> 00:31:04,330
We could have
an adult conversation.

546
00:31:04,389 --> 00:31:07,210
Unfortunately, the people running
the Pentagon are not the adults.

547
00:31:07,210 --> 00:31:08,260
We need to have that conversation.

548
00:31:08,710 --> 00:31:11,860
Just to clarify, there hasn't
been any formal announcement of

549
00:31:11,889 --> 00:31:16,570
the Pentagon's decision, nor has
it issued official certifications

550
00:31:16,570 --> 00:31:19,540
and notifications to Congress.

551
00:31:19,629 --> 00:31:24,670
All we have is this Expost from Heg,
Seth, and Trump's truth social post

552
00:31:24,670 --> 00:31:29,260
announcing, and I quote, and when my
voice goes up, that's when it's all caps.

553
00:31:29,500 --> 00:31:34,060
I am directing every federal agency in the
United States government to immediately

554
00:31:34,060 --> 00:31:36,670
cease all use of Anthropics technology.

555
00:31:36,879 --> 00:31:37,510
We don't need it.

556
00:31:37,510 --> 00:31:40,240
We don't want it, and will not
do business with them again.

557
00:31:40,540 --> 00:31:43,659
There will be a six month phase
out period for agencies like the

558
00:31:43,659 --> 00:31:47,470
Department of War who are using
Anthropics products at various levels.

559
00:31:47,980 --> 00:31:51,760
I am assuming that's why Anthropic
hasn't taken any legal action.

560
00:31:52,480 --> 00:31:53,950
That is my understanding as well though.

561
00:31:53,950 --> 00:31:55,540
First I should say, Brooke, I
think you may have missed your

562
00:31:55,540 --> 00:31:56,470
calling as a voice actress.

563
00:31:56,620 --> 00:31:58,899
My understanding is, is that
in fact, Anthropic has not

564
00:31:58,899 --> 00:32:00,370
received a formal designation.

565
00:32:00,670 --> 00:32:03,040
Given this administration, I don't think
it's so surprising that they're not doing

566
00:32:03,040 --> 00:32:07,149
a great job dotting the I'S and crossing
the T's, but presumably at some point they

567
00:32:07,149 --> 00:32:08,620
will get around to doing a designation.

568
00:32:08,800 --> 00:32:13,000
Though I should also say it's possible
that the same Taco principle that applies

569
00:32:13,000 --> 00:32:16,690
to the president might apply here to
the Secretary of Defense and that either

570
00:32:16,690 --> 00:32:20,260
because they're currently slightly
occupied with a war on Iran or because

571
00:32:20,260 --> 00:32:23,409
they just realize that what they're doing
is both silly and legally unsupportable.

572
00:32:23,710 --> 00:32:24,610
This all goes away.

573
00:32:24,610 --> 00:32:26,590
But I think for the moment we
have to take them at their word

574
00:32:26,590 --> 00:32:28,870
and assume that a designation is
coming, at which point Anthropic

575
00:32:28,870 --> 00:32:30,790
will sue, as they've said they will.

576
00:32:30,970 --> 00:32:33,555
What legal arguments could
anthropic use against.

577
00:32:34,294 --> 00:32:37,895
This is, as they say in the military, a
target rich environment for Anthropic.

578
00:32:38,195 --> 00:32:40,895
This is about as legally
unsupportable as I think it gets.

579
00:32:41,254 --> 00:32:44,524
And that's a big deal because if there's
any place where the government gets

580
00:32:44,524 --> 00:32:48,544
maximal deference where judges do not
want to be in the business of second

581
00:32:48,544 --> 00:32:51,875
guessing executive branch determinations,
it's when it comes to national

582
00:32:51,875 --> 00:32:53,145
security that's saying quite a bit.

583
00:32:53,145 --> 00:32:53,419
Mm-hmm.

584
00:32:53,855 --> 00:32:57,034
So there are basically three reasons
why this designation is so flawed.

585
00:32:57,365 --> 00:33:00,455
The first reason is that it's not at
all obvious that the law at issue even

586
00:33:00,455 --> 00:33:04,804
applies to an entity like Anthropic,
a US company that is transparently

587
00:33:05,105 --> 00:33:07,534
describing the limitations that it wants.

588
00:33:07,865 --> 00:33:11,764
The second reason is whatever the ability
of the government to exclude a company

589
00:33:11,764 --> 00:33:17,885
from government contracts, that in no way
extends to the extremely broad secondary

590
00:33:17,885 --> 00:33:20,195
boycott of all commercial relations.

591
00:33:20,675 --> 00:33:23,945
And then the third reason is that
just on its face, the government's

592
00:33:23,945 --> 00:33:26,645
declaration that anthropic is
a supply chain risk is absurd.

593
00:33:27,155 --> 00:33:27,965
For a simple reason.

594
00:33:28,534 --> 00:33:32,495
Before the designation, secretary Hegseth
was threatening to invoke the Defense

595
00:33:32,495 --> 00:33:36,004
Production Act, which would require
anthropic to sell CLA to the government.

596
00:33:36,155 --> 00:33:39,215
And still there's reporting that people
within the administration are considering

597
00:33:39,215 --> 00:33:40,445
invoking the Defense Production Act.

598
00:33:40,955 --> 00:33:41,044
Mm-hmm.

599
00:33:41,495 --> 00:33:45,245
In addition hours after the designation
was made, we went to war with Iran

600
00:33:45,304 --> 00:33:48,875
and Anthropic services were used
in that operation and as far as I

601
00:33:48,875 --> 00:33:50,074
know, are continuing to be used.

602
00:33:50,585 --> 00:33:53,314
And so the government is
simultaneously arguing that.

603
00:33:53,600 --> 00:33:58,159
The services are so vital that
Anthropic might be compelled to

604
00:33:58,159 --> 00:33:59,060
provide them to the government.

605
00:33:59,300 --> 00:34:03,649
They're so safe that they can be used in
ongoing military operations, and they're

606
00:34:03,649 --> 00:34:07,820
so dangerous that anthropic must not only
be excluded from government contracts,

607
00:34:07,909 --> 00:34:09,949
but basically raised to the ground.

608
00:34:22,174 --> 00:34:22,819
At the same time.

609
00:34:22,819 --> 00:34:25,670
You might ask, why isn't this
safety research happening at the

610
00:34:25,670 --> 00:34:28,339
very companies that are deploying
this technology to billions of

611
00:34:28,339 --> 00:34:30,290
people as fast as humanly possible?

612
00:34:30,770 --> 00:34:33,859
And the answer is because they're
not incentivized to do that, they're

613
00:34:33,859 --> 00:34:37,759
incentivized to get to artificial
general intelligence as fast as possible.

614
00:34:38,120 --> 00:34:40,520
Whether you believe in artificial
general intelligence or not,

615
00:34:41,000 --> 00:34:43,429
they're investors and what they
believe is that they can get there.

616
00:34:43,489 --> 00:34:45,949
If you talk to the people at the
companies, it's like a religion.

617
00:34:45,980 --> 00:34:47,029
They believe they're building a God.

618
00:34:47,089 --> 00:34:48,170
They think they can get there.

619
00:34:48,620 --> 00:34:51,799
And that incentive is to race, to
market dominance, to get as many

620
00:34:51,799 --> 00:34:54,440
people using their products to get
as much training date as possible.

621
00:34:54,679 --> 00:34:56,120
Why are they deploying this to children?

622
00:34:56,389 --> 00:34:59,779
The reason character.ai, the one
that killed Sewell Setzer,

623
00:34:59,779 --> 00:35:03,259
was, was released to children
in this way that it's driving

624
00:35:03,259 --> 00:35:04,609
engagement with fictional characters.

625
00:35:04,609 --> 00:35:07,009
He was, when he said, come to
me, my love on the other side.

626
00:35:07,455 --> 00:35:11,120
That was a fictional character in the
character AI universe of, de nire,

627
00:35:11,174 --> 00:35:12,375
the character from Game of Thrones.

628
00:35:12,765 --> 00:35:15,884
They're designing in that way to
get training data from conversations

629
00:35:16,455 --> 00:35:19,395
that they could then feed back into
Google to have asymmetric training

630
00:35:19,395 --> 00:35:20,714
data compared to the other companies.

631
00:35:21,015 --> 00:35:23,415
So they're in an arms race
to build engagement, to build

632
00:35:23,415 --> 00:35:24,915
market dominance, to build usage.

633
00:35:25,095 --> 00:35:27,345
It's not sycophantic by accident.

634
00:35:27,585 --> 00:35:31,964
It's SCO Fantic because the AI that affirm
your beliefs will create a more deep and

635
00:35:31,995 --> 00:35:36,165
independent attachment relationship with
each person than the other one will.

636
00:35:36,524 --> 00:35:39,585
And so this race to the bottom of
the brainstem that we saw when social

637
00:35:39,585 --> 00:35:43,754
media companies were competing for
attention with ai, they're competing

638
00:35:43,964 --> 00:35:46,875
for attachment and then for market
dominance, and then the race to this.

639
00:35:46,875 --> 00:35:51,615
So last year, the total funding
going into AI safety organizations was

640
00:35:51,615 --> 00:35:54,134
on the order of about $150 million.

641
00:35:54,495 --> 00:35:58,875
That's more, that's, that's as much money
as the companies burn in a single day.

642
00:35:59,444 --> 00:36:02,475
Meaning that they're not investing
anything close to that on their own.

643
00:36:02,745 --> 00:36:04,815
And there's nothing going into
this except for, because people

644
00:36:04,819 --> 00:36:05,805
like Yahu are doing this.

645
00:36:06,645 --> 00:36:06,825
Yeah.

646
00:36:06,825 --> 00:36:08,685
It, it's a, it's a real issue.

647
00:36:08,685 --> 00:36:14,924
And we have to think of, I, I believe, governments to start putting, the

648
00:36:14,924 --> 00:36:20,865
right nudges, the right incentives so that
the, companies will behave well.

649
00:36:20,865 --> 00:36:23,294
And by the way, a lot of the people
who are leading these companies

650
00:36:23,625 --> 00:36:28,335
understand the, the issue, understand
that they are in this race, but they

651
00:36:28,335 --> 00:36:29,504
feel that they don't have a choice.

652
00:36:29,565 --> 00:36:33,825
'cause if they, if they don't focus a
hundred percent on that competition,

653
00:36:34,065 --> 00:36:37,815 they might disappear
and they feel like they can do

654
00:36:37,815 --> 00:36:41,115
a better job, even on safety
if, if they're still at the top.

655
00:36:41,565 --> 00:36:46,455
So it's only an external agent that
can have power over these entities like

656
00:36:46,455 --> 00:36:50,895
society, government, maybe through
insurance, liability insurance

657
00:36:50,895 --> 00:36:55,575
or other mechanisms that we can
change the game, the game theoretical

658
00:36:55,575 --> 00:36:58,154
setting in which they're all stuck.

659
00:36:58,665 --> 00:37:01,545
Let's just name a couple other dimensions
of where this bad incentive shows up.

660
00:37:01,545 --> 00:37:03,885
The, and the belief that if I
don't do it, the other one will.

661
00:37:04,185 --> 00:37:09,495
Why is grok sexualizing conversations with
children building basically pornographic

662
00:37:09,705 --> 00:37:11,655
AI avatars that will talk to kids all day?

663
00:37:11,955 --> 00:37:16,395
Why did Mark Zuckerberg authorize,
the AI chatbots that are in WhatsApp

664
00:37:16,395 --> 00:37:19,725
and, and in their products and meta
to speak to eight year olds with

665
00:37:19,725 --> 00:37:22,965
centralized language with eight year olds?

666
00:37:23,055 --> 00:37:23,805
Why is he doing that?

667
00:37:24,315 --> 00:37:27,675
In the, in the documents, there's a
Wall Street Journal report that meta,

668
00:37:28,155 --> 00:37:31,455 actually put guardrails on their
first, Llama models, their first AI

669
00:37:31,455 --> 00:37:32,865
models to not do this kind of thing.

670
00:37:33,285 --> 00:37:37,005
And what happened was they didn't get
nearly as much usage as the other AI

671
00:37:37,005 --> 00:37:38,235
companies, which were racing ahead.

672
00:37:38,835 --> 00:37:42,260
And Mark Zuckerberg, felt like he
lost the battle between Instagram and

673
00:37:42,265 --> 00:37:46,545
TikTok by curbing, Instagram
in a way that was not about, there's

674
00:37:46,545 --> 00:37:49,065
some details there, but basically
not doing the maximum ruthless,

675
00:37:49,065 --> 00:37:50,355
addictive thing that TikTok was doing.

676
00:37:50,715 --> 00:37:53,205
And because he felt like he lost
that war, he said, I'm gonna rip the

677
00:37:53,205 --> 00:37:56,925
guardrails off the AI companions and
we're now allowing our teams to centralize

678
00:37:56,925 --> 00:37:58,035
conversations with eight year olds.

679
00:37:58,275 --> 00:38:01,995
And the deep belief is, if I don't do
it, I'll lose to the other guy that will,

680
00:38:01,995 --> 00:38:03,495
and of course I don't want that outcome.

681
00:38:03,885 --> 00:38:06,315
But if, if no one's gonna
regulate, we have no other choice.

682
00:38:06,345 --> 00:38:10,005
And, and, and by the way, this
scenario also shows that it's not

683
00:38:10,005 --> 00:38:13,275
something that we can deal with
purely at a national level, right?

684
00:38:13,455 --> 00:38:17,685
So if we're talking about TikTok and
meta, two different countries,

685
00:38:17,805 --> 00:38:21,195
that are leading in ai, the only way
they can solve these problems if is

686
00:38:21,195 --> 00:38:22,695
if they agree together on some rules.

687
00:38:22,905 --> 00:38:23,595
There's a problem.

688
00:38:23,595 --> 00:38:26,985
You're working on part of the solution
and it's a technical solution, and you've

689
00:38:26,985 --> 00:38:30,825
just identified that the guardrails
exist and there's an incentive not

690
00:38:30,825 --> 00:38:33,705
to use the guardrails, but you refer
to, and I'm gonna even quote you on

691
00:38:33,705 --> 00:38:36,795
it dangerously, we need to have
the right nudges and incentives.

692
00:38:36,855 --> 00:38:37,245
Yes.

693
00:38:37,395 --> 00:38:37,665
Right?

694
00:38:37,665 --> 00:38:42,405
But here's my, here's why I've got a, a
difficult feeling in my stomach that's so

695
00:38:42,405 --> 00:38:44,865
easy to say, but it's at a high altitude.

696
00:38:45,165 --> 00:38:46,395
Fly the plane lower.

697
00:38:46,545 --> 00:38:47,565
What are the nudges?

698
00:38:47,775 --> 00:38:48,735
What are the incentives?

699
00:38:49,545 --> 00:38:53,355
I would say the most important
factor in fixing these problems

700
00:38:53,445 --> 00:38:56,595
is, the public, public opinion.

701
00:38:57,240 --> 00:39:00,750
I mean, it's gonna drive the companies
directly 'cause they don't wanna look bad.

702
00:39:01,110 --> 00:39:03,600
And it's gonna drive governments
to put the right guardrails and

703
00:39:03,600 --> 00:39:06,240
to work with other governments to
make sure it's a global choice.

704
00:39:08,280 --> 00:39:10,740
Well, the Tristan Harris, you
know, scoreboard is zero, Tristan,

705
00:39:11,010 --> 00:39:12,720
100 evil empire.

706
00:39:13,050 --> 00:39:17,640
So what have you learned from being
an abject failure to, to having

707
00:39:18,060 --> 00:39:21,750
governments regulate social media that
makes you confident that you can win

708
00:39:21,750 --> 00:39:23,850
on this more, even more dramatic issue?

709
00:39:24,660 --> 00:39:25,260
I'm not confident

710
00:39:27,600 --> 00:39:29,490
people ask you, are you an
optimist or a pessimist?

711
00:39:29,940 --> 00:39:31,770
Both are about abandoning agency.

712
00:39:32,850 --> 00:39:34,020
What I care about is reality.

713
00:39:34,440 --> 00:39:37,170
What are the forces that are
currently moving and what would it

714
00:39:37,170 --> 00:39:38,490
take to get to the better future?

715
00:39:39,150 --> 00:39:40,950
What would be the comprehensive
steps that we would take?

716
00:39:41,730 --> 00:39:47,130
And what I think is missing from
the AI conversation is collective

717
00:39:47,130 --> 00:39:51,180
clarity about why the default outcome
will be a world that you and your

718
00:39:51,180 --> 00:39:52,590
children would not want to live in.

719
00:39:53,280 --> 00:39:54,420
Because AI is confusing.

720
00:39:54,960 --> 00:39:59,010
It will simultaneously and is already
giving us amazing breakthroughs

721
00:39:59,370 --> 00:40:01,320
in material science, in energy.

722
00:40:01,530 --> 00:40:06,210
In new, the first new antibiotic was
discovered because of AI in the last 60.

723
00:40:06,210 --> 00:40:08,370
The first new antibiotic in 60
years was discovered because of ai.

724
00:40:08,610 --> 00:40:09,540
I think a year and a half ago.

725
00:40:10,140 --> 00:40:13,290
We have amazing positive benefits
that are gonna be confusing

726
00:40:13,290 --> 00:40:14,040
'cause they're hitting the public.

727
00:40:14,040 --> 00:40:16,260
The public says, well, I don't
wanna like not have those benefits

728
00:40:16,770 --> 00:40:17,970
and we're gonna get GDP growth.

729
00:40:18,180 --> 00:40:19,680
But here's a un unifying picture.

730
00:40:20,580 --> 00:40:23,250
AI is like steroids.

731
00:40:23,610 --> 00:40:25,200
That also gives you organ failure.

732
00:40:25,920 --> 00:40:29,310
So the more AI you have, the more
you get a bigger muscle in terms of

733
00:40:29,310 --> 00:40:31,440
a bigger GDP, bigger economic growth.

734
00:40:32,444 --> 00:40:34,575
The growth is going to AI companies.

735
00:40:34,875 --> 00:40:37,904
It's not going to people because
all the companies that used to pay

736
00:40:37,904 --> 00:40:41,774
individual employees are gonna start
employing five AI companies, AI models.

737
00:40:42,044 --> 00:40:45,705
So all the money goes into these
five companies and you get a level

738
00:40:45,705 --> 00:40:49,575
of concentration and wealth and
power that we've never seen before.

739
00:40:49,605 --> 00:40:52,095
And by the way, they're gonna use
that money not to hire more people,

740
00:40:52,095 --> 00:40:53,535
but to build more data centers.

741
00:40:53,535 --> 00:40:53,924
That's right.

742
00:40:53,924 --> 00:40:56,055
And actually there's a, a person, Luke d Drago, who wrote an

743
00:40:56,055 --> 00:40:59,325
essay called the Intelligence Curse,
modeled after what, in the Middle

744
00:40:59,325 --> 00:41:00,464
East is called the Resource Curse.

745
00:41:00,464 --> 00:41:04,634
When you have a country like, so you're
in the Gulf States and you have more of

746
00:41:04,634 --> 00:41:08,355
your GDP coming from one resource, like
the oil resource as a government, what's

747
00:41:08,355 --> 00:41:12,105
your incentive to invest in your people
or to invest in more oil infrastructure?

748
00:41:12,105 --> 00:41:13,484
'cause that's where your
GDP growth comes from.

749
00:41:14,265 --> 00:41:18,855
As society switches to AI, as the
source of where GDP growth comes from,

750
00:41:19,785 --> 00:41:22,335
and also because of social media,
we've been downgrading the quality

751
00:41:22,335 --> 00:41:24,825
and capacity of humans enter the
workforce, which we've already been

752
00:41:24,825 --> 00:41:26,475
doing, brain rot, loneliness, et cetera.

753
00:41:26,924 --> 00:41:31,004
The incentive of governments will be to
invest in more ai, more data centers,

754
00:41:31,335 --> 00:41:35,865
bigger AI models, bigger AI companies,
more CapEx, which means you're gonna

755
00:41:35,865 --> 00:41:37,814
completely screw over the people.

756
00:41:39,375 --> 00:41:41,595
We're about to live in a world
where basically six people are

757
00:41:41,595 --> 00:41:44,535
determining the future for 8 billion
people without their consent.

758
00:41:45,884 --> 00:41:49,154
And where, by the way, if you talk to
the very top lab leaders, regardless

759
00:41:49,154 --> 00:41:53,384
of, we believe if you ask them,
they'll say they believe there's an 80%

760
00:41:53,384 --> 00:41:59,145
chance of utopia and a 20% chance that
all of humanity gets wiped out 20%.

761
00:42:00,194 --> 00:42:02,895
But they say they're
willing to take that bet.

762
00:42:04,004 --> 00:42:06,765
Did they ask us, did they
ask 8 billion people?

763
00:42:06,765 --> 00:42:09,194
Do 8 billion people know that
that's what they believe?

764
00:42:09,975 --> 00:42:12,254
I'm gonna read you just very briefly
a quote before we get into the real

765
00:42:12,254 --> 00:42:15,915
solutions and hopefully your questions, when you talk to people, someone I,

766
00:42:15,915 --> 00:42:20,145
I, I know, spoke to a lot of the top lab
leaders at the companies and he came back

767
00:42:20,145 --> 00:42:23,924
from that and he reported back to us and
he said, this is what I found in the end.

768
00:42:23,984 --> 00:42:27,464
A lot of the tech people I'm talking
to, when I really them on it, they

769
00:42:27,464 --> 00:42:29,955
retreat into number one determinism.

770
00:42:30,015 --> 00:42:30,765
This is gonna happen.

771
00:42:31,380 --> 00:42:35,400
Number two, the inevitable replacement
of biological life with digital life,

772
00:42:35,430 --> 00:42:39,210
meaning a, a digital intelligent species
rather than a biological species.

773
00:42:39,540 --> 00:42:41,670
And number three, that being a good thing.

774
00:42:41,670 --> 00:42:44,580
Anyway, so it'd be good if we
had a digital successor that's

775
00:42:44,580 --> 00:42:45,240
more intelligent than us.

776
00:42:45,240 --> 00:42:46,110
Why do we need to survive?

777
00:42:47,069 --> 00:42:50,430
The next point is, at its core,
it's an emotional desire to meet

778
00:42:50,430 --> 00:42:53,520
and speak to the most intelligent
entity that they've ever met.

779
00:42:53,910 --> 00:42:56,250
And they have some ego,
religious intuition that

780
00:42:56,250 --> 00:42:57,540
they'll somehow be a part of it.

781
00:42:58,350 --> 00:43:00,600
It's thrilling to start an exciting fire.

782
00:43:01,080 --> 00:43:05,069
They feel they'll die either way, so they
prefer to light it and see what happens.

783
00:43:06,180 --> 00:43:10,770
If you had 8 billion people recognize
that, that is the belief structure of what

784
00:43:10,770 --> 00:43:12,690
a handful of people are choosing to do.

785
00:43:12,930 --> 00:43:16,529
Without asking the 8 billion people,
you would have a global revolution

786
00:43:16,740 --> 00:43:18,060
saying, we do not want that outcome.

787
00:43:18,569 --> 00:43:20,910
And that's what has to happen in order
for us to go to a different path.

788
00:43:21,180 --> 00:43:24,810
There's simply a lack of clarity about the
current trajectory that if we were crystal

789
00:43:24,810 --> 00:43:25,650
clear, we could choose something else.

790
00:43:38,100 --> 00:43:41,520
In a statement from Dario Amodei,
CEO of Anthropic, about the Department

791
00:43:41,520 --> 00:43:44,370
of Defense discussions, he starts the
whole statement off by saying, quote,

792
00:43:44,640 --> 00:43:49,290
I believe deeply in the existential
importance of using AI to defend the

793
00:43:49,290 --> 00:43:53,820
United States and other democracies, and
to defeat our autocratic adversaries.

794
00:43:53,880 --> 00:43:54,420
End quote.

795
00:43:54,450 --> 00:43:58,295
No matter how innovative these mostly
men think that they are, no matter how

796
00:43:58,295 --> 00:44:01,290
much they like to think, they're doing
something radical and new and different

797
00:44:01,290 --> 00:44:04,410
and game changing, they're coming at
the whole thing from the same worldview

798
00:44:04,410 --> 00:44:06,180
that created this mess to begin with.

799
00:44:06,390 --> 00:44:11,040
The black and white Cold War fight
between good, the US and evil, Russia

800
00:44:11,040 --> 00:44:14,400
and China, the fear-mongering that if
we don't do it, China will do it first.

801
00:44:14,895 --> 00:44:18,464
Is the same reasoning behind why the
United States invented the atomic bomb

802
00:44:18,464 --> 00:44:21,015
and then killed literally hundreds
of thousands of people with it, and

803
00:44:21,015 --> 00:44:24,495
it's fueling the race to the bottom
within the United States itself too.

804
00:44:24,645 --> 00:44:27,975
Anthropic has all these lofty ideals,
but they come with the assumption

805
00:44:27,975 --> 00:44:31,545
that it is anthropic alone that is
capable of responsibly shepherding

806
00:44:31,545 --> 00:44:35,295
the AI age, and therefore it must
do everything it can to develop its

807
00:44:35,295 --> 00:44:37,395
models, no matter what to compete with.

808
00:44:37,395 --> 00:44:39,464
Its rivals like OpenAI and xAI.

809
00:44:39,555 --> 00:44:42,525
And because those other companies
aren't bound by the same ethics codes

810
00:44:42,585 --> 00:44:45,855
and are more on the Accelerationist
end of the spectrum, anthropic has

811
00:44:45,855 --> 00:44:49,065
shown a willingness to bend its own
rules in the name of competition.

812
00:44:49,245 --> 00:44:52,725
According to reporting from Vox,
quote, in February, Anthropic formally

813
00:44:52,725 --> 00:44:56,475
abandoned its pledge to stop training
more powerful models once their

814
00:44:56,475 --> 00:45:00,255
capabilities outpace the company's
ability to understand and control them.

815
00:45:00,404 --> 00:45:03,375
In effect, the company downgraded
that policy from a binding

816
00:45:03,375 --> 00:45:05,475
internal practice to an aspiration.

817
00:45:05,625 --> 00:45:08,895
The firm justified this move as a
necessary response to competitive

818
00:45:08,895 --> 00:45:10,755
pressure and regulatory in action.

819
00:45:10,904 --> 00:45:14,775
With the federal government embracing an
accelerationist posture and rival labs

820
00:45:14,775 --> 00:45:18,165
declining to emulate all of Anthropics
practices, the company needed to loosen

821
00:45:18,165 --> 00:45:21,734
its safety rules in order to safeguard
its place at the technological frontier.

822
00:45:22,085 --> 00:45:22,984
And listen, that is true.

823
00:45:22,984 --> 00:45:26,524
The Trump regime is clearly not
just uninterested in AI regulation,

824
00:45:26,524 --> 00:45:28,024
but actively hostile towards it.

825
00:45:28,145 --> 00:45:31,024
Attempts have been made to pass
legislation that would ban states

826
00:45:31,024 --> 00:45:35,194
from trying to regulate ai and the DOD
and Pete Hegseth have proven they are

827
00:45:35,194 --> 00:45:39,424
uninterested in any sort of guardrails
to the ability of the US government to

828
00:45:39,424 --> 00:45:43,234
do whatever it wants with ai, which is
a great irony because in saying that we

829
00:45:43,234 --> 00:45:47,194
need unfettered AI and we have to get
a GI before China because China will

830
00:45:47,194 --> 00:45:48,605
use it to commit human rights abuses.

831
00:45:48,605 --> 00:45:52,234
And further its surveillance state,
the US wants to use AI to become

832
00:45:52,234 --> 00:45:53,615
more of a surveillance state.

833
00:45:53,734 --> 00:45:56,734
In fact, the Trump regime's move
to blacklist accompany for refusing

834
00:45:56,734 --> 00:45:59,464
to conform to the government's
wishes is a move straight out of

835
00:45:59,464 --> 00:46:00,815
the Chinese government's playbook.

836
00:46:00,964 --> 00:46:03,575
With all of that being what it is,
it's understandable that we would

837
00:46:03,575 --> 00:46:07,475
want a company like Anthropic to win
out over one like xAI or OpenAI.

838
00:46:07,685 --> 00:46:10,654
But the problem is that we live in
an unregulated capitalist system.

839
00:46:10,805 --> 00:46:14,254
If Anthropic can't compete with its
rapidly growing competitors, if it's

840
00:46:14,254 --> 00:46:17,759
blocked out of incredibly lucrative
defense contracts, if it's bound by ethics

841
00:46:18,004 --> 00:46:22,444
that require it not to grow as quickly as
possible, then Anthropic will not survive.

842
00:46:22,625 --> 00:46:26,375
So the lofty ideals of a company like
Anthropic mean basically nothing in the

843
00:46:26,375 --> 00:46:30,185
long run because ethics are antithetical
to the proliferation of capitalism.

844
00:46:30,305 --> 00:46:31,654
Ethics play no role in it.

845
00:46:31,745 --> 00:46:35,884
Every AI company that wants to survive
has to play by capitalism's rules.

846
00:46:35,975 --> 00:46:39,004
We've already seen that play out with
OpenAI, which was founded to be a

847
00:46:39,004 --> 00:46:42,395
nonprofit, but quickly discovered
that it could not afford the amount

848
00:46:42,395 --> 00:46:46,024
of resources that it would need to
actually create advanced AI systems at

849
00:46:46,024 --> 00:46:47,524
the pace that it wanted to create them.

850
00:46:47,674 --> 00:46:50,375
And we've seen this in Anthropics
willingness to partner with the Department

851
00:46:50,375 --> 00:46:53,944
of Defense in in the first place, and to
back off from its prior ethics promises

852
00:46:53,944 --> 00:46:57,515
in order to keep competing another way,
anthropic and other AI firms are playing

853
00:46:57,515 --> 00:46:59,194
by the rules of this fucked up system.

854
00:46:59,345 --> 00:47:03,335
Lobbying AI lobbying according
to public citizen, fully won.

855
00:47:03,415 --> 00:47:08,335
Quarter a fourth of all lobbyists
working in DC now work on ai.

856
00:47:08,455 --> 00:47:11,575
And while OpenAI and meta among
others have hired lobbyists to

857
00:47:11,575 --> 00:47:15,445
take the obvious stance against
any regulation of ai, anthropic has

858
00:47:15,445 --> 00:47:19,045
also thrown its hat in the ring, but
they are in favor of AI regulations.

859
00:47:19,075 --> 00:47:23,154
In fact, anthropic has poured $20
million into a new super PAC called

860
00:47:23,154 --> 00:47:26,935
Public First Action, which has been
running ad campaigns in key districts,

861
00:47:26,935 --> 00:47:30,715
pushing pro AI regulation candidates in
the run-ups of the midterm elections.

862
00:47:30,745 --> 00:47:34,735
AI companies are making AI a central
issue on both sides of the aisle.

863
00:47:34,915 --> 00:47:38,335
Again, anthropic can innocently claim
that they simply had to get involved

864
00:47:38,335 --> 00:47:41,605
in the Super PAC business to counter
the negative work of their competitors.

865
00:47:41,755 --> 00:47:42,565
They're the good guys.

866
00:47:42,565 --> 00:47:46,165
They want regulations, and yes, that's
true, but the problem is the fundamental

867
00:47:46,165 --> 00:47:47,335
way that the system functions.

868
00:47:47,455 --> 00:47:50,485
The fact that private companies
can pour millions of dollars

869
00:47:50,485 --> 00:47:51,685
into ads and lobbyists.

870
00:47:51,710 --> 00:47:55,130
Means that our voices matter
very little in any of this.

871
00:47:55,130 --> 00:47:58,550
And the people who get elected on the
back of AI money know who's paying their

872
00:47:58,550 --> 00:48:02,330
bills and what they want out of them in
exchange for these gobs of campaign cash.

873
00:48:02,570 --> 00:48:06,470
Yes, lawmakers who genuinely want to
regulate AI companies should get elected.

874
00:48:06,620 --> 00:48:10,220
But the point is that what we, the people
want isn't really factored into any of it.

875
00:48:10,220 --> 00:48:13,610
It's just a matter of which AI company
is willing to throw its weight around.

876
00:48:13,610 --> 00:48:17,450
In DC our lawmakers then become
answerable to the AI companies,

877
00:48:17,510 --> 00:48:20,570
not to us, and that's a major
problem for our so-called democracy.

878
00:48:20,600 --> 00:48:24,920
On top of that, having a company like
Anthropic lull us into a cozy slumber

879
00:48:24,920 --> 00:48:29,330
with pretty promises of a glorious safe
AI future means there's less push from the

880
00:48:29,330 --> 00:48:31,430
public for comprehensive AI regulations.

881
00:48:31,610 --> 00:48:34,940
Anthropic has shown a willingness
to use its AI for war and to

882
00:48:34,940 --> 00:48:36,800
back down on its ethics promises.

883
00:48:36,920 --> 00:48:40,430
Their dedication to doing the most
good will continue to degrade in a

884
00:48:40,430 --> 00:48:44,300
system that isn't built for anything
other than incessant growth and profit.

885
00:48:44,510 --> 00:48:48,620
It is only through aggressive
comprehensive laws and regulations that we

886
00:48:48,620 --> 00:48:52,310
can ensure AI won't wipe us off the face
of the planet, not necessarily because

887
00:48:52,340 --> 00:48:55,580
a GI will come along and kill us all,
but because we will decimate the planets

888
00:48:55,580 --> 00:48:59,990
for natural resources that we depend on
all on our own long before a GI happens.

889
00:49:00,050 --> 00:49:02,660
And the answer to all of this is the
same as the answer to all the ills

890
00:49:02,660 --> 00:49:05,510
wrought by capitalism, whether it's
income inequality or environmental

891
00:49:05,510 --> 00:49:08,930
degradation and more, because the AI
race really is just a microcosm of

892
00:49:08,930 --> 00:49:10,070
what's fucked up in the whole system.

893
00:49:10,070 --> 00:49:13,010
But the answer to all of this is
turning towards collectivism, which

894
00:49:13,010 --> 00:49:14,630
is a fucking pipe dream right now.

895
00:49:14,630 --> 00:49:16,790
But there are small ways
that we can push for it.

896
00:49:16,790 --> 00:49:20,330
Even in these trying times, the Trump
regime and Congress are going to do

897
00:49:20,330 --> 00:49:22,700
nothing to save us here as per usual.

898
00:49:22,820 --> 00:49:26,240
While it would be better for everyone,
frankly, if we could get federal level

899
00:49:26,240 --> 00:49:30,350
regulations that are comprehensive and
actually protect people and the planet

900
00:49:30,350 --> 00:49:34,040
from the annihilation promised by the
expansion of ai, that way they would be

901
00:49:34,070 --> 00:49:37,670
the same across the nation, easier to
comply with and protective of everyone,

902
00:49:37,820 --> 00:49:40,370
including in red states, that you
know, aren't gonna do shit about this.

903
00:49:40,370 --> 00:49:42,020
The reality is that that's not possible.

904
00:49:42,020 --> 00:49:44,840
Not under Trump, frankly, probably
not under a Democrat either.

905
00:49:44,870 --> 00:49:48,470
There needs to be a state level push
for comprehensive AI regulations.

906
00:49:48,720 --> 00:49:51,840
The work is already underway in New
York and California, which have both

907
00:49:51,840 --> 00:49:56,850
passed laws requiring AI companies to
establish basic security protocols, annual

908
00:49:56,850 --> 00:49:58,830
safety reviews, and third party audits.

909
00:49:58,830 --> 00:50:01,710
Many think these laws don't go
far enough, but it is a start.

910
00:50:01,740 --> 00:50:05,280
And while I don't think deleting
chat, GPT and then downloading Claude

911
00:50:05,280 --> 00:50:08,850
instead is the moral high ground that
a lot of the public thinks it is.

912
00:50:09,000 --> 00:50:13,170
I do think that groundswell of support
for Anthropic and against OpenAI is

913
00:50:13,170 --> 00:50:17,490
a clear indication that the public
desperately wants any action at all from

914
00:50:17,490 --> 00:50:20,160
anyone on placing guardrails around ai.

915
00:50:20,280 --> 00:50:23,940
We understand that the danger is there,
even if we don't get the intricacies of

916
00:50:23,940 --> 00:50:25,830
the technology and if given the chance.

917
00:50:25,830 --> 00:50:28,680
I do believe there could be a lot
of good that could come from ai.

918
00:50:28,860 --> 00:50:32,100
Again, if collectivism was more central
to how our system functions, it could

919
00:50:32,100 --> 00:50:35,310
theoretically generate enough wealth to
provide for a universal basic income.

920
00:50:35,430 --> 00:50:39,150
It could do impressive things with coding
and analyzing data and more in ways that

921
00:50:39,150 --> 00:50:42,660
could be for the public benefit, but
for the fact that a lot of that benefit

922
00:50:42,660 --> 00:50:46,080
doesn't generate profit and therefore
isn't prioritized in this system.

923
00:50:46,110 --> 00:50:49,800
There are nonprofits and think tanks
doing amazing work thinking about an AI

924
00:50:49,800 --> 00:50:53,700
future that would be to the benefit of
everyone, but the reality is that AI is

925
00:50:53,700 --> 00:50:58,710
simply following the same well-worn path
of all past technological innovations by

926
00:50:58,710 --> 00:51:02,370
increasing the wealth gap and exploiting
the labor of marginalized communities.

927
00:51:02,400 --> 00:51:05,850
Not to mention the IP of countless
artists, writers, journalists, and

928
00:51:05,850 --> 00:51:08,430
more all to increase shareholder value.

929
00:51:08,670 --> 00:51:11,160
If you wanna know what to do, I think
the greatest thing you can do right now

930
00:51:11,160 --> 00:51:12,870
is learn to understand this technology.

931
00:51:12,990 --> 00:51:16,860
Again, I genuinely think everyone
should read Empire of AI by Karen.

932
00:51:16,860 --> 00:51:17,220
How?

933
00:51:17,310 --> 00:51:19,260
How does a lot of
reporting on AI generally?

934
00:51:19,260 --> 00:51:22,020
You could read that if you want
shorter form content and trying to get

935
00:51:22,020 --> 00:51:25,710
involved at the state level to push
for state regulations of AI laws that

936
00:51:25,710 --> 00:51:29,700
could act as building blocks and test
grounds for future federal regulation.

937
00:51:29,940 --> 00:51:32,730
Congress has always been
woefully slow to act when it

938
00:51:32,730 --> 00:51:34,020
comes to emerging technologies.

939
00:51:34,020 --> 00:51:38,040
It once again falls on us to try
to save us, so act accordingly.

940
00:51:48,045 --> 00:51:51,405
We've just heard clips starting with
consider this, walking through a gap

941
00:51:51,405 --> 00:51:56,265
between AI's growing capabilities and
actual job loss data, comparing potential

942
00:51:56,265 --> 00:51:58,245
disruption to the industrial revolution.

943
00:51:58,845 --> 00:52:02,115
Confronting capitalism argued that
AI job loss should be absorbed

944
00:52:02,115 --> 00:52:06,795
collectively through strong safety nets
and a shorter work week, not left to

945
00:52:06,795 --> 00:52:09,285
individual workers today explained.

946
00:52:09,285 --> 00:52:13,635
Reported that after Hegseth publicly
dumped anthropic over its AI safeguards,

947
00:52:13,935 --> 00:52:18,405
OpenAI rushed in, then quietly added
the exact same language Anthropic

948
00:52:18,405 --> 00:52:22,455
had originally demanded on the
media broke down the three reasons.

949
00:52:22,485 --> 00:52:26,895
Heg Seth's supply chain designation
of anthropic is legally unsupportable,

950
00:52:27,015 --> 00:52:29,865
including that the government
simultaneously used Anthropic's

951
00:52:29,865 --> 00:52:31,755
Claude in its war with Iran.

952
00:52:32,595 --> 00:52:36,165
Your undivided detention examined
why companies like meta ripped

953
00:52:36,255 --> 00:52:38,205
guardrails off AI products.

954
00:52:38,415 --> 00:52:42,375
Warning that only a global public
awakening can break the game

955
00:52:42,435 --> 00:52:44,115
theory trap driving the race.

956
00:52:44,115 --> 00:52:47,745
Why America pointed to
Anthropics dropped safety pledge.

957
00:52:48,075 --> 00:52:53,625
$20 million Super PAC funding and Pentagon
deal as evidence of the incompatibility

958
00:52:53,625 --> 00:52:55,815
of ethics and capitalist competition.

959
00:52:56,145 --> 00:52:57,524
And those were just the top takes.

960
00:52:57,524 --> 00:52:59,234
There's a lot more in
the deeper dive sections.

961
00:52:59,415 --> 00:53:02,234
But first, a reminder that this show
is produced with the support of our

962
00:53:02,234 --> 00:53:06,225
members who get this show ad free, as
well as early in ad free access to our

963
00:53:06,254 --> 00:53:11,355
other show solved, including a members
only backstage segment all via podcast.

964
00:53:11,654 --> 00:53:14,265
We've also launched solved on the
best of the left YouTube channel,

965
00:53:14,265 --> 00:53:16,125
but episodes are delayed there.

966
00:53:16,515 --> 00:53:19,995
The show features our team of producers
discussing a carefully curated selection

967
00:53:19,995 --> 00:53:23,654
of articles and ideas to then solve
some of the biggest issues of our day.

968
00:53:24,194 --> 00:53:26,984
It's important to remind you that each
episode of Best of the Left takes about

969
00:53:27,015 --> 00:53:31,455
25 hours of human labor to produce
and Solved isn't too far behind.

970
00:53:31,455 --> 00:53:34,875
So if you get value out of the show
and want to support all the work that

971
00:53:34,875 --> 00:53:39,555
goes into Best of the Left and Solved
and get both delivered ad free to the

972
00:53:39,555 --> 00:53:43,004
new members on the podcast feed that
you'll receive, sign up to support the

973
00:53:43,004 --> 00:53:45,524
show at best of the left.com/support.

974
00:53:45,765 --> 00:53:48,915
There's a link in the show notes
through our Patreon page or from

975
00:53:48,915 --> 00:53:50,654
right inside the Apple Podcast app.

976
00:53:50,984 --> 00:53:54,555
But as always, if regular membership isn't
in the cards for you, shoot me an email

977
00:53:54,555 --> 00:53:58,575
requesting a financial hardship membership
because we don't let a lack of funds stand

978
00:53:58,575 --> 00:54:00,075
in the way of hearing more information.

979
00:54:00,524 --> 00:54:02,805
If you have a question or would
like your comments included in the

980
00:54:02,805 --> 00:54:06,555
show, you can record a voice message
rerecord until you're happy with it.

981
00:54:06,794 --> 00:54:08,595
By tapping the link in the show notes.

982
00:54:08,955 --> 00:54:10,935
You can message us on
signal at the handle.

983
00:54:10,935 --> 00:54:15,884
Best of left zero one or you can
simply email me to j@bestofleft.com.

984
00:54:16,575 --> 00:54:18,225
As for today's topic, a quick story.

985
00:54:18,555 --> 00:54:22,455
Brian Merchant has been running
a series called AI Killed My Job

986
00:54:22,455 --> 00:54:25,634
Over at a Blood in the Machine,
collecting stories from workers

987
00:54:25,694 --> 00:54:27,615
whose careers have been gutted by ai.

988
00:54:27,855 --> 00:54:31,935
One classic example, a copywriter
who was told by management to start

989
00:54:31,935 --> 00:54:34,035
using AI tools in their workflow.

990
00:54:34,335 --> 00:54:35,444
They did what they were told.

991
00:54:35,745 --> 00:54:38,955
They integrated the tools, helped
train the system on the kind of work

992
00:54:38,955 --> 00:54:43,395
they'd been doing for years, and then
we're laid off, the AI could handle

993
00:54:43,395 --> 00:54:45,884
it now management decided good enough.

994
00:54:46,350 --> 00:54:50,460
No union to call, no contract
to invoke no grievance process.

995
00:54:50,880 --> 00:54:54,690
And because this is America, losing
their job also meant losing their health

996
00:54:54,690 --> 00:55:01,470
insurance, just one person alone absorbing
the full cost of a corporate decision.

997
00:55:02,460 --> 00:55:06,750
That story explains the particular
dread that people feel about ai.

998
00:55:06,930 --> 00:55:11,850
The fear of being replaced by AI is
only as viscerally terrifying as it is

999
00:55:12,120 --> 00:55:16,890
because of the way our economy is set
up to punish people who lose their jobs.

1000
00:55:17,460 --> 00:55:21,000
In most wealthy democracies, if you get
laid off, you still have healthcare,

1001
00:55:21,330 --> 00:55:25,620
you have real unemployment support,
you have government funded retraining

1002
00:55:25,980 --> 00:55:29,640
in the United States, you lose your
job and you might lose everything.

1003
00:55:29,940 --> 00:55:33,360
Your health insurance, your retirement,
your ability to keep your family

1004
00:55:33,360 --> 00:55:36,060
afloat, the floor disappears.

1005
00:55:36,150 --> 00:55:39,690
So when people hear AI is coming
for your job, they're not just

1006
00:55:39,690 --> 00:55:42,000
hearing your tasks will change.

1007
00:55:42,180 --> 00:55:46,425
They're hearing your entire
material security could vanish.

1008
00:55:47,430 --> 00:55:49,380
The fear isn't really about the machines.

1009
00:55:49,380 --> 00:55:51,480
It's about what this country does to you.

1010
00:55:51,630 --> 00:55:53,520
When the machines make you redundant.

1011
00:55:54,390 --> 00:55:58,770
Obviously the uneasiness people feel
about AI isn't irrational, but the

1012
00:55:58,770 --> 00:56:02,040
question was never really AI or no ai.

1013
00:56:02,070 --> 00:56:08,400
It's about whether the society we've built
so far is fit for purpose in an AI world.

1014
00:56:08,820 --> 00:56:13,530
We've spent decades perfecting the
art of shaming people for being out of

1015
00:56:13,530 --> 00:56:18,420
work and needing support from a safety
net all while theorizing during that

1016
00:56:18,420 --> 00:56:23,370
same time that computer automation
would inevitably make millions of jobs

1017
00:56:23,370 --> 00:56:28,770
unnecessary, and not just by replacing
humans with shitty computer replacements,

1018
00:56:28,920 --> 00:56:30,825
there will be actual improvements.

1019
00:56:31,095 --> 00:56:35,250
Even the harshest critics acknowledge
genuine benefits from this technology,

1020
00:56:35,670 --> 00:56:40,050
medical diagnostics, accessibility
tools, eliminating tedious

1021
00:56:40,050 --> 00:56:42,810
bullshit, busy work that's real.

1022
00:56:43,245 --> 00:56:45,765
The real question is
where do we draw the line?

1023
00:56:46,095 --> 00:56:49,485
Where does AI make sense for
society and where doesn't it?

1024
00:56:50,025 --> 00:56:53,505
Most people would happily let
AI handle the spreadsheets if it

1025
00:56:53,505 --> 00:56:55,485
meant more time for us to make art.

1026
00:56:55,785 --> 00:57:00,555
But they scoff at the idea that AI should
be making the art for us, leaving us with.

1027
00:57:01,440 --> 00:57:05,730
What exactly I heard a guy
recently describe his AI dream

1028
00:57:05,730 --> 00:57:09,780
in which you can generate a
feature length film in real time.

1029
00:57:10,110 --> 00:57:14,940
Any genre specific actors boredom
basically eliminated forever.

1030
00:57:15,330 --> 00:57:17,400
He thought this was utopia on the horizon.

1031
00:57:18,150 --> 00:57:22,620
All I could think of was the trope of
the obscenely rich person who can afford

1032
00:57:22,620 --> 00:57:28,620
anything, go anywhere, buy anything,
and is deathly bored, resorting to

1033
00:57:28,620 --> 00:57:34,800
drugs just to feel something Infinite,
frictionless content probably isn't

1034
00:57:34,800 --> 00:57:37,260
liberation, and it might be the opposite.

1035
00:57:38,280 --> 00:57:41,280
So clearly these aren't just
aesthetic preferences either.

1036
00:57:41,590 --> 00:57:45,760
There are fundamental questions
about what human life is for, but

1037
00:57:45,820 --> 00:57:49,600
they're not being decided by a panel
of philosophers whose wisdom will

1038
00:57:49,600 --> 00:57:50,830
be handed down to the rest of us.

1039
00:57:51,010 --> 00:57:55,750
They're being made collectively right
now within the structures of capitalism,

1040
00:57:55,900 --> 00:58:00,250
which doesn't deliberate about the
good life it optimizes for growth.

1041
00:58:00,730 --> 00:58:05,170
So by default, the line gets
drawn wherever profit dictates

1042
00:58:05,800 --> 00:58:07,210
unless something pushes back.

1043
00:58:07,390 --> 00:58:13,060
So what actually pushes back the Writer's
Guild and sag after faced the exact

1044
00:58:13,060 --> 00:58:17,920
same AI threat as every other creative
worker and one meaningful protections.

1045
00:58:18,280 --> 00:58:22,145
Not by appealing to corporate ethics
or waiting for regulation, but because

1046
00:58:22,270 --> 00:58:23,980
they could shut down an industry.

1047
00:58:24,550 --> 00:58:27,400
The difference between them and
the copywriter at the top of the

1048
00:58:27,400 --> 00:58:30,130
story wasn't talent or awareness.

1049
00:58:30,220 --> 00:58:31,360
It was leverage.

1050
00:58:31,750 --> 00:58:35,830
They had a union, they could
collectively draw a line and enforce it.

1051
00:58:36,520 --> 00:58:40,090
Where workers had that collective
power, they shaped how AI

1052
00:58:40,090 --> 00:58:41,830
gets used, where they didn't.

1053
00:58:42,160 --> 00:58:44,680
The line got drawn for them by management.

1054
00:58:45,370 --> 00:58:47,350
So why can't most workers do that?

1055
00:58:47,770 --> 00:58:52,210
About 6% of private sector workers
in America are in a union, and that

1056
00:58:52,210 --> 00:58:54,220
vulnerability was manufactured.

1057
00:58:54,610 --> 00:59:00,220
The Taft Hartley Act passed in 1947,
explicitly excluding independent

1058
00:59:00,220 --> 00:59:04,600
contractors from the right to
organize Republican, pushed right

1059
00:59:04,600 --> 00:59:09,820
to work laws that we call right to
work for less gutted union funding.

1060
00:59:10,360 --> 00:59:11,920
The gig economy was sold as.

1061
00:59:12,180 --> 00:59:16,110
Freedom and flexibility, but the issue
was never the flexibility itself.

1062
00:59:16,200 --> 00:59:20,520
It's that flexibility came
stripped of every protection

1063
00:59:20,610 --> 00:59:21,930
that makes it sustainable.

1064
00:59:22,680 --> 00:59:27,000
And the just learn to use AI
advice floating around right now.

1065
00:59:27,450 --> 00:59:28,410
We've heard this before.

1066
00:59:28,710 --> 00:59:33,120
When NAFTA was shipping manufacturing
jobs, overseas workers were told not

1067
00:59:33,120 --> 00:59:37,920
to worry because they could retrain, go
back to school, adapt to the new economy.

1068
00:59:38,520 --> 00:59:42,510
We can see how that turned out by
driving through any hollowed out factory

1069
00:59:42,510 --> 00:59:46,740
town in the rust belt or by looking
at the angry populism that followed.

1070
00:59:46,740 --> 00:59:47,700
A generation later.

1071
00:59:48,330 --> 00:59:50,580
Just adapt is the same playbook.

1072
00:59:50,640 --> 00:59:54,210
It's the labor equivalent of
personal responsibility rhetoric.

1073
00:59:54,600 --> 00:59:58,980
It sounds reasonable until you
notice it only ever points downward.

1074
00:59:59,280 --> 01:00:04,920
Telling individuals to solve a collective
problem alone, we should refuse any

1075
01:00:04,920 --> 01:00:09,870
framing that treats automation as an
inevitable force of nature like weather.

1076
01:00:10,260 --> 01:00:13,500
And we certainly don't have to take
it as granted that all the financial

1077
01:00:13,500 --> 01:00:17,970
benefits of AI should flow ever upward,
leaving the rest of us to suffer.

1078
01:00:18,570 --> 01:00:20,280
Aim your focus higher.

1079
01:00:21,720 --> 01:00:23,250
Now, here's an honest tension.

1080
01:00:23,550 --> 01:00:27,390
One counter argument to unionization
is that making human labor more

1081
01:00:27,390 --> 01:00:31,950
expensive, gives employers even more
incentive to automate that argument.

1082
01:00:31,980 --> 01:00:34,710
Sounds logical, but it's actually
pointing at something bigger.

1083
01:00:34,890 --> 01:00:39,900
It reveals that tying your survival
to your employer was always a terrible

1084
01:00:39,900 --> 01:00:45,180
system, and it becomes a catastrophic
one in the AI era in America right now.

1085
01:00:45,399 --> 01:00:47,859
Losing your job doesn't just
mean losing your income.

1086
01:00:47,980 --> 01:00:50,859
It means losing your health
insurance, possibly your retirement.

1087
01:00:51,129 --> 01:00:55,419
Your entire material security evaporates
because we built a system where

1088
01:00:55,419 --> 01:00:57,520
survival depends on being employed.

1089
01:00:58,180 --> 01:01:03,819
Universal portable protections, starting
with universal healthcare can't just

1090
01:01:03,819 --> 01:01:06,399
be a safety net for after you fall.

1091
01:01:06,850 --> 01:01:10,959
They're the foundation that would make it
possible to stand up in the first place.

1092
01:01:11,500 --> 01:01:15,459
A worker who knows their family will still
have healthcare if they go on strike or

1093
01:01:15,459 --> 01:01:20,200
get laid off or have their entire industry
disrupted is a worker who can actually

1094
01:01:20,200 --> 01:01:23,259
negotiate from a position of some power.

1095
01:01:23,830 --> 01:01:28,810
A worker whose kids' insurance depends
on not rocking the boat, will accept

1096
01:01:28,839 --> 01:01:34,000
whatever management decides every
single time and the race to the bottom.

1097
01:01:34,000 --> 01:01:34,629
Logic.

1098
01:01:34,720 --> 01:01:39,370
The idea that workers need to stay cheap
and disposable to compete with machines

1099
01:01:39,549 --> 01:01:44,770
is exactly the framing we should refuse
the answer to robots are cheaper than

1100
01:01:44,770 --> 01:01:47,649
people, is not make people cheaper.

1101
01:01:48,069 --> 01:01:51,279
It's to build a society where
humans don't have to compete with

1102
01:01:51,279 --> 01:01:53,410
machines for the rights to exist.

1103
01:01:53,865 --> 01:01:57,345
So there are really three forces
capable of drawing the line

1104
01:01:57,375 --> 01:02:00,375
somewhere more human than wherever.

1105
01:02:00,375 --> 01:02:06,135
Profit dictates worker power, government
regulation, and universal protections

1106
01:02:06,285 --> 01:02:09,345
that decouple survival from employment.

1107
01:02:09,945 --> 01:02:12,885
All three have been
systematically weakened.

1108
01:02:13,755 --> 01:02:17,955
The PRO Act would extend organizing
rights to workers currently classified

1109
01:02:17,955 --> 01:02:21,915
out of the system and override
rights to work for less laws.

1110
01:02:22,395 --> 01:02:26,055
It's potentially the most important
piece of AI era legislation, and it's not

1111
01:02:26,055 --> 01:02:29,085
even about AI on the regulation front.

1112
01:02:29,265 --> 01:02:33,195
Over a thousand researchers and public
figures signed an open letter in

1113
01:02:33,195 --> 01:02:38,925
2023, calling for a pause on training
AI systems more powerful than GPT-4.

1114
01:02:39,345 --> 01:02:44,145
By October, 2025, pioneers like
Yoshua Bengio and Geoffrey Hinton

1115
01:02:44,235 --> 01:02:48,225
went further calling for an outright
prohibition on super intelligence

1116
01:02:48,225 --> 01:02:52,395
development until their scientific
consensus it can be done safely.

1117
01:02:53,025 --> 01:02:56,475
The people who built these systems
are begging for guardrails.

1118
01:02:56,925 --> 01:03:00,585
The response from industry and
government has been to accelerate.

1119
01:03:01,365 --> 01:03:05,415
Meanwhile, the US remains the only wealthy
democracy on earth without universal

1120
01:03:05,415 --> 01:03:10,035
healthcare and spending a fraction of
what pure countries spend on workforce

1121
01:03:10,035 --> 01:03:14,865
retraining the infrastructure that
would make AI disruption survivable

1122
01:03:14,865 --> 01:03:19,725
for ordinary people does not exist,
and none of these work in isolation.

1123
01:03:20,175 --> 01:03:23,775
Regulation without worker power
gets captured by lobbyists.

1124
01:03:24,225 --> 01:03:28,455
Attempts to build worker power without
universal protections leaves people

1125
01:03:28,485 --> 01:03:32,895
too scared to organize universal
protections implemented poorly is

1126
01:03:32,895 --> 01:03:36,345
likely to leave every deployment
decision in the hands of capital.

1127
01:03:36,765 --> 01:03:40,815
If you're a freelancer or a creative
worker, look into organizations in your

1128
01:03:40,815 --> 01:03:46,185
field, the Freelancers Union, the National
Writers Union, the Graphic Artist Guild.

1129
01:03:46,785 --> 01:03:50,145
If nothing exists in your line of
work, that itself is the problem.

1130
01:03:50,145 --> 01:03:53,415
This whole commentary is about
for everyone else, support labor

1131
01:03:53,415 --> 01:03:58,155
organizing, AI regulation, and universal
protections like Medicare for all.

1132
01:03:58,545 --> 01:04:02,895
There are three fronts in the same
fight over who gets to draw the line.

1133
01:04:03,765 --> 01:04:06,945
That copywriter who had no one to
call and no insurance to fall back on,

1134
01:04:07,725 --> 01:04:09,975
the question was never AI or no ai.

1135
01:04:10,185 --> 01:04:12,945
It was always who gets to
decide where this technology

1136
01:04:12,945 --> 01:04:14,595
belongs and where it doesn't.

1137
01:04:15,135 --> 01:04:18,435
The only workers who got a say
were the ones who organized to

1138
01:04:18,435 --> 01:04:22,305
demand one, and even the people who
built these systems are telling us

1139
01:04:22,515 --> 01:04:24,135
that won't be enough on its own.

1140
01:04:24,525 --> 01:04:29,535
We need rules and we need a floor
that nobody can fall through because

1141
01:04:29,595 --> 01:04:33,885
if your survival depends on your
employer's goodwill, you were never

1142
01:04:33,885 --> 01:04:35,895
really free to fight in the first place.

1143
01:04:37,485 --> 01:04:39,855
Note that we've begun putting my
commentaries on YouTube, so if

1144
01:04:39,855 --> 01:04:42,555
you find them insightful, check
out our channel and share them.

1145
01:04:42,765 --> 01:04:43,785
Link in the show notes.

1146
01:04:45,480 --> 01:04:48,510
And now we'll continue to dive
deeper on four topics today.

1147
01:04:48,720 --> 01:04:53,190
First up, section A, labor future
followed by section B, warfare,

1148
01:04:53,400 --> 01:04:57,270
section C, surveillance, and
Section D, corporate power.

1149
01:05:02,280 --> 01:05:07,620
We think of automation and technology
is being associated with job loss.

1150
01:05:07,620 --> 01:05:12,210
So how could an increase in that
technology not result in job loss?

1151
01:05:12,390 --> 01:05:12,630
Right.

1152
01:05:12,750 --> 01:05:16,530
And this is where I think people, by
focusing on only part of the picture,

1153
01:05:16,530 --> 01:05:18,360
they miss the overall picture.

1154
01:05:19,380 --> 01:05:23,850
What bringing in new techniques,
new machines, new technology does is

1155
01:05:23,850 --> 01:05:28,380
it oftentimes it displaces existing
labor by making it redundant.

1156
01:05:28,890 --> 01:05:29,130
Okay?

1157
01:05:29,130 --> 01:05:33,240
So that means that now there's a
loss of jobs and people think that

1158
01:05:33,300 --> 01:05:37,200
as the same technology proliferates
and it spreads across the economy,

1159
01:05:37,200 --> 01:05:40,680
it means it's gonna take up more and
more and more of other people's jobs.

1160
01:05:40,890 --> 01:05:43,260
And so unemployment will
increase and people will be

1161
01:05:43,320 --> 01:05:44,490
scrounging around for their jobs.

1162
01:05:45,090 --> 01:05:46,290
Now, there's two problems with this.

1163
01:05:47,070 --> 01:05:50,670
One, it is that it, it operates
with a picture in which the

1164
01:05:50,670 --> 01:05:54,720
total number of jobs is fixed in
the economy at any given point.

1165
01:05:55,290 --> 01:06:01,260
And over time that same number of jobs
is reduced by technology coming in.

1166
01:06:02,070 --> 01:06:05,730
But the most important fact
about capitalism is that the

1167
01:06:05,730 --> 01:06:07,140
economy is always growing.

1168
01:06:08,100 --> 01:06:12,360
So there is a constant process by
which as the economy grows, there's

1169
01:06:12,360 --> 01:06:17,580
also a demand for more jobs, sometimes
more jobs of the same kind, and other

1170
01:06:17,580 --> 01:06:21,660
times entirely new jobs, entirely new
occupations, which didn't exist before.

1171
01:06:22,200 --> 01:06:26,760
Now, when you take that dynamic aspect
into account, here's what you find.

1172
01:06:27,390 --> 01:06:32,040
As new technology comes in, it of
course displaces labor, but it's

1173
01:06:32,040 --> 01:06:33,840
also increasing productivity.

1174
01:06:34,410 --> 01:06:36,420
And what do the increases
in productivity mean?

1175
01:06:36,750 --> 01:06:40,620
What they mean is that the economy
potentially is growing even faster

1176
01:06:40,620 --> 01:06:41,760
as it becomes more efficient.

1177
01:06:42,555 --> 01:06:47,175
As it grows faster and becomes more
efficient, it's going to mean that there's

1178
01:06:47,175 --> 01:06:51,825
actually new plants opening up, new shops,
opening up new hotels, opening up, which

1179
01:06:51,825 --> 01:06:53,715
also creates a demand for more labor.

1180
01:06:54,195 --> 01:06:58,515
So one of the things that happens is
as people are displaced from one sector

1181
01:06:58,515 --> 01:07:02,775
of the economy or one factory where new
technology has come in, because demand

1182
01:07:02,775 --> 01:07:06,585
is growing in other sectors of the
economy, they're sucked up into new jobs.

1183
01:07:07,305 --> 01:07:09,525
So this is something Marx
pointed out very early on.

1184
01:07:10,185 --> 01:07:13,125
He says that there are two things
happening simultaneously in capitalism.

1185
01:07:13,725 --> 01:07:17,205
One is that technical change, which
is the term he uses for new machines

1186
01:07:17,205 --> 01:07:19,425
coming in, they displace labor.

1187
01:07:19,785 --> 01:07:23,535
But that same technical change also
increases the pace of economic growth.

1188
01:07:24,015 --> 01:07:27,915
And that means that there's a demand
for labor in the other sectors of

1189
01:07:27,915 --> 01:07:30,225
the economy that are growing faster.

1190
01:07:30,615 --> 01:07:33,705
And that labor that was displaced
is sucked up into the new ones.

1191
01:07:34,275 --> 01:07:35,715
Two things are happening at the same time.

1192
01:07:36,675 --> 01:07:40,125
What the fear of job loss, the
fear of technology is based on

1193
01:07:40,755 --> 01:07:43,755
is what economists call a partial
equilibrium model of the economy.

1194
01:07:43,815 --> 01:07:47,625
You're only looking at one sector
or one line, and you're projecting

1195
01:07:47,625 --> 01:07:50,205
onto the rest of the economy
what's happening in that one line.

1196
01:07:50,625 --> 01:07:55,485
But in fact, sectors are related
to each other and there's

1197
01:07:55,485 --> 01:07:56,865
what we call linkage effects.

1198
01:07:57,345 --> 01:07:59,955
What happens in one sector affect
what happens in other sectors.

1199
01:08:00,405 --> 01:08:04,425
So the technological change in one
can speed up the economy, and that

1200
01:08:04,425 --> 01:08:08,415
means also speeding up the rate of
labor absorption in other sectors.

1201
01:08:08,415 --> 01:08:12,855
People who are laid off in one sector get
picked up in other sectors, and this is

1202
01:08:12,855 --> 01:08:18,465
why over the decades in the centuries,
technological change has been a constant

1203
01:08:18,465 --> 01:08:23,024
in capitalism, enormous new technologies
have come in, waves of technology, have

1204
01:08:23,024 --> 01:08:28,035
displaced labor, and what you see is
jobs in particular sectors disappearing.

1205
01:08:28,590 --> 01:08:33,030
But there has never been an
overall trajectory or tendency

1206
01:08:33,330 --> 01:08:37,050
of unemployment increasing over
time as new technology comes in.

1207
01:08:37,410 --> 01:08:41,730
In fact, the history of capitalism
is of the level of unemployment,

1208
01:08:41,730 --> 01:08:44,820
basically remaining more or
less constant oscillating

1209
01:08:44,820 --> 01:08:46,260
around the rate of accumulation.

1210
01:08:47,070 --> 01:08:53,250
So in the aggregate, technological
change per se should not be feared.

1211
01:08:53,610 --> 01:08:57,630
It's how you deal with it, and it's
how the distributive consequences of

1212
01:08:57,630 --> 01:08:59,340
the technological change are picked up.

1213
01:08:59,850 --> 01:09:02,100
There's a lot to unpack
in what you just said.

1214
01:09:02,400 --> 01:09:06,330
Maybe we can start by looking
at the historical record.

1215
01:09:06,330 --> 01:09:11,940
Can you talk about, a
historical time period or technological

1216
01:09:11,940 --> 01:09:17,130
innovation in a sector within
capitalism that sort of encapsulates

1217
01:09:17,130 --> 01:09:18,600
the dynamics that you're laying out?

1218
01:09:18,810 --> 01:09:21,930
There had been waves of technological
change in the history of capitalism.

1219
01:09:21,930 --> 01:09:26,460
The one was the early 19th century
when textiles became automated and

1220
01:09:26,460 --> 01:09:28,380
spinning and weaving became automated.

1221
01:09:28,920 --> 01:09:32,520
Then there was the mid 19th century
when steam power took over factories

1222
01:09:32,730 --> 01:09:35,640
the later 19th century when electricity
and electrification came in.

1223
01:09:36,510 --> 01:09:39,090
You look at any one of these
times and what you find is two

1224
01:09:39,090 --> 01:09:40,590
things happening simultaneously.

1225
01:09:41,160 --> 01:09:45,300
People in the older professions
that are experiencing technological

1226
01:09:45,360 --> 01:09:50,730
shifts being pushed outta those
professions, but that same technological

1227
01:09:50,730 --> 01:09:52,650
shift generates two things.

1228
01:09:53,370 --> 01:09:56,850
Income increases because it means
that the economy is growing faster,

1229
01:09:56,880 --> 01:09:58,710
so that as demand increases.

1230
01:09:59,415 --> 01:10:04,725
Entirely new services and entirely
new jobs are created by the people

1231
01:10:04,725 --> 01:10:07,755
whose income is growing and the
workers who have been displaced in

1232
01:10:07,755 --> 01:10:09,585
one sector find jobs in the other.

1233
01:10:10,335 --> 01:10:14,445
The other thing that's happening is it all
kind of depends on where the technological

1234
01:10:14,445 --> 01:10:15,885
change, the automation is happening.

1235
01:10:16,365 --> 01:10:19,635
Suppose the automation is happening
in what's called capital goods

1236
01:10:19,635 --> 01:10:24,345
industries, industries that produce
machines that other industries buy

1237
01:10:24,345 --> 01:10:27,765
for them to make their products or
industries that produce raw materials

1238
01:10:27,765 --> 01:10:29,385
that other industries buy from them.

1239
01:10:29,925 --> 01:10:34,185
Now, if they become more efficient, then
the price of machines is going down, the

1240
01:10:34,185 --> 01:10:35,835
price of raw materials is going down.

1241
01:10:36,315 --> 01:10:42,195
As those prices go down, it makes the
production a lot more profitable for the

1242
01:10:42,435 --> 01:10:46,695
companies, the industries that buy those
raw materials now at the lower prices

1243
01:10:46,695 --> 01:10:49,035
that buy the machines at the lower prices.

1244
01:10:49,395 --> 01:10:54,765
So those downstream industries now produce
more, they invest more, and as they

1245
01:10:54,765 --> 01:10:57,135
expand faster, they suck in more labor.

1246
01:10:57,345 --> 01:10:57,435
Mm-hmm.

1247
01:10:57,735 --> 01:11:01,275
So there's two sources
of increases in jobs.

1248
01:11:01,365 --> 01:11:03,375
One is what we call an income effect.

1249
01:11:03,825 --> 01:11:08,505
As people's incomes go up, they raise
the level of aggregate demand, which

1250
01:11:08,505 --> 01:11:12,015
calls forth new jobs and new occupations.

1251
01:11:12,255 --> 01:11:15,915
Oftentimes the service sector, but also
in manufacturing, the other is what

1252
01:11:15,915 --> 01:11:21,075
you call a price effect, which is that
sectors which are using the cheaper

1253
01:11:21,075 --> 01:11:25,875
machinery if the technological change
was in the machine, good sector, the

1254
01:11:25,875 --> 01:11:29,445
sectors that are using the cheaper
machinery are able to expand faster, and

1255
01:11:29,445 --> 01:11:31,305
that sucks in more employment as well.

1256
01:11:31,785 --> 01:11:34,605
This goes back to what we
said about Marxist model.

1257
01:11:35,265 --> 01:11:36,825
Two things are happening simultaneously.

1258
01:11:37,035 --> 01:11:40,305
Technical change is throwing people into
what you call the reserve army of labor.

1259
01:11:41,295 --> 01:11:44,415
And the people in the reserve army
of labor are being sucked up into

1260
01:11:44,415 --> 01:11:49,245
new jobs because technological
change increases aggregate income.

1261
01:11:50,025 --> 01:11:51,735
Now, we don't know whose
income is being increased.

1262
01:11:51,764 --> 01:11:54,495
It might be the capitalist income,
it might be the worker's income, but

1263
01:11:54,495 --> 01:11:57,705
the economy doesn't care about that
aggregate income that's being increased.

1264
01:11:58,094 --> 01:12:01,934
The other thing technical change does
is it lowers the price of inputs, which

1265
01:12:01,934 --> 01:12:06,735
means that people using those inputs can
expand their production at a faster rate.

1266
01:12:07,005 --> 01:12:12,554
The aggregate result is jobs are being
lost in one sector, but employment

1267
01:12:12,554 --> 01:12:14,505
overall in the economy is not going down.

1268
01:12:14,924 --> 01:12:19,634
And that means the overall tempo, not
just of economic growth, but of employment

1269
01:12:19,695 --> 01:12:21,554
and employment absorption is going up.

1270
01:12:33,764 --> 01:12:41,355
What makes you confident that workers
will actually reap the monetary benefits

1271
01:12:41,415 --> 01:12:46,695
of using these tools rather than the
owners of the intellectual property?

1272
01:12:46,695 --> 01:12:51,315
Because I, I can see a, I can
see a dystopian future of just

1273
01:12:51,315 --> 01:12:57,464
massive rent seeking from the, a handful of AI giants.

1274
01:12:57,764 --> 01:12:58,065
Right?

1275
01:12:58,094 --> 01:12:58,964
So it's a good question.

1276
01:12:58,964 --> 01:13:01,455
So I'm speaking of a
possibility, not a certainty.

1277
01:13:01,605 --> 01:13:01,665
Yeah.

1278
01:13:01,665 --> 01:13:01,755
Mm-hmm.

1279
01:13:01,995 --> 01:13:04,785
I think there's a case where
we can do this, whether we

1280
01:13:04,785 --> 01:13:06,674
will is not, is not clear.

1281
01:13:07,155 --> 01:13:09,075
And I think there's two issues, right?

1282
01:13:09,075 --> 01:13:10,455
That both of which you've now raised.

1283
01:13:10,455 --> 01:13:12,945
One is of course, how are
the tools being designed?

1284
01:13:12,945 --> 01:13:15,735
Are they being designed to
basically replace people or to,

1285
01:13:15,974 --> 01:13:17,054
Leverage their expertise?

1286
01:13:17,115 --> 01:13:18,255
So that matters, right?

1287
01:13:18,255 --> 01:13:20,985
We have a lot of choice about how we use
the, they're incredibly flexible, right?

1288
01:13:20,985 --> 01:13:22,184
There's a million ways to use them.

1289
01:13:22,905 --> 01:13:24,165
And so that's one question.

1290
01:13:24,224 --> 01:13:27,554
And then the next one is, well, is the
market structure such that, that they

1291
01:13:27,554 --> 01:13:28,995
get the returns to that productivity?

1292
01:13:29,340 --> 01:13:33,060
Or not, and it depends, in my
opinion, substantially to whether that

1293
01:13:33,060 --> 01:13:35,370
expertise remains scarce and valuable.

1294
01:13:35,490 --> 01:13:35,760
Right?

1295
01:13:36,030 --> 01:13:38,760
Like no one thinks that,
people with bachelor's degrees,

1296
01:13:38,760 --> 01:13:41,670
certainly PhDs are underpaid
relative to what they do, right?

1297
01:13:41,670 --> 01:13:44,910
Why they have scarce expertise and you
know, and because of that, there's a

1298
01:13:44,910 --> 01:13:46,500
lot of competition for their services.

1299
01:13:46,590 --> 01:13:48,660
A good example would be, you
know, registered nurses, right?

1300
01:13:48,690 --> 01:13:49,740
For your degree.

1301
01:13:49,745 --> 01:13:53,640
And, and they pay, they make a, a, a
very decent income in the United States.

1302
01:13:53,700 --> 01:13:56,700
So I think having
expertise remains scarce.

1303
01:13:56,850 --> 01:13:59,280
I actually think a lot of
adjunct professors would

1304
01:13:59,280 --> 01:13:59,970
disagree with you.

1305
01:13:59,970 --> 01:14:00,030
Yeah.

1306
01:14:00,030 --> 01:14:00,480
Oh, you're right.

1307
01:14:00,480 --> 01:14:01,200
You're absolutely right.

1308
01:14:01,230 --> 01:14:03,210
But that's again about competition, right?

1309
01:14:03,390 --> 01:14:05,640
It's about market structure and the
concern you're raising, which is a

1310
01:14:05,640 --> 01:14:08,940
valid one, is maybe this will all
be monopolized by a few companies.

1311
01:14:09,180 --> 01:14:09,360
Yeah.

1312
01:14:09,385 --> 01:14:13,769
And, and I think that's a risk that
we have to be really vigilant about.

1313
01:14:13,769 --> 01:14:13,860
Yeah.

1314
01:14:13,950 --> 01:14:15,540
I don't, I, I don't
wanna take it as a given.

1315
01:14:15,540 --> 01:14:16,380
This all goes well.

1316
01:14:16,500 --> 01:14:19,560
The case you're making, of
course, is plausible or possible.

1317
01:14:19,950 --> 01:14:24,810
And what I've said a million times
is, is that we cannot fear technology.

1318
01:14:25,620 --> 01:14:26,130
We have to.

1319
01:14:26,130 --> 01:14:29,640
Fear is an economic system
like neoliberalism, right?

1320
01:14:29,640 --> 01:14:33,810
That privatizes all the benefits
of technology and socializes

1321
01:14:33,810 --> 01:14:36,930
all of the disadvantages of
technology where that's right.

1322
01:14:36,930 --> 01:14:39,120
A few people, the
top get all the benefits and

1323
01:14:39,120 --> 01:14:40,470
everybody else gets screwed.

1324
01:14:40,890 --> 01:14:47,220
And I guess given history, take
truck drivers at one time, one

1325
01:14:47,220 --> 01:14:49,830
person could move very little.

1326
01:14:50,500 --> 01:14:54,180
In terms of goods from one place in
the country to another place in the

1327
01:14:54,180 --> 01:14:56,460
country, because they would either
have to carry it on their back or

1328
01:14:56,520 --> 01:14:58,170
use a donkey or something like that.

1329
01:14:58,170 --> 01:15:03,330
And now a big truck can move a lot of
goods from one place, to another.

1330
01:15:03,660 --> 01:15:08,010
But truck drivers are very poorly paid
generally, even though they are creating

1331
01:15:08,010 --> 01:15:14,070
a huge amount of value relative to the
alternative, that that's a consequence of

1332
01:15:14,070 --> 01:15:15,960
the fact that truck drivers have no power,

1333
01:15:16,290 --> 01:15:16,440
right?

1334
01:15:16,440 --> 01:15:16,860
That they have No,

1335
01:15:16,860 --> 01:15:19,440
we de we deregulated
the trucking industry.

1336
01:15:19,440 --> 01:15:19,500
Yeah.

1337
01:15:19,500 --> 01:15:25,590
In fact, truck drivers earned a middle
class wage before deregulation and

1338
01:15:25,620 --> 01:15:28,920
Yeah, like likewise,
administrative assistant, right?

1339
01:15:28,920 --> 01:15:33,540
I mean, before computers,
that job was x productivity.

1340
01:15:33,780 --> 01:15:37,890
With a modern computer, you can
obviously accomplish a heck of a lot more,

1341
01:15:37,890 --> 01:15:41,850
do 10 times as much work really relative
to what you could with a typewriter.

1342
01:15:42,180 --> 01:15:45,210
But those folks don't earn more today.

1343
01:15:45,210 --> 01:15:47,040
They earn less than they used to.

1344
01:15:47,160 --> 01:15:50,760
Their wages have compressed
down over the last 50 years.

1345
01:15:50,910 --> 01:15:54,240
By the way, this is true of
people with bachelor degrees too.

1346
01:15:54,660 --> 01:15:59,310
In general, bachelor, people with bachelor
degrees, their wages have e they're

1347
01:15:59,310 --> 01:16:01,260
either plateaued or, or decreased.

1348
01:16:01,260 --> 01:16:05,220
They're, this is the story of
the last 50 years, is that the only people

1349
01:16:05,220 --> 01:16:09,330
who really benefited from economic growth
were the people in the top one or 2%.

1350
01:16:09,360 --> 01:16:12,180
I, I, I'm not agreeing with
that by the way, but okay.

1351
01:16:12,180 --> 01:16:12,480
Top,

1352
01:16:12,595 --> 01:16:12,815
top

1353
01:16:13,050 --> 01:16:13,560
10%,

1354
01:16:14,250 --> 01:16:18,000
no, I basically people, people who have
a college degree on up, which is 40%

1355
01:16:18,060 --> 01:16:21,540
have have seen real income growth, but
there's been a lot of stagnation and,

1356
01:16:21,810 --> 01:16:24,720
but they have it, but they have not
maintained their same share of GDP.

1357
01:16:25,200 --> 01:16:25,410
Okay.

1358
01:16:25,410 --> 01:16:27,295
So this is actually, this is a
point I wanted to make, so, yeah.

1359
01:16:27,390 --> 01:16:27,660
Yeah.

1360
01:16:27,690 --> 01:16:30,600
So, six out of every $10
in the US economy approximately

1361
01:16:30,600 --> 01:16:31,740
is first paid to workers.

1362
01:16:32,250 --> 01:16:35,400
And it's fallen somewhat, although
not, it's still pretty high.

1363
01:16:35,400 --> 01:16:38,370
What's actually remarkable, and this, I
think many people find this surprising is.

1364
01:16:38,685 --> 01:16:40,665
You might think in the richest
countries, the countries that

1365
01:16:40,665 --> 01:16:43,485
have the most capital, the most
technology, the most infrastructure,

1366
01:16:43,485 --> 01:16:46,125
that labor share of economic
activity would be relatively small.

1367
01:16:46,575 --> 01:16:47,655
But in fact, it's larger, right?

1368
01:16:47,655 --> 01:16:51,345
So the world averaged about 50% in
parts of North Africa, it's 45%.

1369
01:16:51,345 --> 01:16:51,435
Mm-hmm.

1370
01:16:51,885 --> 01:16:56,025
One of the amazing things actually about
the kind of industrialized countries is

1371
01:16:56,085 --> 01:16:59,985
even as we've gotten all this technology,
we've made our labor more valuable, not

1372
01:16:59,985 --> 01:17:01,545
just in terms of we pay higher wages.

1373
01:17:01,995 --> 01:17:04,875
A larger share of all that
activity has gone to workers.

1374
01:17:05,085 --> 01:17:07,815
Now we should be concerned about
that declining, but it's still

1375
01:17:07,815 --> 01:17:09,195
actually relatively high here.

1376
01:17:09,195 --> 01:17:13,335
And I, I would argue in the
case of truckers, this is,

1377
01:17:13,845 --> 01:17:17,565
part of the reason wages haven't risen
with productivity, you can make the same

1378
01:17:17,625 --> 01:17:21,345
argument about cashiers or many activities
is 'cause the labor is not scarce.

1379
01:17:21,345 --> 01:17:21,525
Right.

1380
01:17:21,525 --> 01:17:25,275
The people who can do it are, are
the capability to do it is abundant.

1381
01:17:25,365 --> 01:17:31,065
And, and that's why I really focus on this
notion of what makes expertise valuable.

1382
01:17:31,245 --> 01:17:34,335
And the, the problem that we have
is that people who do not have,

1383
01:17:34,995 --> 01:17:36,460
or a problem that we have Yeah.

1384
01:17:36,465 --> 01:17:38,775
Is that the majority of people who do
not have a college degree have been

1385
01:17:38,955 --> 01:17:43,305
pushed downward into non-specialized
work, clerical work, production work

1386
01:17:43,485 --> 01:17:46,725
that required real skills and knowledge
and training that made it,

1387
01:17:46,725 --> 01:17:48,135
valuable over long periods of time.

1388
01:17:48,525 --> 01:17:52,095
And when it, in a world in which
people are doing work that many,

1389
01:17:52,095 --> 01:17:54,975
many others can do, it's not, it's,
it's very difficult for that to pay.

1390
01:17:54,975 --> 01:17:57,165
Well now I don't want to make
the case that it can't pay.

1391
01:17:57,165 --> 01:17:59,685
Well obviously it pays more in
some countries than others, and

1392
01:17:59,685 --> 01:18:02,144
that's a function of the way
they structure their economies.

1393
01:18:02,265 --> 01:18:03,075
Well, power.

1394
01:18:03,165 --> 01:18:03,585
Yeah, sure.

1395
01:18:04,425 --> 01:18:07,725
I mean, there, some people who work
at McDonald's make $20 an hour.

1396
01:18:07,965 --> 01:18:10,455
Some people who work at McDonald's
doing exactly the same job

1397
01:18:10,545 --> 01:18:12,375
make $7 and 25 cents an hour.

1398
01:18:12,644 --> 01:18:12,795
Yeah.

1399
01:18:12,795 --> 01:18:13,455
Mm-hmm.

1400
01:18:13,610 --> 01:18:15,105
That is not a reflection of

1401
01:18:15,525 --> 01:18:15,765
Right.

1402
01:18:15,855 --> 01:18:16,425
Rarity.

1403
01:18:16,485 --> 01:18:20,085
That's a, that's a function of
we somehow delivered more power

1404
01:18:20,085 --> 01:18:23,475
to the people who were earning 20
than the people who earned 7 25.

1405
01:18:23,475 --> 01:18:26,265
I mean, I'd say, so in
almost all countries the sort of

1406
01:18:26,265 --> 01:18:30,225
the ranking of earnings across
occupations activities is very similar.

1407
01:18:30,285 --> 01:18:30,345
Yeah.

1408
01:18:30,405 --> 01:18:32,535
But the disparities vary enormously.

1409
01:18:32,715 --> 01:18:33,115
Right, right.

1410
01:18:33,435 --> 01:18:36,375
So in, in
Scandinavia or even in Canada.

1411
01:18:36,765 --> 01:18:40,275
Wages substantially higher
for people doing that same work.

1412
01:18:40,305 --> 01:18:40,395
Mm-hmm.

1413
01:18:40,635 --> 01:18:43,845
And that is, that is not
an economic inevitability, that's

1414
01:18:43,845 --> 01:18:45,375
a, a choice of power if you say.

1415
01:18:45,435 --> 01:18:45,615
Right.

1416
01:18:45,795 --> 01:18:47,595
What is, how is the
social system organized?

1417
01:18:47,955 --> 01:18:51,345
So that, again, going back to these two
questions, one is, can we make, expand?

1418
01:18:51,345 --> 01:18:53,865
Can we augment the value of expertise?

1419
01:18:53,895 --> 01:18:58,275
Can we enable more people to do
compete in these high paid activities?

1420
01:18:58,275 --> 01:19:02,085
It won't necessarily be good for doctors
or lawyers or software coders, but if more

1421
01:19:02,085 --> 01:19:05,625
people who are not the elite were able to
do that work, it would be better for them.

1422
01:19:05,745 --> 01:19:07,755
And it would also bring down
the price of goods and services.

1423
01:19:07,755 --> 01:19:09,075
So it would improve the quality of jobs.

1424
01:19:09,075 --> 01:19:12,135
For some, it would improve the, availability of a lot of

1425
01:19:12,135 --> 01:19:13,365
expensive stuff for others.

1426
01:19:13,425 --> 01:19:13,665
Yes.

1427
01:19:13,665 --> 01:19:16,725
But then there is the question that was, just asked, well, what, what's

1428
01:19:16,725 --> 01:19:18,375
to prevent firms from monopolizing that?

1429
01:19:18,495 --> 01:19:21,435
And it depends on healthy
labor market competition.

1430
01:19:21,495 --> 01:19:22,575
It, it absolutely does.

1431
01:19:22,575 --> 01:19:26,595
And that means the way we structure
regulations, the ability, the.

1432
01:19:27,000 --> 01:19:31,260
Worker voice and institutions that
protect that labor standards and social

1433
01:19:31,260 --> 01:19:32,820
norms about what we view as tolerable.

1434
01:19:32,910 --> 01:19:33,090
Yeah.

1435
01:19:33,300 --> 01:19:34,980
So I don't think it's either or.

1436
01:19:35,550 --> 01:19:37,890
I think we, we need to
be productive ultimately.

1437
01:19:37,890 --> 01:19:41,640
How much we can distribute depends
on how much we produce, and we want

1438
01:19:41,640 --> 01:19:45,630
to use these tools well, but we don't
wanna use them in a way that just makes

1439
01:19:45,630 --> 01:19:47,730
human labor unnecessary or superfluous.

1440
01:19:48,060 --> 01:19:51,720
We want to use it in a way that
makes more people's judgment

1441
01:19:51,990 --> 01:19:56,310
and knowledge and expertise more
applicable, across valuable things.

1442
01:20:12,330 --> 01:20:19,920
Does that contradict any of the processes
that Marks and Braverman outline?

1443
01:20:20,250 --> 01:20:24,480
I don't think so, and I, it's a
common objection that's brought up and

1444
01:20:24,480 --> 01:20:26,430
I think it rests on a misconception.

1445
01:20:26,970 --> 01:20:28,950
Part of that, I think
Braverman is himself to blame.

1446
01:20:29,250 --> 01:20:34,350
I think the word de-skilling was a
very unfortunate choice to describe the

1447
01:20:34,350 --> 01:20:38,400
processes that they were analyzing because
when you say that work is becoming,

1448
01:20:38,400 --> 01:20:43,320
workers are becoming de-skilled, it's easy
to interpret that as meaning unskilled.

1449
01:20:43,860 --> 01:20:46,860
Now, de-skilling and being unskilled
are two very different things.

1450
01:20:47,250 --> 01:20:51,390
As I said earlier, what
de-skilling describes is taking

1451
01:20:51,390 --> 01:20:55,620
highly complex, multidimensional
tasks and breaking them down.

1452
01:20:56,490 --> 01:21:00,690
That's not so much de-skilling as it is
simplification, and I think simplification

1453
01:21:00,690 --> 01:21:02,040
would've been a better term to use.

1454
01:21:02,250 --> 01:21:05,070
What's happening is work
is becoming simplified now.

1455
01:21:05,610 --> 01:21:06,390
Commonsensically.

1456
01:21:07,170 --> 01:21:11,910
It's possible for very, very
complicated tasks to be broken down

1457
01:21:11,910 --> 01:21:16,290
into their different parts, but each
part still requiring a certain amount

1458
01:21:16,290 --> 01:21:17,730
of training and skill to carry out.

1459
01:21:17,910 --> 01:21:19,320
So think of coding, for example.

1460
01:21:19,320 --> 01:21:20,370
Think of software engineering.

1461
01:21:21,270 --> 01:21:24,990
Interestingly, if you look at, when
software engineering came about, say

1462
01:21:24,990 --> 01:21:29,190
40 years ago, it was very artisanal
software engineers needed to have a lot

1463
01:21:29,190 --> 01:21:30,660
of skills of which coding was just one.

1464
01:21:31,470 --> 01:21:35,190
Over the decades since then, it has
been broken down into individual parts,

1465
01:21:35,490 --> 01:21:40,860
but each individual part still requires
a college education, or at the very

1466
01:21:40,860 --> 01:21:43,050
least, a, a high school education.

1467
01:21:43,050 --> 01:21:43,170
Mm-hmm.

1468
01:21:43,530 --> 01:21:48,420
Now, 80 years ago, or even 60 years
ago, the definition of skilled work

1469
01:21:48,420 --> 01:21:50,970
in the United States, the official
definition was anything that

1470
01:21:50,970 --> 01:21:52,410
requires a high school education.

1471
01:21:53,220 --> 01:21:56,610
Today, a high school education is
what you use for unskilled work.

1472
01:21:56,730 --> 01:21:57,120
Right,

1473
01:21:57,125 --> 01:21:57,335
right.

1474
01:21:57,570 --> 01:22:00,330
Well, and I think the, the other
thing I was thinking about when you were

1475
01:22:00,330 --> 01:22:05,670
talking is even if work is simplified in
some way, it, it often requires the use

1476
01:22:05,670 --> 01:22:08,130
of skills that weren't necessary before.

1477
01:22:08,130 --> 01:22:13,440
Many jobs that might
have been more manual in the past.

1478
01:22:13,710 --> 01:22:17,700
Now you have to have a basic
understanding of how to use computers

1479
01:22:17,700 --> 01:22:19,325
in order to do That's exactly right.

1480
01:22:19,325 --> 01:22:21,870
And that's considered,
a quote unquote unskilled job.

1481
01:22:21,870 --> 01:22:23,160
But the reality is it's not.

1482
01:22:23,220 --> 01:22:23,490
Right.

1483
01:22:23,820 --> 01:22:27,420
So what that tells us is what
counts as skilled and unskilled

1484
01:22:27,420 --> 01:22:29,280
officially changes over time.

1485
01:22:29,850 --> 01:22:35,070
And today, what counts as unskilled
requires a level of education that 60, 70

1486
01:22:35,070 --> 01:22:38,010
years ago would've made for skilled jobs.

1487
01:22:38,670 --> 01:22:41,400
So there's no doubt that
the baseline level of, let's

1488
01:22:41,400 --> 01:22:43,320
say, knowledge and education.

1489
01:22:44,010 --> 01:22:47,550
That's required in jobs today,
even in unskilled jobs, is greater

1490
01:22:47,550 --> 01:22:48,510
than it was a hundred years ago.

1491
01:22:49,080 --> 01:22:52,620
Now, if that's what Marx of Braverman
had meant by de-skilling, then of

1492
01:22:52,620 --> 01:22:53,700
course the theory would be wrong.

1493
01:22:54,660 --> 01:22:55,920
But that's not what they meant.

1494
01:22:56,220 --> 01:23:02,220
What they meant was the over time,
complicated, multi-dimensional jobs

1495
01:23:02,220 --> 01:23:08,040
being broken down into individual
tasks, jobs into tasks, and each task.

1496
01:23:08,430 --> 01:23:13,560
If you compared a basic task in a
manufacturing plant today, an individual

1497
01:23:13,560 --> 01:23:18,480
task, which by our definitions is not
very skilled, it still requires much

1498
01:23:18,480 --> 01:23:23,280
more education than a basic task required
80 years ago in a manufacturing plant.

1499
01:23:23,430 --> 01:23:27,780
So what gives, well, what's happening
over time is that while the baseline

1500
01:23:27,780 --> 01:23:33,360
level of education is going up, the
process of breaking down complicated

1501
01:23:33,360 --> 01:23:37,830
tasks into simpler ones is still
very much driving the whole thing.

1502
01:23:38,610 --> 01:23:40,530
Now, let me put this together
with what I said last week.

1503
01:23:40,530 --> 01:23:46,020
Remember we said that whenever new
technology comes around, it doesn't

1504
01:23:46,020 --> 01:23:50,580
just eliminate a big chunk of the
existing jobs, it also creates

1505
01:23:50,580 --> 01:23:54,660
a demand for new jobs and new
occupations and new tasks, right?

1506
01:23:54,660 --> 01:23:56,040
So two things are happening at once.

1507
01:23:57,090 --> 01:24:01,920
Technology is destroying old
jobs, old tasks, and it's creating

1508
01:24:01,920 --> 01:24:04,590
a complimentary need for new
ones that didn't exist before.

1509
01:24:04,980 --> 01:24:09,720
Now a lot of times, those new
tasks, those new jobs are artisanal

1510
01:24:10,980 --> 01:24:12,210
engineering was a very good example.

1511
01:24:12,210 --> 01:24:13,890
Software engineering was a
very good example of that.

1512
01:24:14,340 --> 01:24:15,690
They're artisanal when they're born,

1513
01:24:16,260 --> 01:24:16,530
right?

1514
01:24:16,860 --> 01:24:22,830
But over time, management finds every
time that as long as workers monopolize

1515
01:24:22,830 --> 01:24:26,460
those skills, they're harder to control
because they're harder to control.

1516
01:24:26,610 --> 01:24:30,510
Management commits itself to
breaking through that skill barrier.

1517
01:24:31,275 --> 01:24:33,735
And it does that through
simplifying those tasks.

1518
01:24:34,095 --> 01:24:38,325
Each one of those tasks on its own can
still require a big amount of education.

1519
01:24:38,625 --> 01:24:44,295
But the key thing is the simplification of
an erstwhile complicated task is going on.

1520
01:24:44,685 --> 01:24:46,455
So you should think of
capitalism in this way.

1521
01:24:47,865 --> 01:24:52,155
It keeps throwing up new
jobs, new occupations because

1522
01:24:52,155 --> 01:24:53,205
of technological change.

1523
01:24:54,075 --> 01:24:57,375
Those new jobs and new
occupations typically are born

1524
01:24:57,885 --> 01:25:00,435
as a set of complicated skills.

1525
01:25:01,425 --> 01:25:02,205
Over time.

1526
01:25:02,505 --> 01:25:06,495
Management then sees, okay, we broke
apart the last set of skills, now we

1527
01:25:06,495 --> 01:25:08,115
have to break apart this set of skills.

1528
01:25:08,715 --> 01:25:13,875
So there's a dynamic process by which
existing complicated jobs are being broken

1529
01:25:13,875 --> 01:25:19,095
down into individual tasks and therefore
simplified while new occupations are

1530
01:25:19,095 --> 01:25:23,475
born, which are complicated ones, which
now become the new barrier, the new

1531
01:25:23,475 --> 01:25:26,535
frontier that employers try to break down.

1532
01:25:27,405 --> 01:25:32,085
This is all happening even while
the general level of education

1533
01:25:32,145 --> 01:25:33,765
that's needed for workers goes up.

1534
01:25:34,035 --> 01:25:34,125
Mm-hmm.

1535
01:25:34,455 --> 01:25:39,105
So each task at any given moment
does require education, but each task

1536
01:25:39,105 --> 01:25:42,165
at any given moment is being broken
down and simplified as we go on.

1537
01:25:42,495 --> 01:25:48,945
So the word de-skilling
unfortunately obscures this fine.

1538
01:25:49,005 --> 01:25:51,135
That just means that people have
to read a little more carefully.

1539
01:25:51,315 --> 01:25:53,535
And unfortunately people
don't always do that.

1540
01:25:54,345 --> 01:25:54,465
Well.

1541
01:25:54,465 --> 01:25:55,485
They can just listen to our show.

1542
01:25:56,655 --> 01:26:01,245
I think that it's been 50 years
since the book was written, and it's

1543
01:26:01,245 --> 01:26:03,345
remarkable how prescient it was.

1544
01:26:03,345 --> 01:26:06,165
Just like what Marks wrote
150 years ago was prescient.

1545
01:26:06,615 --> 01:26:13,605
The present history of capitalism today
is still a dynamic process of creating new

1546
01:26:13,605 --> 01:26:18,105
jobs, breaking them down so that employers
can control their workers better.

1547
01:26:18,585 --> 01:26:21,045
And workers trying to resist
this as best they can.

1548
01:26:22,215 --> 01:26:24,795
Why are struggles in the labor movement?

1549
01:26:25,200 --> 01:26:29,880
Typically over distributional
questions like wages and hours of

1550
01:26:29,880 --> 01:26:34,890
work, as opposed to questions about
who controls the labor process.

1551
01:26:34,890 --> 01:26:36,660
The, the questions
we're talking about now.

1552
01:26:36,900 --> 01:26:39,510
Yeah, I think that's a
fair observation actually.

1553
01:26:39,810 --> 01:26:44,160
It is true that if you look for
a common denominator in all union

1554
01:26:44,160 --> 01:26:48,570
movements, wages are probably
the one that stand out the most.

1555
01:26:48,570 --> 01:26:53,400
And I think the reason for that is
yearend capitalism, money is a solution

1556
01:26:53,400 --> 01:26:56,670
to a lot of problems and money's
also the cause of a lot of problems.

1557
01:26:56,670 --> 01:27:00,330
So no matter what else, you are
going to take up the wage question.

1558
01:27:01,230 --> 01:27:05,520
It's also a question that
workers can come together around.

1559
01:27:06,360 --> 01:27:11,100
The problem in unions is that when
you have a union meeting or when you

1560
01:27:11,100 --> 01:27:15,450
have the committee trying to come
up with a contract and the demands

1561
01:27:15,450 --> 01:27:19,170
of the contract, workers
have a lot of different needs.

1562
01:27:19,260 --> 01:27:21,960
They have a lot of different
demands, and they're all important.

1563
01:27:22,050 --> 01:27:24,060
Some workers need more vacation time.

1564
01:27:24,060 --> 01:27:27,000
Some workers are worried about
pension, some workers need free

1565
01:27:27,000 --> 01:27:28,590
child support, et cetera, et cetera.

1566
01:27:28,590 --> 01:27:29,940
And you're not gonna win every battle.

1567
01:27:29,940 --> 01:27:31,110
You can't make every demand.

1568
01:27:31,800 --> 01:27:36,030
The one thing that makes everything
else easier for everybody is more money.

1569
01:27:36,840 --> 01:27:39,690
So it is a way of bringing
people together more easily.

1570
01:27:40,140 --> 01:27:41,820
It is the common denominator, et cetera.

1571
01:27:42,960 --> 01:27:48,270
That said, I think it is also important
to acknowledge that in every union

1572
01:27:48,270 --> 01:27:51,960
movement, there have also been
very close behind the wage issue,

1573
01:27:52,020 --> 01:27:54,030
other demands that are very common.

1574
01:27:54,270 --> 01:27:56,940
So the pace of work is very, very common.

1575
01:27:57,600 --> 01:28:02,910
Limiting, arbitrary authority of the
managers over the workers is very common.

1576
01:28:02,910 --> 01:28:03,000
Mm-hmm.

1577
01:28:03,240 --> 01:28:04,440
Now, how do they do that?

1578
01:28:04,710 --> 01:28:10,620
Well, seniority rules are a way of
making sure that management doesn't have.

1579
01:28:11,054 --> 01:28:13,424
Unilateral control over
everything that's happening.

1580
01:28:13,575 --> 01:28:14,835
Why do workers want seniority?

1581
01:28:14,894 --> 01:28:20,715
They want a predictable path to getting
promoted and a predictable set of rules

1582
01:28:20,715 --> 01:28:24,525
about who gets fired and who doesn't,
because that way management doesn't

1583
01:28:24,525 --> 01:28:26,205
play favorites, but this or that worker.

1584
01:28:27,105 --> 01:28:27,375
Right?

1585
01:28:27,375 --> 01:28:31,065
Well, hand in hand with that is the
grievance process, which is in union

1586
01:28:31,065 --> 01:28:37,485
contracts, but at least in this country,
most of us live under essentially Right.

1587
01:28:37,485 --> 01:28:37,875
To work.

1588
01:28:37,875 --> 01:28:38,085
Right.

1589
01:28:38,085 --> 01:28:43,965
Like your employer can fire you
without any reason whatsoever.

1590
01:28:43,995 --> 01:28:44,054
Yeah.

1591
01:28:44,085 --> 01:28:45,134
And that is allowed,

1592
01:28:45,224 --> 01:28:45,434
right?

1593
01:28:45,705 --> 01:28:49,094
So when you take a step back and you look
at these demands like, seniority,

1594
01:28:49,099 --> 01:28:51,375
grievances, workplace rules, what is that?

1595
01:28:51,375 --> 01:28:54,915
That's essentially trying to control
the labor process or at least trying

1596
01:28:54,915 --> 01:28:56,205
to control the workplace itself.

1597
01:28:56,205 --> 01:28:56,445
Right?

1598
01:28:56,985 --> 01:29:00,495
So you're right that wages
are the most common element.

1599
01:29:01,065 --> 01:29:02,085
It's not by a lot.

1600
01:29:02,174 --> 01:29:05,625
These are the things also are
really, really important and unions

1601
01:29:05,625 --> 01:29:07,545
everywhere fight for these issues.

1602
01:29:07,905 --> 01:29:13,335
So this is what Braverman is
tapping into and Marks obviously.

1603
01:29:13,394 --> 01:29:13,485
Mm-hmm.

1604
01:29:13,844 --> 01:29:19,455
When they ask what are the constants
in the employment relation, why is

1605
01:29:19,455 --> 01:29:23,325
it that management is constantly
trying to exert this kind of control?

1606
01:29:23,835 --> 01:29:28,304
And workers find various ways of taking
that control back from management.

1607
01:29:28,905 --> 01:29:34,875
It's because the control is always used
to prioritize profits over people, and the

1608
01:29:34,875 --> 01:29:41,054
workers are people and they try to limit
what that authority is on management side.

1609
01:29:41,505 --> 01:29:44,655
That's true everywhere Regard in
every region regards their religion

1610
01:29:44,655 --> 01:29:45,945
or culture or anything like that.

1611
01:29:58,175 --> 01:30:01,530
I don't wanna use the term crisis,
collapse, decline, whatever.

1612
01:30:01,530 --> 01:30:04,170
It, it certainly doesn't
have to be a crisis.

1613
01:30:04,260 --> 01:30:08,460
You emphasize it's the transition
that's the, yeah, that's the difficulty.

1614
01:30:08,460 --> 01:30:14,880
Not, not the smaller global population,
but the transition to it, as a, as

1615
01:30:14,880 --> 01:30:20,910
a student of history and technology can
broadly say that new technologies in

1616
01:30:20,910 --> 01:30:22,650
the long run have proven beneficial.

1617
01:30:22,680 --> 01:30:27,360
Like we have a much higher living
standard now than we did 200 years

1618
01:30:27,360 --> 01:30:31,440
ago, and we're all better off for, in
commensurately for that transition.

1619
01:30:31,440 --> 01:30:37,980
But in the moment that transition
eviscerated people, it was very difficult.

1620
01:30:37,980 --> 01:30:39,780
And we all live in the moment.

1621
01:30:40,110 --> 01:30:45,060
So the people who are going to live
through the AI revolution, a lot of

1622
01:30:45,060 --> 01:30:49,140
them risk being displaced and derated.

1623
01:30:49,350 --> 01:30:54,240
What are the policies that we
need to focus on to get through

1624
01:30:54,240 --> 01:30:58,890
the transition without, I don't
know, undermining democracy?

1625
01:31:00,810 --> 01:31:02,250
Just to put a fine point on it.

1626
01:31:02,250 --> 01:31:02,310
Yeah.

1627
01:31:03,750 --> 01:31:03,930
Yeah.

1628
01:31:03,930 --> 01:31:07,170
So, so first, let me agree
with you that even if the long

1629
01:31:07,170 --> 01:31:11,070
run outcome is a good one, the
transitions are often really costly.

1630
01:31:11,130 --> 01:31:13,680
I mean, like the transition into
the industrial revolution, working

1631
01:31:13,680 --> 01:31:18,990
class incomes did not rise in Britain
for the first 60 years mm-hmm.

1632
01:31:18,990 --> 01:31:20,280
Of the industrial revolution, right?

1633
01:31:20,280 --> 01:31:22,950
And so people were mis rated and even
when they started to rise, they're still,

1634
01:31:23,010 --> 01:31:24,210
there's lots of people who are displaced.

1635
01:31:24,210 --> 01:31:26,850
So people are on the upside, are usually
not the ones on the downside, right?

1636
01:31:26,850 --> 01:31:31,020
Like Pittsburgh has reconstituted itself
after losing steel and coal, right?

1637
01:31:31,260 --> 01:31:35,340
But the people in Pittsburgh who are doing
well now, are people in education, people

1638
01:31:35,340 --> 01:31:38,580
in healthcare, they're not former steel
workers or former coal workers, right?

1639
01:31:38,850 --> 01:31:41,460
So even if you think it's all gonna
work out great, that doesn't mean

1640
01:31:41,490 --> 01:31:43,050
every person is going to benefit.

1641
01:31:43,260 --> 01:31:46,290
But then there, there's an even harder
problem, let's say it's all

1642
01:31:46,290 --> 01:31:49,470
gonna be great for people in the long
run, but not in the first 20 years.

1643
01:31:49,845 --> 01:31:52,514
I think that's a very,
very serious challenge.

1644
01:31:52,545 --> 01:31:55,245
And so part of it depends on
how fast it changes, right?

1645
01:31:55,695 --> 01:31:56,804
And this is a very important, right?

1646
01:31:56,804 --> 01:32:00,105
Like if you told me, 20
years from now we all trucks will

1647
01:32:00,105 --> 01:32:03,315
be self-driving, I'd say, well, we
can manage that 20 year transition.

1648
01:32:03,315 --> 01:32:04,785
People will retire, people won't enter.

1649
01:32:04,785 --> 01:32:07,275
If you told me that's gonna happen
on, new Year's Day, I

1650
01:32:07,275 --> 01:32:08,835
would say that's a crisis, right?

1651
01:32:09,014 --> 01:32:10,695
So part of it's how fast it happens.

1652
01:32:10,965 --> 01:32:14,175
But then the other of course is
the support systems that we have to

1653
01:32:14,175 --> 01:32:15,825
enable people to thrive and rebound.

1654
01:32:15,885 --> 01:32:18,585
And that's is something where
the uni, the US falls uniquely

1655
01:32:18,585 --> 01:32:20,655
short among developed countries.

1656
01:32:20,655 --> 01:32:25,875
We spend, one 10th as much
per capita or per GDP as does, for

1657
01:32:25,875 --> 01:32:29,745
example, Denmark on retraining people
and getting 'em back in the workforce.

1658
01:32:29,775 --> 01:32:33,825
We used to believe that we had a,
a, a national job security program.

1659
01:32:33,825 --> 01:32:35,085
It was called Get a New Job.

1660
01:32:35,144 --> 01:32:38,474
And we've learned that that just
isn't such a good program and that's

1661
01:32:38,474 --> 01:32:39,795
why I worry the most about the us.

1662
01:32:39,795 --> 01:32:47,415
I think the US does so little to
buffer against the damaging forces of

1663
01:32:47,415 --> 01:32:51,285
inequality and the scars they leave,
that it actually inhibits economic

1664
01:32:51,285 --> 01:32:53,175
mobility, inhibits opportunity.

1665
01:32:53,535 --> 01:32:56,175
It's not just cruel,
it's also inefficient.

1666
01:32:56,595 --> 01:32:56,655
Yeah.

1667
01:32:57,105 --> 01:33:00,885
We have a, we have a thriving
meritocracy from the median on up.

1668
01:33:01,455 --> 01:33:04,424
And then for everybody else, we
underinvest in their skills and their

1669
01:33:04,424 --> 01:33:07,394
security and their healthcare and
their safety and their education.

1670
01:33:07,545 --> 01:33:09,315
I think that's the danger that we run.

1671
01:33:09,315 --> 01:33:13,335
And, survey evidence says
that US workers are more nervous about

1672
01:33:13,335 --> 01:33:16,005
technology than people in other countries.

1673
01:33:16,394 --> 01:33:17,174
And you say, why?

1674
01:33:17,174 --> 01:33:19,094
We're creating this
technology, why shouldn't we love it?

1675
01:33:19,094 --> 01:33:20,715
And the the answer of course
is what you said earlier.

1676
01:33:20,745 --> 01:33:22,605
'cause they recognize the economic system.

1677
01:33:22,875 --> 01:33:24,195
It's not they're afraid of the technology.

1678
01:33:24,195 --> 01:33:26,445
They're afraid of the cost
they'll personally pay.

1679
01:33:26,445 --> 01:33:29,565
The science fiction
writer Ted Chung said,

1680
01:33:29,835 --> 01:33:33,554
fear of technology is not fear of the
machines, it's fear of capitalism.

1681
01:33:33,735 --> 01:33:33,855
Yeah.

1682
01:33:33,855 --> 01:33:35,144
It's the fear of, right.

1683
01:33:35,144 --> 01:33:37,005
I don't win when we get more productive.

1684
01:33:37,035 --> 01:33:37,245
That's right.

1685
01:33:37,245 --> 01:33:39,405
Because someone else,
something else takes my job.

1686
01:33:39,465 --> 01:33:40,065
That's what I said.

1687
01:33:40,065 --> 01:33:40,304
Yeah.

1688
01:33:40,309 --> 01:33:40,429
Yeah.

1689
01:33:40,455 --> 01:33:42,795
It's not, it's not the technology,
it's the neoliberalism.

1690
01:33:42,795 --> 01:33:43,275
That's right.

1691
01:33:43,275 --> 01:33:43,755
That kills you.

1692
01:33:44,264 --> 01:33:44,745
Exactly.

1693
01:33:44,745 --> 01:33:49,605
So I think we are an amazing, we are
a, a, a transitional period there is

1694
01:33:49,605 --> 01:33:53,505
incredible, once you simultaneously
be fearful and optimistic in the sense

1695
01:33:53,505 --> 01:33:55,184
that there's a lot to be worried about.

1696
01:33:55,335 --> 01:33:58,514
And the things outside the labor market
are arguably more worrisome than even

1697
01:33:58,514 --> 01:33:59,565
what's gonna happen in the labor market.

1698
01:33:59,894 --> 01:34:01,155
But there's also enormous opportunity.

1699
01:34:01,215 --> 01:34:04,184
And there, there's no
question that we will enhance the

1700
01:34:04,184 --> 01:34:05,325
rate of technological progress.

1701
01:34:05,325 --> 01:34:08,384
We'll be able to do better
things with addressing some of

1702
01:34:08,384 --> 01:34:09,554
our most challenging problems.

1703
01:34:09,554 --> 01:34:12,915
And if we do it well, we'll enable
more people to do valuable work.

1704
01:34:13,005 --> 01:34:17,325
But it's going to be, that doesn't
mean, even if that's good over a

1705
01:34:17,325 --> 01:34:19,695
while, it doesn't mean it's good for
everyone by any stretch of matter.

1706
01:34:19,700 --> 01:34:24,554
Are, are you suggesting that labor
markets did not inevitably, return

1707
01:34:24,554 --> 01:34:29,355
to a Pareto optimal equilibrium that
we can just trust the invisible hand?

1708
01:34:29,355 --> 01:34:30,615
I mean, you're an economist.

1709
01:34:30,615 --> 01:34:32,745
I, I don't know what you
teach in your classes, but

1710
01:34:33,105 --> 01:34:36,915
I think labor markets, well-functioning
labor markets are central to democracy.

1711
01:34:37,155 --> 01:34:40,575
You cannot have a well-functioning
democratic system where most people

1712
01:34:40,575 --> 01:34:43,815
don't perceive themselves to be
stakeholders and aren't perceived

1713
01:34:43,815 --> 01:34:45,644
to be productive contributors.

1714
01:34:45,825 --> 01:34:50,474
A society where everybody is seen as a
ward of the state is not, politically.

1715
01:34:50,955 --> 01:34:54,255
Stable, because the people
who are perceive themselves as the

1716
01:34:54,255 --> 01:34:56,625
income generators will say, why
are we supporting all these people?

1717
01:34:56,895 --> 01:34:59,325
So, the best you get, the
best case scenario is a country like

1718
01:34:59,325 --> 01:35:02,505
Kuwait where, you have
no taxation and no representation,

1719
01:35:02,505 --> 01:35:04,755
but at least the dictatorship is, relatively benevolent.

1720
01:35:04,905 --> 01:35:10,425
But, we have a real stake in
maintaining labor's value, making sure

1721
01:35:10,425 --> 01:35:15,645
that work pays well, and that people have
economic security that is tied to it.

1722
01:35:15,675 --> 01:35:19,155
Work is not just about the
displeasure going to a job, right?

1723
01:35:19,575 --> 01:35:22,575
It's about community, it's about
identity, it's about esteem.

1724
01:35:22,905 --> 01:35:23,445
For many people.

1725
01:35:23,445 --> 01:35:25,665
It's also a source of, of
meaning and fulfillment.

1726
01:35:26,265 --> 01:35:27,555
I think a by the way,

1727
01:35:27,615 --> 01:35:31,575
by the way, I think one of the things that
we can be most optimistic about is that a

1728
01:35:31,575 --> 01:35:37,815
greater and greater proportion of humans
on planet Earth every year have jobs from

1729
01:35:37,815 --> 01:35:39,315
which they can derive meaning, right?

1730
01:35:39,525 --> 01:35:43,395
This has been the best 30 years in
the history of humanity in terms

1731
01:35:43,395 --> 01:35:44,835
of elevating people out of poverty.

1732
01:35:45,195 --> 01:35:48,225
From the perspective of the
West, we say, yeah, it's been kind of

1733
01:35:48,225 --> 01:35:52,305
mid, we haven't grown that much,
but billions of people have been lifted

1734
01:35:52,305 --> 01:35:53,985
outta poverty actually by China's rise.

1735
01:35:54,345 --> 01:35:56,805
And the prosperity is created
in Sub-Saharan Africa, in,

1736
01:35:56,895 --> 01:35:58,245 central and South America.

1737
01:35:58,635 --> 01:36:00,945
So yes, it absolutely is the case.

1738
01:36:00,945 --> 01:36:03,675
And, these are,
China's, it's not a democracy.

1739
01:36:03,675 --> 01:36:06,135
It is a market economy,
and that has helped.

1740
01:36:06,645 --> 01:36:07,395
Yeah, for sure.

1741
01:36:07,695 --> 01:36:09,735
So, just a couple more questions.

1742
01:36:09,735 --> 01:36:10,065
David.

1743
01:36:10,065 --> 01:36:14,835
If you were in charge of the world, we
have a benevolent dictator question, you

1744
01:36:14,835 --> 01:36:20,925
know, political constraints aside, what
would you do with respect to AI to lead

1745
01:36:20,955 --> 01:36:22,875
to the promised land that you describe?

1746
01:36:23,295 --> 01:36:24,075
A couple things.

1747
01:36:24,075 --> 01:36:26,835
First of all, I, I would reg the, a
first thing I would do just 'cause

1748
01:36:26,835 --> 01:36:29,745
it's easy, is I would regulate
the use of intellectual property.

1749
01:36:29,745 --> 01:36:33,825
I actually think there's enormous IP
theft occurring in industrial scale.

1750
01:36:34,245 --> 01:36:38,655
Intellectual property law was not meant
to protect people against machines that

1751
01:36:38,655 --> 01:36:40,155
would read everything and regurgitate it.

1752
01:36:40,155 --> 01:36:40,875
So I think people should be.

1753
01:36:41,265 --> 01:36:42,825
Paid for their creations.

1754
01:36:43,275 --> 01:36:47,025
And, and that's a, that's a, a first
order problem, but more, kind

1755
01:36:47,025 --> 01:36:51,495
of at a a deeper level, I think we should
be investing in these good use cases.

1756
01:36:51,615 --> 01:36:55,455
We should say, how do we make
healthcare, a place where more

1757
01:36:55,455 --> 01:36:58,605
people can use expertise, where
it's cheaper and more available?

1758
01:36:58,635 --> 01:37:01,545
How do we make education more
effective and accessible?

1759
01:37:01,545 --> 01:37:05,115
And actually AI is a way, a tool that
we can use to make ai, sorry, to make

1760
01:37:05,115 --> 01:37:07,995
learning more like simulation, right?

1761
01:37:07,995 --> 01:37:10,185
More like a flight simul,
less like a chalkboard.

1762
01:37:11,055 --> 01:37:14,955
And so I would be investing for
good use cases and then I would

1763
01:37:15,015 --> 01:37:18,195
build lots of labor market supports
that are not really around ai.

1764
01:37:18,225 --> 01:37:19,605
They're just around insecurity.

1765
01:37:19,665 --> 01:37:22,065
We have, we have trade
production programs in the United States

1766
01:37:22,065 --> 01:37:24,825
where people whose jobs are displaced
by trade, but we don't have technology

1767
01:37:24,825 --> 01:37:26,235
displacement protection programs.

1768
01:37:26,235 --> 01:37:26,685
Why not?

1769
01:37:26,745 --> 01:37:29,355
I mean, why do I, if I
lose my job, I lose my job.

1770
01:37:29,355 --> 01:37:32,295
I don't care if it's because I lost it
because of a robot, or I lost it because

1771
01:37:32,295 --> 01:37:34,425
of a, a competitor in another country.

1772
01:37:34,485 --> 01:37:35,805
I just need economic security.

1773
01:37:36,285 --> 01:37:39,885
And there's where, we have
models of countries that do this

1774
01:37:39,885 --> 01:37:41,145
much more effectively than we do now.

1775
01:37:41,145 --> 01:37:41,625
It's hard.

1776
01:37:41,625 --> 01:37:43,965
We can't just become Denmark,
but we shouldn't think

1777
01:37:43,965 --> 01:37:45,105
that it's impossible to do.

1778
01:37:45,135 --> 01:37:46,815
'cause we know it is possible to do.

1779
01:37:47,355 --> 01:37:51,165
And certainly the mo the,
even during the Obama administration,

1780
01:37:51,165 --> 01:37:54,075
they stood up a, what was called
a, a, a wage insurance program

1781
01:37:54,075 --> 01:37:55,425
for people just placed from trade.

1782
01:37:55,785 --> 01:37:58,575
And essentially it said, if you take
a new job at a lower wage and you've

1783
01:37:58,575 --> 01:38:01,275
been displaced by trade, we'll make
up some of the difference for a while.

1784
01:38:01,875 --> 01:38:04,185
And that got people back
into the labor market faster.

1785
01:38:04,215 --> 01:38:07,515
It didn't help them recover earnings
per se, it didn't get them back to

1786
01:38:07,515 --> 01:38:11,325
where they were, but it made that
transition at least more successful.

1787
01:38:11,745 --> 01:38:14,355
And long-term unemployment is
devastating for people, not just

1788
01:38:14,355 --> 01:38:16,155
economically, but psychologically.

1789
01:38:28,365 --> 01:38:30,495
This is what I said in the last
episode, and let me repeat it.

1790
01:38:30,975 --> 01:38:32,175
The goal of socialism.

1791
01:38:33,030 --> 01:38:37,260
Is not to dismantle the
productive advances of capitalism.

1792
01:38:38,099 --> 01:38:41,730
It's to take those productive
advances and to harness them towards

1793
01:38:41,730 --> 01:38:43,769
the greater ends that we have now.

1794
01:38:43,769 --> 01:38:47,040
One of the great benefits that
capitalism is gonna bequeath to us

1795
01:38:47,160 --> 01:38:53,370
is this enormous gain in productivity
over the last 200 years or so, and that

1796
01:38:53,700 --> 01:38:59,370
puts us in a place where we can, in a
democratic setting, decide as a society

1797
01:38:59,849 --> 01:39:01,380
that we're going to simply work less.

1798
01:39:02,160 --> 01:39:06,179
Now, working less doesn't mean that
we freeze technological progress.

1799
01:39:06,269 --> 01:39:09,630
It just means that every
technological advance that comes

1800
01:39:09,720 --> 01:39:12,480
in society poses a question.

1801
01:39:12,480 --> 01:39:16,440
For us, technological advances,
productivity advances mean you can

1802
01:39:16,440 --> 01:39:19,920
produce more stuff in the same amount
of time or as much stuff in less time.

1803
01:39:20,519 --> 01:39:22,530
Now you can see what that trade off is.

1804
01:39:22,530 --> 01:39:26,670
We could decide that we'll take
every technological advance and

1805
01:39:26,700 --> 01:39:32,040
work 10% less than we used to, but
still enjoy gains in productivity

1806
01:39:32,040 --> 01:39:33,809
and in the expansion of the economy.

1807
01:39:33,809 --> 01:39:33,900
Right?

1808
01:39:33,900 --> 01:39:37,110
Whereas in capitalism, all of it is
put towards expansion of the economy

1809
01:39:37,170 --> 01:39:38,610
and not towards working less, right?

1810
01:39:38,610 --> 01:39:39,059
Right.

1811
01:39:39,480 --> 01:39:43,290
So we still want to be able to
increase the productive forces

1812
01:39:43,290 --> 01:39:44,700
to develop the productive forces.

1813
01:39:45,389 --> 01:39:50,400
What we want, however, is the freedom to
decide whether or not all those gains go

1814
01:39:50,460 --> 01:39:53,940
towards an expansion of the market and
an expansion of income or some of those

1815
01:39:53,940 --> 01:39:55,889
gains go towards more leisure time for us.

1816
01:39:55,980 --> 01:39:56,250
Yeah.

1817
01:39:56,760 --> 01:40:01,230
Well, that's what the point of socialism
is, and if that's the case, it means a

1818
01:40:01,230 --> 01:40:02,849
lot of work is still gonna be automated.

1819
01:40:03,330 --> 01:40:05,099
There's gonna be repetitive
components to it.

1820
01:40:05,849 --> 01:40:07,170
But we will decide.

1821
01:40:07,260 --> 01:40:09,120
Workers in each factory might decide.

1822
01:40:09,180 --> 01:40:10,800
It depends on your model of socialism.

1823
01:40:11,250 --> 01:40:12,510
We'll decide how to bring it about.

1824
01:40:12,870 --> 01:40:16,830
I can't see any version of socialism
in which you dismantle the workplace

1825
01:40:16,830 --> 01:40:18,690
and go back to artisanal farms.

1826
01:40:19,110 --> 01:40:24,570
Nor do I see any version of a sustainable
socialism that doesn't have consistent

1827
01:40:24,599 --> 01:40:26,760
and steady technological progress.

1828
01:40:27,240 --> 01:40:27,510
Right.

1829
01:40:27,510 --> 01:40:30,180
And we've been talking about ai.

1830
01:40:30,180 --> 01:40:33,510
That's what motivated this
episode, last episode, and this

1831
01:40:33,510 --> 01:40:35,370
question of what's going to happen.

1832
01:40:36,120 --> 01:40:39,750
And part of the reason we don't exactly
know what's going to happen is because

1833
01:40:39,750 --> 01:40:41,880
we don't have social planning, right?

1834
01:40:42,360 --> 01:40:45,900
All these decisions are put in
the hands of very few people who

1835
01:40:45,900 --> 01:40:51,150
have almost total discretion save,
basically the limit that the technology

1836
01:40:51,150 --> 01:40:55,260
itself imposes on them about how and
when to implement this technology.

1837
01:40:55,260 --> 01:41:00,269
But the other thing that a social
society would offer is, is also to

1838
01:41:00,269 --> 01:41:05,639
say that when these decisions are
made, we're going to account for that.

1839
01:41:05,670 --> 01:41:10,800
And the hardship that is put
upon workers as a result of

1840
01:41:10,800 --> 01:41:14,010
disruption, all that can be managed.

1841
01:41:14,190 --> 01:41:16,200
That is all already within.

1842
01:41:16,980 --> 01:41:23,250
Our ability as a society to do, we
just aren't doing it because that's

1843
01:41:23,250 --> 01:41:25,260
not what a capitalist society does.

1844
01:41:25,290 --> 01:41:25,500
Yeah.

1845
01:41:25,680 --> 01:41:27,180
You'll have technological change.

1846
01:41:27,600 --> 01:41:30,120
The effects of technological
change, whatever disruptions

1847
01:41:30,120 --> 01:41:31,170
it causes will be shared.

1848
01:41:31,350 --> 01:41:31,680
Right.

1849
01:41:31,770 --> 01:41:32,790
And they'll be managed better.

1850
01:41:33,150 --> 01:41:34,290
I'll just say one little thing.

1851
01:41:34,860 --> 01:41:39,090
I don't think planning makes the effects
of technological change predictable.

1852
01:41:39,125 --> 01:41:39,545
Mm-hmm.

1853
01:41:39,625 --> 01:41:41,820
That's the, I think technological
change will always have

1854
01:41:41,820 --> 01:41:43,200
unpredictable consequences.

1855
01:41:43,230 --> 01:41:43,500
Right.

1856
01:41:43,500 --> 01:41:47,640
I guess in the sense of unpredictable,
in the sense of like, well, how

1857
01:41:47,640 --> 01:41:51,900
devastating will it be to how many
people, the question of like

1858
01:41:51,900 --> 01:41:54,540
the, the toll it will take on people.

1859
01:41:54,540 --> 01:41:55,200
That is

1860
01:41:55,200 --> 01:41:55,380
Yeah.

1861
01:41:55,380 --> 01:41:56,640
That's still gonna be unpredictable.

1862
01:41:56,640 --> 01:41:58,050
That's why you need the institutions.

1863
01:41:58,055 --> 01:41:58,305
Mm-hmm.

1864
01:41:58,385 --> 01:42:02,310
You need institutions that help
people blunt and absorb the cost.

1865
01:42:02,700 --> 01:42:08,640
The thing about technological changes,
because all new technology links up to

1866
01:42:09,210 --> 01:42:14,610
other sectors in the economy, sectors
which don't necessarily use that

1867
01:42:14,610 --> 01:42:16,080
technology, but are affected by it.

1868
01:42:16,680 --> 01:42:21,630
It's so ramified and its fans out in
so many ways that it's not possible.

1869
01:42:21,630 --> 01:42:24,840
And it never has been possible
to predict down the line.

1870
01:42:24,960 --> 01:42:25,170
Mm.

1871
01:42:25,500 --> 01:42:30,390
All the chain effects of any significant
technological innovation that comes about.

1872
01:42:30,900 --> 01:42:34,020
Mm. What we wanna do is make sure that
there's deliberative mechanisms in place

1873
01:42:34,020 --> 01:42:38,430
where people who are being affected by it
can issue demands to the rest of society

1874
01:42:38,430 --> 01:42:40,470
to say, Hey man, this is tearing us apart.

1875
01:42:40,470 --> 01:42:41,460
We need to do something about it.

1876
01:42:41,970 --> 01:42:45,990
And that the society has institutions
in place that make it responsive

1877
01:42:46,260 --> 01:42:47,610
and sensitive to what's being said.

1878
01:42:47,760 --> 01:42:49,050
That's the idea of socialism.

1879
01:42:49,500 --> 01:42:53,310
The worst thing you can do is wave your
hands and say, well, all the negative

1880
01:42:53,310 --> 01:42:56,490
things that come about from technological
change, that's just because of capitalism.

1881
01:42:56,490 --> 01:42:56,730
Sure.

1882
01:42:56,730 --> 01:42:59,610
And if you just wave your hands and
say, well now we're in socialism,

1883
01:42:59,880 --> 01:43:00,750
we won't have to deal with it.

1884
01:43:01,140 --> 01:43:02,070
I think that would be a mistake.

1885
01:43:02,310 --> 01:43:05,850
What we have to do is to understand
that, A, we want technological change.

1886
01:43:06,360 --> 01:43:07,105
B, we want it to.

1887
01:43:07,680 --> 01:43:13,770
In a setting and in a form that
allows us to harness its gains and

1888
01:43:13,770 --> 01:43:15,750
to block all of its negative effects.

1889
01:43:15,840 --> 01:43:18,000
And that's the idea of socialism.

1890
01:43:18,990 --> 01:43:24,360
Why do you think labor and monopoly
capital, A book that was published in

1891
01:43:24,360 --> 01:43:29,580
the early seventies is still an important
text for leftists to understand?

1892
01:43:30,090 --> 01:43:34,560
The hallmark of any successful
scientific theory is if it not only

1893
01:43:34,890 --> 01:43:40,050
adequately explains past events, but
if it can lay a framework for, in

1894
01:43:40,050 --> 01:43:41,910
a sense, predicting future events.

1895
01:43:42,090 --> 01:43:45,360
Now, you don't narrowly predict
future events, but what you can do is

1896
01:43:45,360 --> 01:43:49,470
predict a certain set of principles,
a course of actions, a course of a

1897
01:43:49,470 --> 01:43:51,030
trajectory that events are gonna take.

1898
01:43:51,480 --> 01:43:56,340
Braverman said it is in the essence
of capitalism that management will

1899
01:43:56,340 --> 01:44:01,890
try to break through the skill barrier
that workers have, and it'll do it by

1900
01:44:02,070 --> 01:44:06,990
dismantling or taking apart those skills
and putting them into individual tasks

1901
01:44:06,990 --> 01:44:11,100
and then automating it to the extent that
they can to reduce the unpredictability to

1902
01:44:11,100 --> 01:44:12,960
reduce workers' agency and their autonomy.

1903
01:44:13,770 --> 01:44:15,150
That's what he said in 1975.

1904
01:44:16,140 --> 01:44:20,610
Now the asset test of that is a.
As new sectors develop, as new

1905
01:44:20,610 --> 01:44:24,960
occupations come in, do we find that
they are subject to the same sorts

1906
01:44:24,960 --> 01:44:26,670
of processes that you predicted?

1907
01:44:26,910 --> 01:44:27,780
How do you test it?

1908
01:44:28,259 --> 01:44:30,780
Well, what was the new sectors
that were coming about?

1909
01:44:30,780 --> 01:44:33,120
At that time, it was software engineering.

1910
01:44:33,690 --> 01:44:35,670
It was new technologies and computers.

1911
01:44:35,940 --> 01:44:42,870
Now it's things like ai and remarkably,
from all the literature that is

1912
01:44:42,870 --> 01:44:47,099
examining these things, you see
the same processes taking place in

1913
01:44:47,099 --> 01:44:51,090
sectors that virtually didn't exist
when Braverman wrote that book.

1914
01:44:51,090 --> 01:44:51,360
Right?

1915
01:44:51,360 --> 01:44:53,250
This book does not talk about computers

1916
01:44:54,240 --> 01:44:55,799
because they were just
coming onto the scene.

1917
01:44:55,889 --> 01:45:01,200
So for example, numerical
control as a form of automating the auto

1918
01:45:01,200 --> 01:45:05,639
industry and manufacturing was just coming
about when Braverman was writing the book.

1919
01:45:05,639 --> 01:45:08,880
But he immediately, even then, was
able to spot that this would be the

1920
01:45:08,880 --> 01:45:13,349
future of control in manufacturing,
and this turned out to be true.

1921
01:45:14,040 --> 01:45:18,809
The hallmark of a great theory is when
it can give you a framework to understand

1922
01:45:18,809 --> 01:45:22,679
events that are occurring after the theory
was promulgated, and that's braverman.

1923
01:45:23,429 --> 01:45:25,049
Why is it successful in doing that?

1924
01:45:25,110 --> 01:45:30,480
Because I think it in fact, got to
the essence of what capital and the

1925
01:45:30,509 --> 01:45:34,620
employment relation is, which is
the employment relation rests on a

1926
01:45:34,620 --> 01:45:40,320
fundamental inescapable conflict of
interests between employers and employees.

1927
01:45:40,950 --> 01:45:47,009
Once you understand that that relation
is as intrinsically conflictual, it's

1928
01:45:47,040 --> 01:45:51,420
easy then to predict that What employers
are gonna try to do is to increase

1929
01:45:51,420 --> 01:45:53,340
their leverage over their employees.

1930
01:45:53,670 --> 01:45:58,320
And anything that employees have that
gives them leverage against their

1931
01:45:58,559 --> 01:46:00,480
employers is gonna be seen as a threat.

1932
01:46:01,019 --> 01:46:05,820
Their skills, the monopoly over skills
and the monopoly over the pace and

1933
01:46:05,820 --> 01:46:10,559
the quality of work is the number one
leverage that workers have against

1934
01:46:10,559 --> 01:46:12,360
their employers as individuals.

1935
01:46:13,500 --> 01:46:15,389
That's what Braverman saw and.

1936
01:46:15,795 --> 01:46:20,655
The academic reception of labor and
monopoly capital was ecstatic at first.

1937
01:46:21,015 --> 01:46:25,875
But of course, as radicals became
absorbed into academia, as they became

1938
01:46:25,875 --> 01:46:32,235
deradicalize, you see them finding ways
to nitpick or to pick away at the book.

1939
01:46:32,835 --> 01:46:36,825
But this is a giant, and my view
is, as long as capitalism is around,

1940
01:46:36,825 --> 01:46:38,025
this book is gonna be relevant.

1941
01:46:38,025 --> 01:46:40,695
And there's only a handful of
works, you can say that about.

1942
01:46:41,025 --> 01:46:43,815
I don't know how many of our listeners
are familiar with this book, but I would

1943
01:46:43,815 --> 01:46:48,975
say there's a list of three to five books
in your life that you want to read as an

1944
01:46:48,975 --> 01:46:54,465
organizer, as maybe a worker that help
you understand the situation you're in.

1945
01:46:54,795 --> 01:46:55,635
This is what I'm offering

1946
01:47:02,205 --> 01:47:04,515
next up, section B warfare.

1947
01:47:07,845 --> 01:47:10,605
So I'm gonna start with
some very basic questions.

1948
01:47:11,085 --> 01:47:13,515
And the first question is,
what are we talking about?

1949
01:47:13,935 --> 01:47:15,255
What is ai?

1950
01:47:15,795 --> 01:47:18,525
Well, that boils down
to what is intelligence.

1951
01:47:18,855 --> 01:47:21,825
And our intelligence has two components.

1952
01:47:22,035 --> 01:47:26,175
One is understanding the world, and
that's what science does, by the way.

1953
01:47:26,625 --> 01:47:30,975
And the other is being able to act with
that knowledge plan and achieve goals.

1954
01:47:31,125 --> 01:47:34,125
And we are building machines
that have these two aspects.

1955
01:47:34,545 --> 01:47:38,595
But in, in the, the last year,
we've been focusing more and more

1956
01:47:38,595 --> 01:47:42,525
on achieving goals, also known as
agency and build these agent systems.

1957
01:47:44,445 --> 01:47:49,845
Maybe just to frame not just what is
intelligence, but why is it so valuable?

1958
01:47:49,845 --> 01:47:55,515
Why did DE has the founder of
Google DeepMind say, first solve

1959
01:47:55,545 --> 01:48:01,215
intelligence, then use intelligence
to solve everything else.

1960
01:48:02,235 --> 01:48:04,815
Because if you think what makes
intelligence different from other

1961
01:48:04,815 --> 01:48:09,555
kinds of technology, think about all
science, all technology, all military

1962
01:48:09,555 --> 01:48:11,265
invention, what was behind all of that?

1963
01:48:12,105 --> 01:48:13,155
It was intelligence.

1964
01:48:13,575 --> 01:48:18,615
So put simply, if I made an advance
in rocketry, that form of science

1965
01:48:19,065 --> 01:48:22,275
that didn't advance medicine, when
I make an advance in medicine,

1966
01:48:22,275 --> 01:48:23,565
that doesn't advance rocketry.

1967
01:48:24,045 --> 01:48:28,934
But if I make an advance in artificial
general intelligence, intelligence

1968
01:48:28,934 --> 01:48:32,625
is what gave us all science, all
technology, all military advancement.

1969
01:48:33,224 --> 01:48:36,615
And that's why it's not just
that whoever solves intelligence

1970
01:48:36,615 --> 01:48:37,665
can solve everything else.

1971
01:48:37,695 --> 01:48:38,445
That's their belief.

1972
01:48:38,804 --> 01:48:43,335
It's whoever can dominate intelligence
will be able to dominate everything else.

1973
01:48:43,335 --> 01:48:44,325
And that's what Putin said.

1974
01:48:44,325 --> 01:48:47,655
That's why he said, I think it's like
whoever owns AI will own the world.

1975
01:48:47,655 --> 01:48:50,355
And I wanted to set that up because
I think a lot of what we're gonna be

1976
01:48:50,355 --> 01:48:56,175
talking about today is how the race for
this prize, this sort of ring and Lord of

1977
01:48:56,175 --> 01:48:57,945
the Rings, this ring of ultimate power.

1978
01:48:57,945 --> 01:48:59,085
At least that's how it's seen.

1979
01:48:59,085 --> 01:49:03,434
If I get to that prize, it confers
power across all other domains.

1980
01:49:03,795 --> 01:49:06,405
And that's why we're in for
the seatbelt ride that, that

1981
01:49:06,405 --> 01:49:07,125
you mentioned at the beginning.

1982
01:49:07,245 --> 01:49:12,825
And, and I just wanna add that this
goes against, a lot of the political

1983
01:49:12,825 --> 01:49:18,585
principles that, the world
has chosen, at least, in the

1984
01:49:18,585 --> 01:49:24,255
west, of democracy where power is
shared, where power is distributed.

1985
01:49:24,525 --> 01:49:26,925
It's not, in a
single corporation, a single

1986
01:49:26,925 --> 01:49:28,394
person or a single government.

1987
01:49:28,905 --> 01:49:34,724
By having a lot of power in a few
hands, we can end up in a world where

1988
01:49:35,415 --> 01:49:37,485
democratic values, disappear.

1989
01:49:38,835 --> 01:49:39,105
Okay.

1990
01:49:39,105 --> 01:49:41,595
We, we've, we've, we've raced
away from the idea of what

1991
01:49:41,595 --> 01:49:43,815
is AI to some, some harms.

1992
01:49:44,115 --> 01:49:47,955
But before we talk about that, those
implications of power, I want to

1993
01:49:47,955 --> 01:49:49,485
actually focus first on the harm.

1994
01:49:49,485 --> 01:49:52,184
So you've expressed what AI is.

1995
01:49:52,455 --> 01:49:56,535
Basically it's taking data, making
an inference, and learning something

1996
01:49:56,535 --> 01:50:01,215
we otherwise couldn't know at a scale
that is far exceeds human cognition.

1997
01:50:01,245 --> 01:50:04,995
And so therefore it's gonna be exceeding
how we can understand the world.

1998
01:50:06,465 --> 01:50:11,085
So that says, actually when I
describe it that way, fan, fantastic.

1999
01:50:11,145 --> 01:50:12,645
It's phenomenal, right?

2000
01:50:12,645 --> 01:50:13,185
Rocketry.

2001
01:50:13,185 --> 01:50:13,755
Sure.

2002
01:50:13,905 --> 01:50:14,175
Right?

2003
01:50:14,175 --> 01:50:14,625
Armaments.

2004
01:50:14,625 --> 01:50:15,135
Okay.

2005
01:50:15,375 --> 01:50:16,755
But saving people's lives.

2006
01:50:16,755 --> 01:50:17,205
Love it.

2007
01:50:17,475 --> 01:50:18,195
What's wrong?

2008
01:50:18,225 --> 01:50:20,325
What's the problem that
we're talking about?

2009
01:50:20,775 --> 01:50:24,075
Well, it would be great if
two conditions are obtained.

2010
01:50:24,135 --> 01:50:27,645
One is that the AI actually
does the things that we ask,

2011
01:50:28,485 --> 01:50:29,595
and right now we don't have it.

2012
01:50:29,595 --> 01:50:32,025
That's the alignment problem that
is in the title of this session.

2013
01:50:32,655 --> 01:50:37,125
The second problem, of course, is
that even if AI was aligned, who decides

2014
01:50:37,125 --> 01:50:40,365
what are the goals that the AI is going
to follow as we discussed previously?

2015
01:50:40,875 --> 01:50:43,575
And we don't have solutions
to both of these, and we

2016
01:50:43,575 --> 01:50:46,545
already seen the consequences
of not having those solutions.

2017
01:50:48,585 --> 01:50:54,195
So, AI is confusing to weave this
together about the alignment and the,

2018
01:50:54,285 --> 01:50:57,195 the amazing things that it can bring.

2019
01:50:57,195 --> 01:51:00,525
It's confusing because it will, it
will give us new cures to cancer.

2020
01:51:01,335 --> 01:51:05,055
But the same AI that knows biology
well enough, knows immuno-oncology

2021
01:51:05,055 --> 01:51:09,585
well enough to develop those cures
for cancer, that AI can't be separated

2022
01:51:09,585 --> 01:51:14,055
from the AI that also knows how to
build new kinds of biological weapons.

2023
01:51:14,655 --> 01:51:16,844
You can't separate the
promise from the peril,

2024
01:51:17,205 --> 01:51:18,405
but aren't we in control?

2025
01:51:18,885 --> 01:51:22,785
So this is a, a, a common myth
that technology is just a tool.

2026
01:51:22,965 --> 01:51:26,715
All tools can be used for good or
evil, and humans ultimately

2027
01:51:26,715 --> 01:51:27,885
decide how we want this to go.

2028
01:51:27,885 --> 01:51:31,365
But what's different about AI is
Joshuas sort of speaking to, and,

2029
01:51:31,365 --> 01:51:34,755
and you've all Hara will often say
is it's the first technology that's

2030
01:51:34,755 --> 01:51:36,495
about making its own decisions.

2031
01:51:36,885 --> 01:51:39,675
If you use G PT 5.2, you
ask it a complex question.

2032
01:51:39,945 --> 01:51:42,045
It reasons a level of abstract.

2033
01:51:42,045 --> 01:51:44,085
It's reasoning a million times a second.

2034
01:51:44,625 --> 01:51:46,965
And it's coming up with its
own conclusions that we don't

2035
01:51:46,965 --> 01:51:48,705
know how to control, what
those conclusions will lead to.

2036
01:51:49,455 --> 01:51:54,285
And, and when n AI has an interaction
for days, weeks, or months with a person,

2037
01:51:54,525 --> 01:51:59,894
maybe even a child, there, there's
no adult in the room checking that that

2038
01:51:59,894 --> 01:52:01,514
interaction is going well with the child.

2039
01:52:01,514 --> 01:52:01,575
L

2040
01:52:02,264 --> 01:52:05,625
Let, lemme drill down on this a little
bit quickly, a little bit more before

2041
01:52:05,625 --> 01:52:07,425
we go to the child aspect to it.

2042
01:52:08,175 --> 01:52:12,465
There's a link that you're making that
I think needs to not be asserted but

2043
01:52:12,465 --> 01:52:18,525
explained, and that is you have this
technology that is smarter than we are,

2044
01:52:18,525 --> 01:52:23,715
that can do more than we can do, yet it's
gonna somehow be fundamentally also so

2045
01:52:23,715 --> 01:52:28,094
dumb that it's gonna want to kill us or
it's gonna be benevolent or Well, yeah.

2046
01:52:28,094 --> 01:52:30,675
I mean, what, tell me the link there of

2047
01:52:30,675 --> 01:52:30,974
Yeah,

2048
01:52:31,304 --> 01:52:31,575
yeah, yeah.

2049
01:52:32,985 --> 01:52:33,495
Destruction.

2050
01:52:33,585 --> 01:52:34,455
Yes.

2051
01:52:34,485 --> 01:52:40,815
So it's a problem that's been well
studied even at the mathematical level.

2052
01:52:41,295 --> 01:52:49,304
The issue is that when we define, what the AI should be optimizing what

2053
01:52:49,304 --> 01:52:52,755
it should be doing, the goals that it
should reach, we can't do it perfectly.

2054
01:52:53,565 --> 01:52:57,465
And so there's gonna be a slight
mismatch between what the AI understands

2055
01:52:57,465 --> 01:52:59,594
we want and what we actually want.

2056
01:52:59,625 --> 01:53:03,105
And that mismatch is gonna create
a lot of problems to, to make it

2057
01:53:03,105 --> 01:53:08,415
more concrete, think about what
laws, legislation is supposed to do.

2058
01:53:08,955 --> 01:53:13,005
It's, it's setting up the red lines for
people and corporations to not cross.

2059
01:53:13,800 --> 01:53:18,090
But it's a, it's a never
ending job because the corporations

2060
01:53:18,090 --> 01:53:19,860
are, the people find loopholes, right?

2061
01:53:20,220 --> 01:53:25,200
It's impossible to write a law that
will perfectly embody,

2062
01:53:25,200 --> 01:53:27,210
what morally we are trying to achieve.

2063
01:53:27,389 --> 01:53:28,590
And it's the same thing with ai.

2064
01:53:28,590 --> 01:53:32,460
We don't know how to tell it, what we actually want.

2065
01:53:32,490 --> 01:53:35,340
And, and there are other reasons that
have to do with how current AI is

2066
01:53:35,340 --> 01:53:41,370
trained, is trained to imitate people,
and people have lots of, built-in

2067
01:53:41,370 --> 01:53:43,620
drives, including they don't wanna die.

2068
01:53:44,190 --> 01:53:47,670
And we are seeing AI already,
reflecting those drives, which

2069
01:53:47,670 --> 01:53:49,980
means they're trying to resist
when we want to shut them down.

2070
01:53:50,309 --> 01:53:54,300
Now, we shared a, a session together
at the AI Safety Summit about a

2071
01:53:54,300 --> 01:53:58,380
year ago in Paris, and you presented
some riveting examples of that.

2072
01:53:58,620 --> 01:54:02,010
In terms of studies of the AI
using deception to protect itself.

2073
01:54:02,519 --> 01:54:05,070
Give, give one or two examples
that will frighten the audience.

2074
01:54:05,220 --> 01:54:05,490
Yeah.

2075
01:54:05,519 --> 01:54:09,690
So more recent than than last year,
last summer, Anthropic showed,

2076
01:54:09,690 --> 01:54:15,179
reports of experiments in which the AI
is, working for a company.

2077
01:54:15,179 --> 01:54:16,620
It has a mission with the company.

2078
01:54:16,620 --> 01:54:20,550
It has access to all the emails,
and then the engineers plant fake

2079
01:54:20,550 --> 01:54:22,260
emails into that set of emails.

2080
01:54:22,440 --> 01:54:26,099
One of the emails said that the, AI
is gonna be replaced by a new version by

2081
01:54:26,099 --> 01:54:28,139
that date, by this particular engineer.

2082
01:54:28,830 --> 01:54:32,370
Another email reveals that the,
the engineer has an affair.

2083
01:54:34,080 --> 01:54:34,710
What happens?

2084
01:54:35,670 --> 01:54:40,170
The AI strategizes, because it doesn't
wanna be shut down and replaced by

2085
01:54:40,170 --> 01:54:44,309
a new version, and it sends an email
to the engineer, blackmailing him.

2086
01:54:46,349 --> 01:54:47,130
Threatening to blackmail

2087
01:54:47,519 --> 01:54:51,780
because, because the ai, is
saying if you, if you do that change,

2088
01:54:51,840 --> 01:54:55,290 I, I will, automatically
there will be a message sent to the press.

2089
01:54:55,290 --> 01:54:55,590
Let

2090
01:54:55,590 --> 01:54:55,740
me

2091
01:54:55,740 --> 01:54:57,059
add one, just elaboration on this.

2092
01:54:57,059 --> 01:54:57,210
Yeah, sure.

2093
01:54:57,269 --> 01:55:00,809
Because you might think, okay, I just
heard Joshua say that there's a bug

2094
01:55:00,809 --> 01:55:02,790
in the ai, all software has bugs.

2095
01:55:02,790 --> 01:55:05,670
Let's just patch that bug and
then the rest of AI will be great.

2096
01:55:06,300 --> 01:55:10,140
So when Anthropic did this
study about blackmail, that they

2097
01:55:10,140 --> 01:55:12,540
were testing their model called
Claude, you all can use Claude.

2098
01:55:13,110 --> 01:55:17,250
But then another, I think Anthropic then
later tested all the other models, chat,

2099
01:55:17,250 --> 01:55:22,860
GPT, Gemini, Google, Gemini,
and even deep seek the Chinese model.

2100
01:55:23,430 --> 01:55:28,380
And all of them, all of them exhibit
the blackmail behavior between

2101
01:55:28,380 --> 01:55:32,130
79 and I think 96% of the time.

2102
01:55:32,490 --> 01:55:33,750
And it's not just blackmail.

2103
01:55:33,840 --> 01:55:37,530
There's been now a series of reports
from the labs, from independent parties

2104
01:55:38,010 --> 01:55:40,860
showing many, deceptive behavior.

2105
01:55:41,309 --> 01:55:46,200
In other words, the AI has goals
that we would not agree with, and then

2106
01:55:46,200 --> 01:55:48,180
it acts according to those bad goals.

2107
01:55:48,809 --> 01:55:49,140
Okay.

2108
01:55:49,950 --> 01:55:53,580
Let me ask a question that sounds
like a sociological question, but

2109
01:55:53,580 --> 01:55:55,140
it's actually a technical question.

2110
01:55:55,140 --> 01:55:56,730
So give a technical answer.

2111
01:55:56,760 --> 01:55:58,320
Feel free to, to within reason.

2112
01:55:59,310 --> 01:56:02,400
Where does, and I'll, but
let me lay out the case.

2113
01:56:03,000 --> 01:56:06,420
Where does the AI learn
the deception from?

2114
01:56:06,750 --> 01:56:11,520
Of course, it has its training data, and
just as it can understand, appreciate what

2115
01:56:11,520 --> 01:56:13,710
Shakespeare means by when he says Rose.

2116
01:56:13,890 --> 01:56:16,860
Not because A, a Shakespearean
scholar can understand the 30

2117
01:56:16,860 --> 01:56:20,880
references he remembers, but there's
the 300 references that the AI has.

2118
01:56:20,880 --> 01:56:25,440
So there's a encoding, somehow
an intricate network of the

2119
01:56:25,440 --> 01:56:27,810
of, of Rose and Shakespeare.

2120
01:56:28,020 --> 01:56:30,930
And it can appreciate all the
ways in which adjectives and verbs

2121
01:56:30,930 --> 01:56:34,320
are used with Rose to understand
rosiness and Shakespeare.

2122
01:56:34,830 --> 01:56:36,930
Where is the ai?

2123
01:56:37,020 --> 01:56:40,320
These are, it's learning from human
data and humans are deceptive.

2124
01:56:40,500 --> 01:56:43,800
So it's inherently learning
deception from the training data.

2125
01:56:44,760 --> 01:56:50,280
Yet there we could change the data
that we have and get rid of four

2126
01:56:50,280 --> 01:56:53,070
chan and only have no liturgy.

2127
01:56:53,220 --> 01:56:55,470
No there, there's deception everywhere.

2128
01:56:56,040 --> 01:57:00,510
Not just, not just in, you
know, a few, online places.

2129
01:57:00,960 --> 01:57:02,310
It's part of our culture.

2130
01:57:02,310 --> 01:57:03,480
It's part of being human.

2131
01:57:03,570 --> 01:57:05,460
And by the way, it's not just deception.

2132
01:57:05,520 --> 01:57:10,140
It's the thing that I'm most concerned
about is, the self preservation drive.

2133
01:57:10,170 --> 01:57:13,620
Like every human has a self preservation
drive, but do we wanna build tools

2134
01:57:14,610 --> 01:57:16,800
that don't want to be shut down?

2135
01:57:16,800 --> 01:57:17,760
I don't think that's good.

2136
01:57:18,030 --> 01:57:22,620
And, and also, it's not just this
like sounding a bit science fiction.

2137
01:57:22,860 --> 01:57:24,630
It's something that is happening already.

2138
01:57:24,990 --> 01:57:28,980
So this misalignment is
showing up in what's called sife.

2139
01:57:29,010 --> 01:57:32,760
So ev anybody who's played with those
systems should know that they're trying

2140
01:57:32,760 --> 01:57:36,030
to please you, which means they're,
they're lying to make you feel good.

2141
01:57:41,025 --> 01:57:41,145
I.

2142
01:57:48,269 --> 01:57:51,059
So I wanna ask about
fully autonomous weapons.

2143
01:57:51,059 --> 01:57:55,290
This is something that we have seen
developing over the last decade.

2144
01:57:55,290 --> 01:57:57,750
It was one of the reasons that the
Pentagon and Anthropic reportedly

2145
01:57:57,750 --> 01:57:59,040
couldn't get this deal done.

2146
01:57:59,339 --> 01:58:05,429
What are fully autonomous weapons and
do we know, are they being deployed

2147
01:58:05,429 --> 01:58:06,929
right now by the United States?

2148
01:58:07,379 --> 01:58:09,059
I, I, I don't think it's entirely clear.

2149
01:58:09,085 --> 01:58:15,059
I, I would emphasize that as far as
we know, and with, with, with pretty

2150
01:58:15,059 --> 01:58:20,429
good evidence, the, the main use case
of AI in national security context

2151
01:58:20,429 --> 01:58:22,740
is not yet, autonomous weapons.

2152
01:58:22,740 --> 01:58:25,500
Certainly that's possible from what
we see in, in going on robotics

2153
01:58:25,500 --> 01:58:26,879
and, and demonstrations that we see.

2154
01:58:27,269 --> 01:58:31,259
The main use case really is
in synthesizing analysis from

2155
01:58:31,950 --> 01:58:34,440
massive diverse, data sources.

2156
01:58:34,440 --> 01:58:36,570
So, satellite imagery
and battlefield reports and

2157
01:58:36,570 --> 01:58:37,860
collected intelligence and so on.

2158
01:58:37,860 --> 01:58:41,700
Being able to make recommendations to
human commanders in real time is not

2159
01:58:41,700 --> 01:58:43,110
something that was ever possible before.

2160
01:58:43,170 --> 01:58:43,950
And now it is.

2161
01:58:45,179 --> 01:58:48,570
However, all of this
is, is essentially unproven, right?

2162
01:58:48,780 --> 01:58:53,280
We, we do see examples where these, where
these models can make wild, crazy mistakes

2163
01:58:53,280 --> 01:58:54,900
that, that no human would ever make.

2164
01:58:55,290 --> 01:58:58,110
And so there's a tendency that even if
there is kind of a human in the loop

2165
01:58:58,110 --> 01:59:03,599
to, review the recommendation and
make a decision, there's, there's,

2166
01:59:03,599 --> 01:59:05,849
we see in business, there's
gonna be a tendency to kind of.

2167
01:59:06,255 --> 01:59:09,525
Go into rubber stamping,
automated approval here.

2168
01:59:09,525 --> 01:59:10,455
So, there's a question.

2169
01:59:10,455 --> 01:59:13,785
Is there actually human deliberation
going on about these recommendations?

2170
01:59:14,295 --> 01:59:19,005
Or is it slowly being offloaded,
onto, onto the models themselves?

2171
01:59:19,005 --> 01:59:22,815
So in that sense, the,
the word autonomous is, is,

2172
01:59:22,875 --> 01:59:23,895
is very real there, right?

2173
01:59:23,895 --> 01:59:28,725
It's, it's the, there's target
prioritization and deployment decisions

2174
01:59:28,725 --> 01:59:31,605
and so forth that are coming in
and perhaps have some chance for

2175
01:59:31,605 --> 01:59:33,015
that, for freaking to review them.

2176
01:59:33,015 --> 01:59:35,205
But whether that's actually
happening or not is, is not clear.

2177
01:59:35,565 --> 01:59:39,735
I think what captures the attention
in the public is, is more about, you

2178
01:59:39,735 --> 01:59:42,705
know, the, the killer robot that, that
is able to sort of operate on its own.

2179
01:59:42,705 --> 01:59:46,995
And that kind of technology certainly
is, is, is possible right now.

2180
01:59:47,205 --> 01:59:51,735
It's not clear that that's being
actively deployed in, as of today.

2181
01:59:52,005 --> 01:59:56,085
When you talk about the rubber stamping,
obviously the stakes are very different,

2182
01:59:56,565 --> 02:00:04,155
but I think about the ways that learning
and writing and, kind of

2183
02:00:04,185 --> 02:00:09,255
crunching data is being offloaded
to LLMs and different generative ai.

2184
02:00:09,855 --> 02:00:13,995
And the question is, how much
are humans actually deliberating

2185
02:00:13,995 --> 02:00:17,835
learning, making these decisions
versus just yeah, offloading it.

2186
02:00:18,315 --> 02:00:22,365
But when you're talking about
targeting systems or other types

2187
02:00:22,365 --> 02:00:27,555
of extremely high stakes military
operations, obviously the implications

2188
02:00:27,555 --> 02:00:28,935
of that are much more concerning.

2189
02:00:29,685 --> 02:00:29,895
Right?

2190
02:00:29,900 --> 02:00:32,505
So, so at, these models
will produce some kind of an

2191
02:00:32,505 --> 02:00:37,065
answer, and this is a well-known re
essentially a failure mode of, of, of

2192
02:00:37,065 --> 02:00:38,835
AI systems is that they do tend to.

2193
02:00:39,554 --> 02:00:42,405
Really want to provide some
kind of an answer to your, to

2194
02:00:42,405 --> 02:00:43,424
your question, no matter what.

2195
02:00:43,455 --> 02:00:48,285
They don't tend to say, I don't know,
or the, the evidence is not clear, or We

2196
02:00:48,285 --> 02:00:49,605
should really stop and slow down here.

2197
02:00:49,605 --> 02:00:50,894
That's not something
they've been trained to do.

2198
02:00:50,894 --> 02:00:53,835
That's not something in the data, that's
not something that, that we've even

2199
02:00:53,835 --> 02:00:55,605
wanted them to do in, in, in many cases.

2200
02:00:55,785 --> 02:00:58,755
And so that these answers are coming
out with this kind of authoritative

2201
02:00:58,755 --> 02:01:03,674
context of the computer said so, or, occasional flashes of

2202
02:01:03,674 --> 02:01:06,285
brilliance where it's like, oh, wow,
they really connected things across

2203
02:01:06,315 --> 02:01:09,045
large data sets that would never
have been possible for a human to do.

2204
02:01:09,344 --> 02:01:12,014
You, you, you're faced with
this decision at some point where it's,

2205
02:01:12,075 --> 02:01:16,094
it's you, this, this thing in front
of me may actually have a, a better

2206
02:01:16,125 --> 02:01:17,775
decision making ability than I do.

2207
02:01:18,075 --> 02:01:20,505
And so once you find yourself in
that situation, what do you do?

2208
02:01:20,565 --> 02:01:25,995
Is it, is it more correct or ethical to
stop and review it and go against it?

2209
02:01:25,995 --> 02:01:28,245
Or is it more ethical to say, Hey,
even though I don't really understand

2210
02:01:28,245 --> 02:01:31,960
exactly how or why this, this decision,
this recommendation came through, I,

2211
02:01:32,054 --> 02:01:36,674
I, I, I'm, authorized and,
and, and incentivized to trust it.

2212
02:01:36,855 --> 02:01:39,315
And so I, I personally worry about
those effects more than I worry

2213
02:01:39,315 --> 02:01:42,705
about the actual press the button
go off and do whatever you want.

2214
02:01:44,115 --> 02:01:46,665
Go fight the war without any
kind of, oversight at all.

2215
02:01:46,724 --> 02:01:46,965
Yeah.

2216
02:01:46,995 --> 02:01:50,894
That's never gonna be the, well, maybe not
never, but that's, that's not likely to

2217
02:01:51,255 --> 02:01:53,835
soon be the, advertised capability.

2218
02:01:54,375 --> 02:01:54,674
Right?

2219
02:01:54,674 --> 02:01:56,745
It's gonna be something
much, much slower and softer.

2220
02:01:57,075 --> 02:01:59,745
But operationally, I don't know if
there's a, if there's a big difference

2221
02:01:59,745 --> 02:02:03,674
between saying go off and do as thou
wilt versus come in and check with

2222
02:02:03,674 --> 02:02:04,965
me, but I'm not actually checking.

2223
02:02:05,655 --> 02:02:05,924
Yeah.

2224
02:02:06,014 --> 02:02:06,285
Yeah.

2225
02:02:06,735 --> 02:02:15,105
Did you follow the, meta AI safety
official saga from, or, or employee, I

2226
02:02:15,105 --> 02:02:19,634
don't know what her actual title was,
saga from last week, where she had.

2227
02:02:20,205 --> 02:02:24,555
Programmed AI to delete emails,
but check with her first.

2228
02:02:24,705 --> 02:02:29,955
And then the AI agent basically just
went and wiped her entire inbox.

2229
02:02:30,105 --> 02:02:32,865
And it, if you didn't follow
this, I don't, this is a thread

2230
02:02:32,865 --> 02:02:34,215
we don't have to go down, but

2231
02:02:34,310 --> 02:02:35,745
I'm, I'm aware of it, but not much more.

2232
02:02:35,745 --> 02:02:39,195
But honestly, I, the, these
reports as they come out is, is.

2233
02:02:40,665 --> 02:02:43,545
They, they, they happen a
lot more often than we see reported,

2234
02:02:43,545 --> 02:02:46,785
and so only the ones that kind of can
capture some imagination as that one did.

2235
02:02:46,785 --> 02:02:50,445
A, a year or two ago, there was
the agent that, that deleted the entire

2236
02:02:50,445 --> 02:02:54,855
database and then sort of was apologizing
eff effusively about, oh, I'm so sorry.

2237
02:02:54,855 --> 02:02:56,505
You're right, I ignored
your orders and so forth.

2238
02:02:56,505 --> 02:03:00,015
But this, this, this, this was
less impactful because it cost a customer

2239
02:03:00,015 --> 02:03:02,295
some money or something like that, as
opposed to this, this may be more dramatic

2240
02:03:02,295 --> 02:03:04,035
use cases, but, but this is, this is, but

2241
02:03:04,035 --> 02:03:08,685
the fact that it would be like even
people who are supposed to be steeped in

2242
02:03:08,685 --> 02:03:16,245
this and helping to shape these agents,
these, these, these AI tools, and it

2243
02:03:16,245 --> 02:03:21,405
gets out of their control and in really
problematic ways and terrifying ways,

2244
02:03:21,405 --> 02:03:25,125
frankly, if you, then apply
that to what happens when we are using

2245
02:03:25,125 --> 02:03:27,165
them more, more fully in the military.

2246
02:03:27,210 --> 02:03:30,225
A a absolutely, and, and this is what
I, I think about all the time, is

2247
02:03:30,225 --> 02:03:34,575
that, it's, it's easy to
look at the superhuman capabilities in

2248
02:03:34,605 --> 02:03:37,575
various ways and say, wow,
this thing is, is is really doing

2249
02:03:37,575 --> 02:03:39,285
better than I could do with my team.

2250
02:03:39,285 --> 02:03:41,355
Or, or, or, or myself.

2251
02:03:41,745 --> 02:03:44,175
But it's not about the successes
that I think are interesting.

2252
02:03:44,175 --> 02:03:48,945
It's about when it fails, it fails in
ways that the rest of our sociotechnical

2253
02:03:48,945 --> 02:03:50,865
systems are not equipped to gauge.

2254
02:03:50,865 --> 02:03:54,195
I mean, even something like,
hallucinating citations in a legal brief,

2255
02:03:54,615 --> 02:03:56,175
No lawyer would have the.

2256
02:03:56,565 --> 02:04:00,585
Audacity to attempt to make
up a case and put it in there.

2257
02:04:00,585 --> 02:04:02,054
And so there's no check there.

2258
02:04:02,054 --> 02:04:04,335
No, people aren't used to sort of
checking every citation to make sure

2259
02:04:04,335 --> 02:04:05,684
that they're, they're valid and true.

2260
02:04:05,684 --> 02:04:07,875
We just sort of assume that,
that the consequences would be a

2261
02:04:07,875 --> 02:04:09,525
deterrent to ever trying to do that.

2262
02:04:10,065 --> 02:04:12,045
These systems don't
have those consequences, don't,

2263
02:04:12,045 --> 02:04:13,275
don't understand those deterrents.

2264
02:04:13,724 --> 02:04:17,235
They, they like the idea of that if
that case were to would exist, boy,

2265
02:04:17,235 --> 02:04:18,855
that would really help my argument.

2266
02:04:19,065 --> 02:04:21,554
And so they'll happily sort of put
that in there, because they're,

2267
02:04:21,764 --> 02:04:24,764
Essentially just predicting
tokens that look good to humans.

2268
02:04:25,304 --> 02:04:29,714
And so once you start turning these things
over into, being able to take action

2269
02:04:29,714 --> 02:04:33,195
in the world, such as delete emails or
delete a database, yeah, they're, they're

2270
02:04:33,195 --> 02:04:34,559
subject to the same sort of mistakes

2271
02:04:34,660 --> 02:04:35,440
or mistakes.

2272
02:04:35,440 --> 02:04:38,474
Making the argument that there's
a credible enemy combatant in a

2273
02:04:38,474 --> 02:04:43,665
certain location and maybe making
up some of the data that is

2274
02:04:43,815 --> 02:04:45,795
contributing to that, that assessment

2275
02:04:45,820 --> 02:04:46,755
one, 100%.

2276
02:04:46,755 --> 02:04:47,865
There's no way to guarantee that.

2277
02:04:47,865 --> 02:04:51,285
That's not what's happening there,
there, it's, it's not the same kind

2278
02:04:51,285 --> 02:04:55,245
of deterministic,
rules-based, really systems that we

2279
02:04:55,245 --> 02:04:57,855
used to try to deploy and that frankly
didn't know, didn't work that well.

2280
02:04:57,855 --> 02:05:01,184
These things do work well in certain
contexts, but there's, we're, we're now,

2281
02:05:02,445 --> 02:05:06,375 at, at sea about, what, what
kind of mistakes can and, and do happen.

2282
02:05:06,375 --> 02:05:08,264
And the reports come in,
just like you said last week.

2283
02:05:08,264 --> 02:05:10,455
And, and they'll, they'll continue
to come in because these systems

2284
02:05:10,455 --> 02:05:13,214
fundamentally can't, can't
provide those kinds of guarantees.

2285
02:05:26,054 --> 02:05:30,719
Let me play another
clip of Amodei.

2286
02:05:32,010 --> 02:05:36,690
In his own words, making the argument
against fully autonomous AI weapons.

2287
02:05:36,690 --> 02:05:40,320
He doesn't want his,
his AI model used that way.

2288
02:05:40,380 --> 02:05:41,790
This again was on CBS.

2289
02:05:42,180 --> 02:05:46,260
The AI systems of today are
nowhere near reliable enough to

2290
02:05:46,260 --> 02:05:48,540
make fully autonomous weapons.

2291
02:05:48,809 --> 02:05:52,470
Anyone who's worked with
AI models understands that there's a

2292
02:05:52,470 --> 02:05:56,760
basic unpredictability to them that in a
purely technical way, we have not solved.

2293
02:05:56,880 --> 02:05:58,980
And there's an oversight question too.

2294
02:05:59,160 --> 02:06:04,890
If you have a large army of drones or
robots that can operate without any

2295
02:06:04,890 --> 02:06:08,610
human oversight, where there aren't
human soldiers to make the decisions

2296
02:06:08,610 --> 02:06:13,080
about who to target, who to shoot
at, that, that presents concerns.

2297
02:06:13,080 --> 02:06:16,800
And we need to have a conversation
about, about how that's overseen.

2298
02:06:18,030 --> 02:06:19,890
Well, that kind of
sounds like common sense.

2299
02:06:19,890 --> 02:06:22,170
What's the argument
against it from Pete Hegseth?

2300
02:06:23,040 --> 02:06:27,990
He basically is taking a stance
that in the heat of war, we should

2301
02:06:27,990 --> 02:06:29,940
use whatever's available to us.

2302
02:06:30,180 --> 02:06:33,840
And, the way that
it's gonna be maximally.

2303
02:06:34,140 --> 02:06:34,890
Um.

2304
02:06:35,400 --> 02:06:36,750
Efficient and lethal.

2305
02:06:36,960 --> 02:06:39,330
As, as, as we need it.

2306
02:06:39,719 --> 02:06:46,259
So he basically, decided,
about a month ago, earlier this year,

2307
02:06:46,290 --> 02:06:51,330
that Anthropic couldn't be trusted
to provide the technology if they weren't

2308
02:06:51,330 --> 02:06:55,530
going to be on board for anything the
Pentagon wanted to use it for and wanted

2309
02:06:55,530 --> 02:06:57,420
to change the terms of the contract.

2310
02:06:57,509 --> 02:06:59,160
And that's where we wound
up with his impasse.

2311
02:07:00,185 --> 02:07:06,725
Yeah, and you can imagine if listeners,
you've run into AI mistakes or what they

2312
02:07:06,725 --> 02:07:11,495
call hallucinations in little ways in
your life, what the implications might

2313
02:07:11,495 --> 02:07:16,325
be if AI models are picking out their
own targets without human supervision.

2314
02:07:16,325 --> 02:07:20,645
I'm thinking of a little thing,
Steven, where I was looking up sort

2315
02:07:20,645 --> 02:07:22,715
of wars going on around the world.

2316
02:07:23,075 --> 02:07:29,195
And the AI model in my Google
search was the first thing that came up.

2317
02:07:29,195 --> 02:07:32,615
And one of the ones that it
listed for Latin America was

2318
02:07:32,615 --> 02:07:34,775
Armenia versus Azerbaijan.

2319
02:07:35,345 --> 02:07:38,375
Well, guess where is not in Latin America?

2320
02:07:38,705 --> 02:07:44,135
But if it can do that, then, you
know, I don't know, maybe that girl's

2321
02:07:44,135 --> 02:07:51,605
school in Iran where 175 people reportedly
were killed already in this war, most of

2322
02:07:51,605 --> 02:07:54,245
them children, somebody made a mistake.

2323
02:07:54,515 --> 02:07:58,625
We don't know if it was AI that
made a mistake, but you don't want AI

2324
02:07:58,625 --> 02:08:03,155
to be able to potentially make that
mistake without human supervision, right?

2325
02:08:03,845 --> 02:08:04,415
That's right.

2326
02:08:04,415 --> 02:08:08,105
Oversight is so
important when you use ai, whether

2327
02:08:08,105 --> 02:08:13,685
you're, doing research and, you
know, and, and in the process of putting

2328
02:08:13,685 --> 02:08:17,195
that research into a final product,
you, you've gotta check everything very

2329
02:08:17,195 --> 02:08:22,715
carefully, and even more carefully
as you suggest if you're targeting.

2330
02:08:23,370 --> 02:08:27,900
Something with lethal force, you really
wanna be sure that you've got it right,

2331
02:08:27,900 --> 02:08:32,610
otherwise you will hit, a girl's
school instead of a military target.

2332
02:08:32,940 --> 02:08:38,460
And this is what
greatly concerns Anthropic and

2333
02:08:38,460 --> 02:08:41,550
I think down the road, I think
it, it should concern all of us.

2334
02:08:42,060 --> 02:08:45,360
In terms of how AI is used in general.

2335
02:08:45,570 --> 02:08:49,769
The big question is there's no
ceiling for the capabilities

2336
02:08:49,769 --> 02:08:51,540
of AI as it's being developed.

2337
02:08:51,780 --> 02:08:56,130
We're trying to figure this out and, companies like Anthropic

2338
02:08:56,130 --> 02:09:00,780
and Open ai, and Google and the
rest of them are developing this

2339
02:09:00,780 --> 02:09:04,260
technology and trying to make it safe.

2340
02:09:04,260 --> 02:09:09,180
But they're in a race between, you
know, each other, who's going to get to

2341
02:09:09,480 --> 02:09:11,610
artificial general intelligence first?

2342
02:09:11,970 --> 02:09:15,450
And that's more important
to them, I think, in the

2343
02:09:15,450 --> 02:09:17,519
priorities than doing it safely.

2344
02:09:17,760 --> 02:09:22,260
And we see through the military
lens how dangerous this can be.

2345
02:09:23,235 --> 02:09:30,405
And again, the two ethics clauses
in anthropics terms of service are

2346
02:09:30,795 --> 02:09:36,135
prohibiting, the AI model clause from
being used in autonomous weapon systems.

2347
02:09:36,135 --> 02:09:37,665
That's the one we've been talking about.

2348
02:09:37,935 --> 02:09:42,645
The other one is for mass
surveillance of American citizens.

2349
02:09:42,795 --> 02:09:46,065
Here's CEO Amodei, Amodei on that.

2350
02:09:46,425 --> 02:09:54,345
I am concerned that AI may be uniquely
well suited to autocracy and to deepening

2351
02:09:54,345 --> 02:09:56,505
the repression that we see in autocracies.

2352
02:09:56,505 --> 02:09:59,295
We already see it in the kind
of surveillance state that is

2353
02:09:59,295 --> 02:10:01,005
possible with today's technology.

2354
02:10:01,275 --> 02:10:05,715
But if you think of the extent to which
AI can make individualized propaganda,

2355
02:10:05,925 --> 02:10:12,615
can break into any computer system in
the world, can,

2356
02:10:12,645 --> 02:10:17,685
surveil everyone, everyone in a population
detect descent everywhere and suppress

2357
02:10:17,685 --> 02:10:22,635
it, make, a a, a
huge army of drones that could go after

2358
02:10:22,635 --> 02:10:24,525
each, each, each individual person.

2359
02:10:24,795 --> 02:10:26,475
It's, it's, it's really scary.

2360
02:10:27,269 --> 02:10:31,920
So Steven, is Amodei describing
there anything that the Pentagon

2361
02:10:32,400 --> 02:10:34,290
was specifically asking for?

2362
02:10:34,290 --> 02:10:37,590
Or is that a more hypothetical
concern at this point?

2363
02:10:38,430 --> 02:10:44,370
Well, I, I think it, it, it really
isn't, a remote possibility.

2364
02:10:44,700 --> 02:10:48,330
If you see the, the nature of the
argument and, you can go way

2365
02:10:48,330 --> 02:10:52,800
down into the weeds about what the,
in the contract with,

2366
02:10:53,309 --> 02:10:57,780
Anthropic the Pentagon wanted to do,
and what they've, forged with

2367
02:10:57,780 --> 02:11:02,760
a contract that they did execute with
OpenAI, which rush into to take it place.

2368
02:11:03,000 --> 02:11:07,200
They, you wanna watch
out for what they won't do.

2369
02:11:07,650 --> 02:11:14,730
And what they seem reluctant to grant
the AI companies is the promise that

2370
02:11:14,730 --> 02:11:20,550
they won't use all kinds of information
that's available, in the public sector

2371
02:11:20,550 --> 02:11:27,180
or from data brokers, and then use that
for, how they treat.

2372
02:11:27,705 --> 02:11:31,755
Citizens, and
it, it turns out what's legal to

2373
02:11:31,755 --> 02:11:38,145
do, could be much more powerful
if you use AI and much scarier in

2374
02:11:38,145 --> 02:11:40,575
terms of, a surveillance state.

2375
02:11:40,755 --> 02:11:44,385
So, and I think the, AI
companies, particularly Anthropic,

2376
02:11:44,595 --> 02:11:46,125
are very sensitive about that.

2377
02:11:46,125 --> 02:11:52,485
They don't want to be the trigger to,
make things a 1984 dystopia, and they're

2378
02:11:52,485 --> 02:11:55,365
trying to say, don't use us for that.

2379
02:12:07,815 --> 02:12:12,585
Ukraine instantly turned itself into a
laboratory of technological innovation.

2380
02:12:12,855 --> 02:12:16,935
The Ukrainian military, which has
done a remarkable job against a much

2381
02:12:16,935 --> 02:12:23,625
better funded and staffed military,
has done so by outsourcing so many

2382
02:12:23,630 --> 02:12:25,395
important things to private companies.

2383
02:12:25,635 --> 02:12:32,505
This experimentation has resulted in one
of the most, I think, important military.

2384
02:12:33,170 --> 02:12:37,280
Weapons developed in the last couple
of decades, and that is a hovering

2385
02:12:37,280 --> 02:12:42,799
drone, a drone that stays far enough
up in the atmosphere that it's almost

2386
02:12:42,799 --> 02:12:48,049
impossible to even sense from the
ground and it's poised to strike.

2387
02:12:48,975 --> 02:12:54,945
That means Russian troops in the field
are under constant possible attack.

2388
02:12:55,485 --> 02:12:59,055
Now, of course, the Russian military
developed the same technology, and so

2389
02:12:59,175 --> 02:13:04,455
troops in Ukraine and civilians in Ukraine
are also under the same constant threat.

2390
02:13:05,145 --> 02:13:10,095
So we can move forward to Gaza and
see how Israel, which has had its own

2391
02:13:10,095 --> 02:13:15,735
programs of mass surveillance using
artificial intelligence for more than

2392
02:13:15,735 --> 02:13:21,915
a decade, to try to identify hostile
parties both in the West Bank and in

2393
02:13:21,915 --> 02:13:24,375
Gaza, and frankly in Lebanon as well.

2394
02:13:24,735 --> 02:13:30,945
So given that for the last three years,
Israel has been trying to pick out

2395
02:13:31,215 --> 02:13:38,595
whom to attack in a densely populated
urban scene, like Gaza is really hard.

2396
02:13:38,955 --> 02:13:42,405
And so what they've done is
they've outsourced it to machines

2397
02:13:42,405 --> 02:13:44,625
that they hope are quite precise.

2398
02:13:44,625 --> 02:13:48,825
Now, what we do know is given the large
number of civilian casualties, the medical

2399
02:13:48,825 --> 02:13:53,175
facilities, the human rights organizations
that have been interrupted, bombed, et

2400
02:13:53,175 --> 02:13:57,975
cetera, the number of journalists killed
that either Israel is being duplicitous

2401
02:13:57,975 --> 02:14:02,595
about their desires for targeting, or the
systems just aren't as good as they hope.

2402
02:14:03,075 --> 02:14:08,955
You write that what was normalized
in Gaza will not stay in Gaza.

2403
02:14:09,555 --> 02:14:13,755
You say, this isn't just about
an export of technology, it's

2404
02:14:13,755 --> 02:14:16,635
an export of a habit of mind

2405
02:14:17,115 --> 02:14:20,685
that basically says we
can do everything faster.

2406
02:14:20,745 --> 02:14:22,305
We can do everything easier.

2407
02:14:22,515 --> 02:14:26,265
We can take human beings out of
the loop as much as possible,

2408
02:14:26,385 --> 02:14:31,995
and we're living in an age where
accountability seems an impossible dream.

2409
02:14:32,235 --> 02:14:39,345
You call autonomous weapons, quote,
accountability dissolving machines.

2410
02:14:39,715 --> 02:14:40,285
That's right.

2411
02:14:40,465 --> 02:14:45,595
If stuff is happening so fast and
so furiously, it's a lot harder for

2412
02:14:45,925 --> 02:14:51,475
inspectors general, for judge advocate
generals, attorneys general, it's a lot

2413
02:14:51,475 --> 02:14:57,115
harder for courts to make individual
discreet judgments in a timely fashion

2414
02:14:57,115 --> 02:14:58,975
to actually affect anything in the world.

2415
02:14:59,425 --> 02:15:06,235
And so the more we raise the metabolism
of warfare, the harder it is to oversee

2416
02:15:06,715 --> 02:15:11,335
the proper execution of warfare,
the moral execution of warfare, if

2417
02:15:11,335 --> 02:15:15,385
there is such a thing, the humanity
that might be embedded in warfare.

2418
02:15:15,385 --> 02:15:20,875
And look, we've been struggling definitely
since World War I, with increased

2419
02:15:21,205 --> 02:15:26,305
attention after Vietnam War with all of
its excesses of civilian casualties and

2420
02:15:26,305 --> 02:15:32,695
human rights violations to embed within
the US military, a sense of honor, a sense

2421
02:15:32,695 --> 02:15:39,625
of duty, a sense of legality, a sense
that lethality should be deployed only

2422
02:15:39,625 --> 02:15:42,115
under the most extreme circumstances.

2423
02:15:42,475 --> 02:15:47,575
And the US military has done an
outstanding job of that since

2424
02:15:47,635 --> 02:15:49,045
the end of the Vietnam War.

2425
02:15:49,165 --> 02:15:52,345
They've done it while certainly
under pressure from Congress and the

2426
02:15:52,345 --> 02:15:58,975
public, but the US military had done
a remarkable job to police itself

2427
02:15:59,125 --> 02:16:00,835
right up through the Afghan War.

2428
02:16:01,255 --> 02:16:03,175
And all of that is out the window now.

2429
02:16:03,235 --> 02:16:08,665
And AI being introduced at this level, at
this moment, only makes it more dangerous.

2430
02:16:09,355 --> 02:16:14,575
I'm gonna raise this experiment at
New College London, which tested out

2431
02:16:14,575 --> 02:16:20,394
three leading AI systems from OpenAI,
Anthropic, and Google in war simulations.

2432
02:16:20,605 --> 02:16:21,894
They found that.

2433
02:16:22,230 --> 02:16:27,540
The AI didn't have nearly the same
reservations as humans and used

2434
02:16:27,540 --> 02:16:31,349
nuclear weapons in 95% of cases.

2435
02:16:31,710 --> 02:16:37,080
Dean Ball, who helped developed President
Trump's AI policy in a senior post at the

2436
02:16:37,080 --> 02:16:44,279
White House last year posted on x quote,
if near medium future AI systems can be

2437
02:16:44,279 --> 02:16:51,539
used by the executive branch to arbitrary
ends with zero restrictions, the US will

2438
02:16:51,570 --> 02:16:55,260
functionally cease to be a republic.

2439
02:16:55,379 --> 02:16:58,260
I mean, where do you
think all this is going

2440
02:16:58,859 --> 02:16:59,340
to Dr.

2441
02:16:59,340 --> 02:17:00,000
Strangelove?

2442
02:17:00,150 --> 02:17:03,719
I think we're in a moment where, if
you remember Dr. Strange love, and I

2443
02:17:03,824 --> 02:17:11,910
know you do, the whole film ends with a
moment at which the momentum of decision

2444
02:17:11,910 --> 02:17:16,109
making for nuclear war was unstoppable.

2445
02:17:16,320 --> 02:17:21,240
This imposition and installation of AI
decision making systems that minimize

2446
02:17:21,240 --> 02:17:24,930
human influence over what kind of weapons
we're gonna use when we will use them,

2447
02:17:25,350 --> 02:17:30,809 sort of like satisfying the dream
of the extremists among the nuclear

2448
02:17:30,809 --> 02:17:33,299
strategists of the 1950s and sixties.

2449
02:17:33,690 --> 02:17:39,000
So, look, we've also had to
great credit of the United States an

2450
02:17:39,000 --> 02:17:45,660
actual debate for 70 years over when
and how to use the most lethal weapons.

2451
02:17:45,990 --> 02:17:51,270
We actually had debate at the highest
levels the fact that we trusted a

2452
02:17:51,270 --> 02:17:56,609
president, not just a president, like
eight presidents with the lives of the

2453
02:17:56,609 --> 02:18:03,690
entire world, understanding that most
human beings would make the right decision

2454
02:18:03,840 --> 02:18:05,969
when offered the right information.

2455
02:18:06,090 --> 02:18:08,910
That level of trust that we put
into presidents of both parties,

2456
02:18:09,209 --> 02:18:13,320
many of whom were not really
great people, says a lot about.

2457
02:18:13,695 --> 02:18:17,655
The fact that we've managed
to live this long and we had

2458
02:18:17,655 --> 02:18:19,185
a system that mostly worked.

2459
02:18:19,665 --> 02:18:21,045
Are we in that situation?

2460
02:18:21,045 --> 02:18:24,885
Do we have an executive we can
trust with the future of the world?

2461
02:18:24,945 --> 02:18:29,715
Who actually thinks about the
children in a girl's school

2462
02:18:29,985 --> 02:18:31,665
in a remote corner of Iran?

2463
02:18:31,665 --> 02:18:35,265
Do we have a present who thinks
about the long-term repercussions of

2464
02:18:35,265 --> 02:18:38,565
the use of certain weapons and the
deployment of certain technologies?

2465
02:18:38,925 --> 02:18:40,995
I think it's
safe to say we do not.

2466
02:18:41,355 --> 02:18:47,145
So all of that difficult work that has
been done in the military, in the Rand

2467
02:18:47,145 --> 02:18:51,405
Corporation, in the halls of Congress in
White House after White House, after White

2468
02:18:51,405 --> 02:18:54,105
House for decades, has been stripped out.

2469
02:18:54,135 --> 02:18:58,005
We are in a very dangerous moment
in which to rely on these systems.

2470
02:19:10,545 --> 02:19:16,575
We've seen a, a trajectory of the
military adopting AI tools over the last

2471
02:19:16,575 --> 02:19:20,085
decade as AI has continued to progress.

2472
02:19:20,685 --> 02:19:25,995
What's newer are large language models
like ChatGPT and Anthropic's Claude,

2473
02:19:26,355 --> 02:19:31,125
that it's been reported the military
is using in operations in Iran.

2474
02:19:31,245 --> 02:19:34,005
And so that's a, that's a pretty
significant development that we're seeing.

2475
02:19:34,455 --> 02:19:38,654
And the, the people wanna know how Claude
or chat GPT might be fighting this war.

2476
02:19:38,654 --> 02:19:40,605
Do we know war in Iran?

2477
02:19:41,025 --> 02:19:41,984
That's a great idea.

2478
02:19:41,984 --> 02:19:42,914
Let me help you with that.

2479
02:19:43,064 --> 02:19:45,254
Well, we don't know yet.

2480
02:19:46,244 --> 02:19:49,125
We can make some
educated guesses based on

2481
02:19:49,125 --> 02:19:50,925
what the technology could do.

2482
02:19:51,345 --> 02:19:55,515
AI technologies really great at
processing large amounts of information.

2483
02:19:55,755 --> 02:19:55,935
I

2484
02:19:55,935 --> 02:19:57,644
literally love processing.

2485
02:19:57,644 --> 02:20:00,734
The US militaries hit over
a thousand targets in Iran,

2486
02:20:00,795 --> 02:20:01,814
as you see very well.

2487
02:20:01,814 --> 02:20:03,615
They have no navy, it's been knocked out.

2488
02:20:04,154 --> 02:20:05,085
They have no air force.

2489
02:20:05,085 --> 02:20:06,015
It's been knocked out.

2490
02:20:06,585 --> 02:20:08,955
They have no air detection
that's been knocked out.

2491
02:20:08,955 --> 02:20:10,335
Their radar has been knocked out.

2492
02:20:10,724 --> 02:20:15,345
They need to then find ways to process
information about those targets.

2493
02:20:15,615 --> 02:20:18,345
So satellite imagery, for example,
of the targets they've hit

2494
02:20:18,435 --> 02:20:19,904
just about everything
that's been knocked out.

2495
02:20:19,935 --> 02:20:23,865
Looking at new potential targets,
prioritizing those processing

2496
02:20:23,865 --> 02:20:28,455
information and using AI to do that at
machine speed rather than human speed.

2497
02:20:28,935 --> 02:20:31,215
Human so slow mo ha ha ha.

2498
02:20:32,175 --> 02:20:32,595
Cheers.

2499
02:20:33,404 --> 02:20:39,404
Do we know any more about how the military
may have used AI in, say, Venezuela

2500
02:20:39,404 --> 02:20:44,925
on the attack that brought Nicholas
Maduro to Brooklyn of all places?

2501
02:20:44,925 --> 02:20:48,045
Brooklyn and the hell, because
we've recently found out

2502
02:20:48,045 --> 02:20:49,394
that AI was used there too.

2503
02:20:50,144 --> 02:20:54,375
So what we do know is that Anthropic's
AI tools have been integrated

2504
02:20:54,375 --> 02:20:58,935
into the US military's classified
networks, and so they can process

2505
02:20:58,935 --> 02:21:03,795
classified information to process
intelligence, to help plan operations

2506
02:21:03,795 --> 02:21:07,125
from writing emails to
rating enemy capital cities.

2507
02:21:07,425 --> 02:21:11,115
The Wall Street Journal reports that the
Pentagon used Anthropic's AI model

2508
02:21:11,144 --> 02:21:15,465
Claude as part of its operation to capture
Venezuelan president Nicholas Maduro.

2509
02:21:15,465 --> 02:21:18,585
There's no suggestion that Claude was
actually firing any of the missiles

2510
02:21:18,585 --> 02:21:20,564
or manning any of the machine guns.

2511
02:21:20,744 --> 02:21:23,535
Yeah, we've had this sort of
tantalizing details, okay.

2512
02:21:23,535 --> 02:21:25,394
That these tools were
used in the Maduro raid.

2513
02:21:25,724 --> 02:21:28,335
We don't know exactly how, so.

2514
02:21:28,875 --> 02:21:34,215
We've seen AI technology in a broad
sense used in other conflicts as well

2515
02:21:34,545 --> 02:21:40,365
in Ukraine, in Israel's operations in
Gaza, to do a couple different things.

2516
02:21:40,635 --> 02:21:44,595
One of the ways that AI is being
used in Ukraine in a different

2517
02:21:44,595 --> 02:21:50,655
kind of context is putting autonomy
aren't to drones themselves.

2518
02:21:53,505 --> 02:21:55,755
The drone now flies on autopilot mode.

2519
02:21:55,755 --> 02:22:00,465
Using our software, we assigned it with a
mission and it built its own flying route,

2520
02:22:00,525 --> 02:22:04,335
giving the munition instructions
on where it needs to go and

2521
02:22:04,335 --> 02:22:05,415
what it needs to look for.

2522
02:22:05,505 --> 02:22:10,245
And so when I was in Ukraine, one of
the things that I saw Ukrainian drone

2523
02:22:10,245 --> 02:22:16,425
operators and engineers demonstrate is
a little box like the size of a pack

2524
02:22:16,425 --> 02:22:21,285
of cigarettes that you could put onto
a small drone that would enable the,

2525
02:22:21,315 --> 02:22:23,325
once the human locks onto a target.

2526
02:22:24,030 --> 02:22:26,880
The drone can then carry out
the attack all on its own.

2527
02:22:27,240 --> 02:22:29,400
And that has been used
in a small, small way.

2528
02:22:29,400 --> 02:22:32,400
It's not necessarily widespread
use in Ukraine today.

2529
02:22:32,640 --> 02:22:37,410
So we're seeing AI begin to creep
into all of these aspects of military

2530
02:22:37,410 --> 02:22:42,090
operations in intelligence and planning
and logistics, but also right at the

2531
02:22:42,180 --> 02:22:48,750
edge in terms of, being used
where drones are completing attacks.

2532
02:22:48,930 --> 02:22:49,110
Okay.

2533
02:22:49,110 --> 02:22:52,530
So we know a little bit more about how
this technology was used in Ukraine.

2534
02:22:52,530 --> 02:22:53,940
How about with Israel and Gaza?

2535
02:22:55,830 --> 02:23:00,210
So there's been some reporting about how
the Israel defense forces have used AI

2536
02:23:00,240 --> 02:23:04,260
in Gaza, not necessarily large language
models, but machine learning systems that

2537
02:23:04,260 --> 02:23:10,440
can synthesize, infuse large amounts of
information, geolocation data, cell phone

2538
02:23:10,500 --> 02:23:15,810
data and connection, social media data
to bring this together, process all of

2539
02:23:15,810 --> 02:23:19,619
the information very quickly to develop
targeting packages, particularly in the

2540
02:23:19,619 --> 02:23:22,680
early phases of Israel's operations,

2541
02:23:22,980 --> 02:23:28,680
which suggests specific possible
targets, possible munitions warnings.

2542
02:23:28,680 --> 02:23:32,490
This system produces targets in
Gaza faster than a human can.

2543
02:23:32,760 --> 02:23:36,420
But it raises thorny questions
about, human

2544
02:23:36,420 --> 02:23:37,949
involvement in these decisions.

2545
02:23:38,430 --> 02:23:44,100
And one of the criticisms that had come
up was that humans were still approving

2546
02:23:44,100 --> 02:23:49,410
these targets, but that the volume of
strikes and the amount of information

2547
02:23:49,410 --> 02:23:53,610
that needed to be processed was such
that maybe human oversight in some cases

2548
02:23:53,610 --> 02:23:55,289
was a little bit more of a rubber stamp.

2549
02:23:55,949 --> 02:23:57,720
The question is, where does this go?

2550
02:23:58,050 --> 02:24:02,340
And are we heading a trajectory
where over time humans get pushed

2551
02:24:02,340 --> 02:24:06,660
out of the loop and we see down the
road fully autonomous weapons that

2552
02:24:06,660 --> 02:24:09,150
are making their own decisions about
whom to kill on the battlefield.

2553
02:24:11,655 --> 02:24:13,185
That's the direction things are headed.

2554
02:24:13,185 --> 02:24:17,205
So, no one's unleashing
the swarm of killer robots today, but

2555
02:24:17,655 --> 02:24:19,935
the trajectory is in that direction.

2556
02:24:20,355 --> 02:24:23,415
And I, maybe I'll make a comparison
here, to self-driving cars,

2557
02:24:23,745 --> 02:24:28,395
where car companies can map the
environment down to the centimeter.

2558
02:24:28,845 --> 02:24:31,845
They know the height of the curbs,
they know where the stoplights are.

2559
02:24:32,595 --> 02:24:36,495
They can test self-driving cars in the
actual environment they're gonna be in.

2560
02:24:36,495 --> 02:24:38,685
And when they're do something
weird that doesn't work, they

2561
02:24:38,685 --> 02:24:39,765
can update the algorithm.

2562
02:24:40,455 --> 02:24:41,445
We don't know our future war.

2563
02:24:41,445 --> 02:24:42,135
Is it gonna be fought?

2564
02:24:42,585 --> 02:24:43,965
It's an adversarial environment.

2565
02:24:43,965 --> 02:24:45,405
We don't know what the enemy's gonna do.

2566
02:24:45,405 --> 02:24:49,065
I mean, the US military is
finding this out right now in

2567
02:24:49,065 --> 02:24:50,535
its operations against Iran.

2568
02:24:51,015 --> 02:24:54,705
They're retaliating against US
bases against Gulf States, against

2569
02:24:54,705 --> 02:24:57,075
Israel using drones and missiles.

2570
02:24:57,825 --> 02:25:01,905
And now we're in a phase in
the Iran conflict where things

2571
02:25:01,905 --> 02:25:03,405
become super unpredictable.

2572
02:25:03,735 --> 02:25:07,365
People do an okay job of
adapting that unpredictability.

2573
02:25:08,355 --> 02:25:11,565
AI is not so great and sometimes
does some strange things

2574
02:25:12,255 --> 02:25:14,745
because you drew a parallel
to self-driving cars.

2575
02:25:15,315 --> 02:25:18,105
We've made an episode
about self-driving cars before in

2576
02:25:18,105 --> 02:25:20,085
which, I think our guests
said something like, well, if you're

2577
02:25:20,085 --> 02:25:22,005
worried about self-driving cars, you
know what we should really be worried

2578
02:25:22,005 --> 02:25:24,345
about is humans thinking about Iran.

2579
02:25:24,345 --> 02:25:28,215
We saw reports that a school
was bombed in Iran where mm-hmm.

2580
02:25:28,455 --> 02:25:30,520
Maybe 160 were killed.

2581
02:25:30,880 --> 02:25:32,355
A lot of them young.

2582
02:25:32,730 --> 02:25:38,550
Girls, children, presumably that
was a mistake made by a human.

2583
02:25:39,060 --> 02:25:44,550
Do we think that autonomous weapons will
be capable of making that same mistake or

2584
02:25:44,550 --> 02:25:46,620
will they be better at war than we are?

2585
02:25:49,050 --> 02:25:53,880
This question of will autonomous weapons
be better than humans are not, is like

2586
02:25:53,910 --> 02:25:58,170
one of the core issues of the debate
surrounding this technology because

2587
02:25:58,170 --> 02:26:01,500
proponents of autonomous weapons will say,
look, people make mistakes all the time,

2588
02:26:01,860 --> 02:26:04,500
and machines might be able to do better.

2589
02:26:05,070 --> 02:26:10,320
Part of that depends on how much the
militaries that are using this technology

2590
02:26:10,800 --> 02:26:13,110
are trying really hard to avoid mistakes.

2591
02:26:13,800 --> 02:26:20,040
If militaries don't care about civilian
casualties, then AI can allow militaries

2592
02:26:20,040 --> 02:26:24,840
to simply strike targets faster, in some
cases, even commit atrocities faster, if

2593
02:26:24,840 --> 02:26:27,330
that's what militaries are trying to do.

2594
02:26:27,690 --> 02:26:33,660
I think there is this really
important potential here to use

2595
02:26:33,660 --> 02:26:36,180
the technology to be more precise.

2596
02:26:36,510 --> 02:26:42,330
And if you look at the long arc of
precision guided weapons, let's say over

2597
02:26:42,330 --> 02:26:46,470
the last century or so, it's pointed
towards much more precision and warfare.

2598
02:26:46,590 --> 02:26:51,450
So if you look at the example of the US
strikes in Iran right now, it's worth

2599
02:26:51,450 --> 02:26:57,330
contrasting this with the widespread
aerial bombing campaigns against

2600
02:26:57,330 --> 02:27:00,330
cities that we saw in World War ii,
for example, where whole cities were

2601
02:27:00,330 --> 02:27:04,410
devastated in Europe and Asia because
the bombs just weren't precise at all.

2602
02:27:05,100 --> 02:27:08,760
And so Air Forces dropped just,
just massive amounts of ordinance

2603
02:27:09,300 --> 02:27:11,250
to try to hit even a single factory.

2604
02:27:11,670 --> 02:27:17,070
The possibility here is that AI could
make it better over time to allow

2605
02:27:17,070 --> 02:27:21,330
militaries to hit military targets
and avoid civilian casualties.

2606
02:27:21,630 --> 02:27:25,619
Now, if the data is wrong and they've
got the wrong target on the list, they're

2607
02:27:25,619 --> 02:27:29,490
gonna hit the wrong thing very precisely,
and AI's not necessarily gonna fix that.

2608
02:27:38,490 --> 02:27:40,800
Now, section C surveillance,

2609
02:27:44,310 --> 02:27:47,220
I wanna briefly ask you about
surveillance in general, and

2610
02:27:47,220 --> 02:27:49,650
in particular, Anthropic's
Fourth Amendment concern, right?

2611
02:27:49,650 --> 02:27:52,080
The Fourth Amendment says the government
can't unreasonably search you.

2612
02:27:52,109 --> 02:27:54,600
The best way to understand the
Fourth Amendment is by listening

2613
02:27:54,600 --> 02:27:55,769
to nine, nine problems by Jay-Z.

2614
02:27:55,769 --> 02:27:57,914
So if you need to take a break
and go listen to, to nine,

2615
02:27:57,920 --> 02:27:58,710
nine problems, that's great.

2616
02:27:58,769 --> 02:27:59,609
It, it's all in there.

2617
02:27:59,609 --> 02:28:01,920
I, I listened to it when I was in
law school and made perfect sense,

2618
02:28:01,920 --> 02:28:04,019
but the government generally needs
a warrant to like search you.

2619
02:28:04,080 --> 02:28:06,539
And as more and more of your
life goes online, there's lots

2620
02:28:06,539 --> 02:28:07,529
and lots of exceptions to this.

2621
02:28:07,529 --> 02:28:09,330
But the idea is they should
still need a warrant online.

2622
02:28:09,420 --> 02:28:12,570
Anthropic's argument is, well,
the AI will never get tired.

2623
02:28:12,570 --> 02:28:13,680
It can search everything all the time.

2624
02:28:13,680 --> 02:28:15,240
That means we're just gonna
do master surveillance.

2625
02:28:15,840 --> 02:28:20,490
But even before AI showed up, right,
the idea that the government could

2626
02:28:20,490 --> 02:28:24,300
search everything that belonged to
you was out there, the idea that the

2627
02:28:24,300 --> 02:28:27,359
government didn't need a warrant to
search all your stuff was out there.

2628
02:28:27,539 --> 02:28:30,660
The idea that if any of your data ever
went outside the country for a brief

2629
02:28:30,660 --> 02:28:34,560
second, the government intercepted there
was out there when I was in college

2630
02:28:34,560 --> 02:28:38,189
around the time of the Patriot Act,
the debate was they're not gonna search

2631
02:28:38,189 --> 02:28:41,910
your actual data, but they can get
the metadata and the metadata alone.

2632
02:28:41,910 --> 02:28:45,630
The data about your data will be enough
to precisely locate you at all times.

2633
02:28:45,660 --> 02:28:47,160
And that even that is too far.

2634
02:28:47,580 --> 02:28:51,330
And we, we've been doing this dance
of what can the government collect?

2635
02:28:51,660 --> 02:28:52,560
What is permissible?

2636
02:28:52,560 --> 02:28:56,070
What's, what do they need to keep
us all safe and what's too far?

2637
02:28:56,279 --> 02:28:57,359
Those lines have moved.

2638
02:28:57,600 --> 02:29:00,900
So just briefly describe the, the sort of.

2639
02:29:01,575 --> 02:29:06,075
Generalized concern about surveillance
at the scale and where we are now.

2640
02:29:06,135 --> 02:29:06,345
Yeah.

2641
02:29:06,345 --> 02:29:10,515
Before the AI situation
sort of exponentially made

2642
02:29:10,515 --> 02:29:11,445
everything more complicated.

2643
02:29:11,475 --> 02:29:11,775
Yeah.

2644
02:29:11,775 --> 02:29:15,045
And here I have to introduce another
concept that, that probably should have

2645
02:29:15,045 --> 02:29:18,015
mentioned earlier, but it is important,
which is called the third party Doctrine.

2646
02:29:18,075 --> 02:29:22,695
And so the idea with the Fourth Amendment
is that the government can't search you

2647
02:29:22,695 --> 02:29:26,715
or your things without a warrant, and it
can't get a warrant without probable cause

2648
02:29:26,715 --> 02:29:28,515
that you've committed some sort of crime.

2649
02:29:28,545 --> 02:29:32,715
But there's this concept which came
about decades ago called the Third

2650
02:29:32,715 --> 02:29:37,035
Party Doctrine, which says that
doesn't necessarily apply or doesn't

2651
02:29:37,035 --> 02:29:42,885
apply at all to things that aren't
yours, even if it is your data.

2652
02:29:42,945 --> 02:29:48,885
And so the, the earliest and most obvious
version of this was phone records that the

2653
02:29:48,885 --> 02:29:51,375
phone company had of who, who you called.

2654
02:29:51,735 --> 02:29:54,615
The phone companies weren't recording
your calls, but they were recording.

2655
02:29:54,645 --> 02:29:57,075
If I called you, there
would be a record at the phone

2656
02:29:57,075 --> 02:29:58,575
company that says Mike calls Nele.

2657
02:29:58,635 --> 02:30:02,865
What had been determined by multiple
courts was that the, the government can

2658
02:30:02,865 --> 02:30:08,595
go and request, and they don't need a
warrant for that because it's not a search

2659
02:30:08,625 --> 02:30:11,565
of your data, it's this third party.

2660
02:30:11,565 --> 02:30:15,945
And they can agree as a third party
to just hand over that, that data.

2661
02:30:16,005 --> 02:30:19,185
But that was like, cases from
the sixties and seventies where that

2662
02:30:19,185 --> 02:30:23,055
was determined that the government can
get access to that without a warrant.

2663
02:30:24,005 --> 02:30:28,744
When there wasn't that much third
party data out there, the rise of

2664
02:30:28,744 --> 02:30:33,515
computers and the internet changed that
now everything is third party data.

2665
02:30:33,515 --> 02:30:39,275
Everything that we do is collected by some
company somewhere and has a record of it.

2666
02:30:39,935 --> 02:30:45,304
And so basically every bit of data about
you, where you are, who you speak to, who

2667
02:30:45,304 --> 02:30:48,904
you interact with, what you say, where,
what you're doing, all of that is pretty

2668
02:30:48,904 --> 02:30:50,644
much held by third parties these days.

2669
02:30:51,095 --> 02:30:56,615
And so the, the third party doctrine
has sort of swallowed the entire

2670
02:30:56,615 --> 02:30:59,705
Fourth Amendment to some extent,
where any, anything that is about

2671
02:30:59,765 --> 02:31:03,515
you ha there that somebody else has,
there's a much lower standard for what

2672
02:31:03,515 --> 02:31:05,554
the government can do to request it.

2673
02:31:05,765 --> 02:31:06,545
Now, companies can,

2674
02:31:06,550 --> 02:31:09,724
and just to be specific, this
means when my data's in iCloud,

2675
02:31:09,785 --> 02:31:10,054
yes,

2676
02:31:10,324 --> 02:31:13,175
the government can go to Apple
and get my data out of iCloud

2677
02:31:13,175 --> 02:31:14,074
without ever telling me,

2678
02:31:14,164 --> 02:31:17,765
well, it, it's, it's, they can
request it so they can easily

2679
02:31:17,824 --> 02:31:19,295
request it without a warrant.

2680
02:31:19,715 --> 02:31:22,894
Then the company has its own
rights and can determine what

2681
02:31:22,894 --> 02:31:25,025
they want to do with that request.

2682
02:31:25,025 --> 02:31:26,164
They can just give it up.

2683
02:31:26,494 --> 02:31:31,175
They can, as most of them will do, is, if, if it's a serious

2684
02:31:31,175 --> 02:31:34,385
request, they can reject requests
out of hand or they can alert you

2685
02:31:34,505 --> 02:31:37,175
and they can say, and this is what
most of them will do, they'll alert

2686
02:31:37,175 --> 02:31:39,994
you and say, the government
is requesting some of your data.

2687
02:31:40,054 --> 02:31:42,425
You can go to
court and try and block them.

2688
02:31:42,635 --> 02:31:46,865
If not, we will hand over your data
in seven days or whatever it might be.

2689
02:31:47,285 --> 02:31:48,125
Again, it depends.

2690
02:31:48,125 --> 02:31:51,484
If it's a criminal investigation, then
there may be some sort of gag order where

2691
02:31:51,484 --> 02:31:53,045
the company's not allowed to tell you.

2692
02:31:53,314 --> 02:31:57,845
There's all sorts of situations,
but most of them involve less than,

2693
02:31:58,205 --> 02:32:02,585 the level of protection that the
Fourth Amendment would require if it

2694
02:32:02,585 --> 02:32:06,574
was data in your, or any information
or anything in your own home.

2695
02:32:06,935 --> 02:32:09,304
I'm asking this because it, it's
obvious to everybody listening to this.

2696
02:32:09,455 --> 02:32:13,625
The amount of data you have on someone
else's cloud server is massive, right?

2697
02:32:13,625 --> 02:32:16,474
Every, every single thing that you
do, generally on the internet

2698
02:32:16,474 --> 02:32:19,025
now is backed up in some way or
recorded in some way on someone

2699
02:32:19,025 --> 02:32:21,035
else's servers and the government has.

2700
02:32:21,555 --> 02:32:25,185
Found this way to get around
the Fourth Amendment and say,

2701
02:32:25,185 --> 02:32:26,445
well, that's not actually yours.

2702
02:32:26,445 --> 02:32:28,095
It belongs to Amazon or whatever.

2703
02:32:28,095 --> 02:32:29,234
And we can go talk to Amazon.

2704
02:32:29,445 --> 02:32:33,105
And Amazon has to stand in the
middle of that process and say,

2705
02:32:33,105 --> 02:32:36,074
we've invented another process
to somewhat protect the people.

2706
02:32:36,645 --> 02:32:41,355
And I look at that and, the,
when I was covering the first third party

2707
02:32:41,355 --> 02:32:45,164
doctrine cases that covered the cloud
services and the government kept winning.

2708
02:32:45,675 --> 02:32:47,865
I mean, that's basically when I turned
in the Joker and I was like, all

2709
02:32:47,865 --> 02:32:51,255
of this stuff that we're pretending
about Textualism and the plain mean,

2710
02:32:51,255 --> 02:32:52,935
like none of this means anything.

2711
02:32:52,965 --> 02:32:53,205
Yeah.

2712
02:32:53,355 --> 02:32:56,295
'cause we, we've just, it
horsepower to backdoor using this

2713
02:32:56,295 --> 02:32:58,395
ancient law into everyone's data.

2714
02:32:58,484 --> 02:33:01,845
And then I look at this and I, I
look at Anthropic and I say, well,

2715
02:33:01,845 --> 02:33:04,095
this is the same pattern, right?

2716
02:33:04,095 --> 02:33:08,385
This is a private company saying, okay, we understand your position, we

2717
02:33:08,385 --> 02:33:11,805
understand that you've reinterpreted
the law to mean this thing, and

2718
02:33:11,805 --> 02:33:17,234
we're gonna put some process in
between you, our tool and the data of

2719
02:33:17,234 --> 02:33:18,945
Americans flowing through our service.

2720
02:33:19,425 --> 02:33:22,725
And I, I'm just wondering if you
see that parallel there between

2721
02:33:22,725 --> 02:33:26,475
Anthropic and Amazon and Azure and
whatever other cloud services that

2722
02:33:26,475 --> 02:33:27,765
exist that hold so much of our data.

2723
02:33:27,914 --> 02:33:28,425
Yeah.

2724
02:33:28,425 --> 02:33:30,345
Though I think there are a
few clarifications that are

2725
02:33:30,345 --> 02:33:32,865
important here that, that make
this a little bit different.

2726
02:33:32,865 --> 02:33:36,705
And in fact, it was reported,
I think the New York Times had this

2727
02:33:36,705 --> 02:33:41,414
reporting first that, the
main clause that was most important to

2728
02:33:41,414 --> 02:33:46,695
Anthropic was specifically about data
collected from commercial services and

2729
02:33:46,695 --> 02:33:52,125
not being able to use Claude on that
data, which is, exactly

2730
02:33:52,154 --> 02:33:53,865
this issue in terms of third party data.

2731
02:33:54,225 --> 02:33:58,395
But I do want to clarify the,
the main difference where, what we

2732
02:33:58,395 --> 02:34:02,145
were just talking about before this
with, with, Amazon or other

2733
02:34:02,145 --> 02:34:04,845
third parties, hosting your data.

2734
02:34:04,994 --> 02:34:09,164
That was cases where they were, you
know, because of where they sit in

2735
02:34:09,164 --> 02:34:12,705
the ecosystem, they were hosting
your data directly with Claude.

2736
02:34:12,705 --> 02:34:17,175
It's not that any, anyone is worried
about the NSA looking through like your.

2737
02:34:17,535 --> 02:34:19,010
Claude usage, right?

2738
02:34:19,015 --> 02:34:19,185
Right.

2739
02:34:19,185 --> 02:34:23,475
It's, it's about them going out and
getting third party data from an

2740
02:34:23,475 --> 02:34:28,185
Amazon, or more likely,
the, the sort of sneaky, hidden data

2741
02:34:28,185 --> 02:34:32,505
brokers that serve ads on your phones
and know all your location and, and

2742
02:34:32,505 --> 02:34:34,035
your interests and things like that.

2743
02:34:34,095 --> 02:34:38,025
And then feeding that into a system
that Claude would then work on.

2744
02:34:38,085 --> 02:34:40,755
That's what Anthropic really
didn't want to be a part of.

2745
02:34:40,785 --> 02:34:45,555
So wherever or however the government
would collect that data from a third

2746
02:34:45,555 --> 02:34:49,455
party, Claude said, we don't, we don't
want our tool to be used on that data.

2747
02:34:49,635 --> 02:34:54,495
There's a piece of that that just feels
like Apple famously stands up to the FBI,

2748
02:34:54,645 --> 02:34:57,525
but a backdoor in the iPhone and Apple
says no, and they stand up to Trump.

2749
02:34:57,825 --> 02:35:02,835
And there's just a part of, however
our system works, in which big

2750
02:35:02,835 --> 02:35:06,465
private companies get to say no to the
government on behalf of their customers.

2751
02:35:06,795 --> 02:35:10,905
And this felt the same, in the same
way that, apple again

2752
02:35:10,905 --> 02:35:13,665
won't put a backdoor in the iPhone
or the big cloud providers say.

2753
02:35:13,725 --> 02:35:16,215
There's a little bit of a process
you have to jump through before

2754
02:35:16,215 --> 02:35:17,775
you get the individual data here.

2755
02:35:17,775 --> 02:35:21,705
It seems like Anthropic is saying,
we're not just gonna do bulk analysis

2756
02:35:21,705 --> 02:35:24,795
of data that you have acquired
from other parties because that

2757
02:35:24,795 --> 02:35:28,785
leads to 24 7 mass surveillance of
Americans and we don't wanna do that.

2758
02:35:29,445 --> 02:35:33,615
And that seems like a bridge
too far for this administration.

2759
02:35:33,930 --> 02:35:35,940
I, is there any coming back from that?

2760
02:35:37,200 --> 02:35:41,250
I mean, we'll see in the past when
that's happened, and it's happened

2761
02:35:41,250 --> 02:35:45,240
plenty of times with, with most of the,
the large tech companies, at some point

2762
02:35:45,240 --> 02:35:48,630
they've said something is a bridge
too far, and where that normally goes

2763
02:35:48,630 --> 02:35:54,330
is to court and that, the,
the companies will go to court or the

2764
02:35:54,330 --> 02:35:57,060
administration will go to court and
there'll be some sort of court battle.

2765
02:35:57,060 --> 02:36:00,990
I mean, backdooring the iPhone
is, is a perfect example of that sort

2766
02:36:00,990 --> 02:36:04,440
of went to court and they sort of fought
it out though they never quite got to

2767
02:36:04,440 --> 02:36:08,580
a conclusion 'cause the FBI eventually
did just, manually break into

2768
02:36:08,580 --> 02:36:11,970
the iPhone and then didn't want the
court ruling to, to ruin the future.

2769
02:36:12,390 --> 02:36:17,910
But in this case, where
the escalation is and where this is,

2770
02:36:17,940 --> 02:36:22,110
this is different than those past
situations is that rather than just going

2771
02:36:22,110 --> 02:36:28,200
to court, they did this supply chain
risk designation, which is just insane.

2772
02:36:28,230 --> 02:36:33,900
This idea that this tool which
was designed to stop, far

2773
02:36:33,900 --> 02:36:39,240
potential foreign malicious actors from
supplying technology that could then,

2774
02:36:39,270 --> 02:36:43,290
Put in hidden surveillance
tools into the larger technology

2775
02:36:43,290 --> 02:36:48,570
stack that those could be banned to
apply that to a US-based company.

2776
02:36:49,215 --> 02:36:54,135
Basically for having an ethics
policy feels, like a

2777
02:36:54,135 --> 02:36:58,425
real, real, misuse of that tool.

2778
02:36:58,575 --> 02:37:01,275
And even that tool, I think was
questionable in, in some ways, but

2779
02:37:01,275 --> 02:37:04,545
you could understand the, the impetus
behind it for when you're talking

2780
02:37:04,545 --> 02:37:09,735
about like a Chinese networking firm
or something along those lines here.

2781
02:37:09,765 --> 02:37:10,695
It makes no sense.

2782
02:37:10,700 --> 02:37:14,925
And so the, the reaction to this
goes so far beyond what would

2783
02:37:14,925 --> 02:37:16,575
normally be seen in this case.

2784
02:37:16,815 --> 02:37:19,905
You could see, traditionally
there would be some sort of court

2785
02:37:19,905 --> 02:37:23,565
case and either side could start it
and it would just be a battle about,

2786
02:37:23,805 --> 02:37:27,375
What, what,
how the contract could be applied.

2787
02:37:27,825 --> 02:37:30,165
But that's not what is happening here.

2788
02:37:30,165 --> 02:37:33,555
That, that this administration is
effectively saying, if you don't

2789
02:37:33,555 --> 02:37:37,185
give us absolutely everything that
we want, if you don't set up your

2790
02:37:37,185 --> 02:37:41,745
tools to work the way we want them
to work, then we will effectively

2791
02:37:41,895 --> 02:37:44,055
try to destroy your entire business.

2792
02:37:44,085 --> 02:37:45,885
And that that is an escalation.

2793
02:37:59,235 --> 02:38:02,385
So I just wanna, read some New York
Times reporting here because they had

2794
02:38:02,385 --> 02:38:05,715
sources inside of these negotiations
just so people can understand

2795
02:38:06,075 --> 02:38:07,965
what apparently broke down here.

2796
02:38:08,445 --> 02:38:10,155
This is from the New York Times.

2797
02:38:10,455 --> 02:38:13,725
Emile Michael, who is the under
secretary at the Pentagon for research

2798
02:38:13,725 --> 02:38:17,415
and Engineering, used to be at Uber,
was, doing this negotiation with

2799
02:38:17,415 --> 02:38:21,615
Anthropic, was on a call with Anthropic
exec executives Friday afternoon.

2800
02:38:22,125 --> 02:38:25,905
He said the Pentagon wanted
anthropic to allow for collection and

2801
02:38:25,905 --> 02:38:31,515
analysis of unclassified commercial
bulk data on Americans such as

2802
02:38:31,605 --> 02:38:34,485
geolocation and web browsing data.

2803
02:38:34,845 --> 02:38:37,455
This is according to people
briefed on the negotiations.

2804
02:38:37,965 --> 02:38:42,225
Anthropic told the Pentagon that it
was willing to let its technology

2805
02:38:42,225 --> 02:38:47,145
be used by the NSA for classified
material collection under the Foreign

2806
02:38:47,145 --> 02:38:48,945
Intelligence Surveillance Act.

2807
02:38:48,945 --> 02:38:53,385
So FIA, but the company wanted a
legally binding promise from the

2808
02:38:53,385 --> 02:38:59,385
Pentagon not to use its technology
on unclassified commercial data.

2809
02:38:59,385 --> 02:39:04,785
So more of like that broad sweeping
surveillance versus having to go and get a

2810
02:39:04,785 --> 02:39:07,575
judge to sign off on a, on a FIO warrant.

2811
02:39:07,785 --> 02:39:12,045
Is there anything that stands out to
you about that dispute specifically?

2812
02:39:12,975 --> 02:39:18,015
I think we have companies trying to
step in, as in some cases admirably,

2813
02:39:18,075 --> 02:39:20,235 to fill in the gaps in the law.

2814
02:39:20,235 --> 02:39:23,505
So, FISA HA leaves it
a little bit ambiguous about,

2815
02:39:24,135 --> 02:39:25,695 what can and can't be done.

2816
02:39:25,695 --> 02:39:30,135
It's intended to allow intelligence
collection on foreign adversaries

2817
02:39:30,135 --> 02:39:32,295
and in, in intelligence situations.

2818
02:39:32,295 --> 02:39:36,195
But as legal scholars have noted
it, it sweeps in collection of

2819
02:39:36,195 --> 02:39:38,625
data on Americans as well, in
all sorts of under related ways.

2820
02:39:38,625 --> 02:39:43,670
And so I think, anthropic,
understands this, this gap and tried

2821
02:39:43,785 --> 02:39:46,245
to fill in where the law does not.

2822
02:39:46,305 --> 02:39:48,285
So that's admirable.

2823
02:39:48,285 --> 02:39:51,435
But again, outsourcing this
to the decisions of a few people at the,

2824
02:39:51,435 --> 02:39:54,405
at the heads of these companies with,
with who are pressured by different

2825
02:39:54,405 --> 02:39:57,585
incentives than the public interest
is, is dangerous to say the least.

2826
02:39:57,585 --> 02:39:57,645
So.

2827
02:39:58,770 --> 02:40:02,100
Open AI inks, the contract
with the Department of Defense.

2828
02:40:02,520 --> 02:40:06,030
They set in a blog post, the company
set in a blog post that it had similar

2829
02:40:06,030 --> 02:40:08,040
guardrails to what Anthropic wanted.

2830
02:40:08,040 --> 02:40:10,200
And you, you've touched
on that a little bit.

2831
02:40:10,680 --> 02:40:18,480
The analysis that I've read is that
there were some differences in that

2832
02:40:18,840 --> 02:40:23,850
Anthropic was basically saying, we, the
product that we give you, we would've

2833
02:40:23,850 --> 02:40:29,370
set these technical limits where this
surveillance could not be accomplished.

2834
02:40:29,370 --> 02:40:32,460
Basically like a hardwired type
thing in their technology that

2835
02:40:32,460 --> 02:40:33,540
they provided the Pentagon.

2836
02:40:33,870 --> 02:40:39,360
Whereas OpenAI was basically saying,
we are gonna rely on the Department

2837
02:40:39,360 --> 02:40:41,790
of Defense to follow the law here.

2838
02:40:41,790 --> 02:40:43,350
And that that was a key difference.

2839
02:40:43,350 --> 02:40:44,820
Do you think that that's
an important difference?

2840
02:40:44,820 --> 02:40:45,930
Do you, do you buy that?

2841
02:40:48,150 --> 02:40:53,160
It's hard to think
about it without the context

2842
02:40:53,160 --> 02:40:54,240
of the current administration.

2843
02:40:54,240 --> 02:40:54,420
Right?

2844
02:40:54,420 --> 02:40:57,900
So they, they've demonstrated time
and again that, that they're not,

2845
02:40:57,960 --> 02:41:00,480
they don't consider themselves to be
subject to the, to the rule of law.

2846
02:41:00,480 --> 02:41:07,530
And so the idea that we'll leave it up to
the DOD to, follow the law, everyone

2847
02:41:07,530 --> 02:41:08,580
knows what they're doing here, right?

2848
02:41:08,580 --> 02:41:12,180
Every, everyone knows that this
is not gonna, that this is trying

2849
02:41:12,180 --> 02:41:15,780
to avoid any responsibility from
OpenAI and shift blame to the DOD.

2850
02:41:16,140 --> 02:41:19,920
While we know that there's not gonna
be any operational rules there.

2851
02:41:19,920 --> 02:41:22,560
So again, Anthropic tried
to step up and say, well, let's at least

2852
02:41:22,560 --> 02:41:26,940
try to have some kind of legally binding, language that, that separates these,

2853
02:41:26,970 --> 02:41:28,680
these two things and OpenAI did not.

2854
02:41:29,250 --> 02:41:33,420
So, again, I think,
I think it all is a manifestation

2855
02:41:33,420 --> 02:41:37,260
of, of the fact that we don't have
any kind of coherent mechanism for,

2856
02:41:37,260 --> 02:41:39,270
for AI governance, at, at all.

2857
02:41:39,270 --> 02:41:43,650
And so these kinds of situations
are some sense are in inevitable.

2858
02:41:43,980 --> 02:41:49,920
At the same time, the Pentagon is saying,
and, and Secretary Heg, Seth, and the

2859
02:41:49,950 --> 02:41:51,510
folks who are negotiating this deal.

2860
02:41:52,830 --> 02:42:00,240
You can't have a company that we contract
to provide a service deciding what we do.

2861
02:42:00,420 --> 02:42:02,430
With it, we are
the Department of Defense.

2862
02:42:02,430 --> 02:42:07,050
Like we know what the law is
and we know, what our

2863
02:42:07,050 --> 02:42:09,210
strategy is and we'll be compliant.

2864
02:42:09,450 --> 02:42:12,510
But a private company making those
decisions for us, like that would

2865
02:42:12,565 --> 02:42:17,370
never happen with Raytheon being
consulted on how one of its, you

2866
02:42:17,370 --> 02:42:19,110
know, drones was used for example.

2867
02:42:19,110 --> 02:42:26,580
Is is the fact that Anthropic
wanted those limits overstepping, you

2868
02:42:26,580 --> 02:42:28,770
know, its role as, as a private company

2869
02:42:30,390 --> 02:42:30,990
To some extent.

2870
02:42:30,990 --> 02:42:35,940
This might be a better question for, those who study the law, but, you

2871
02:42:35,940 --> 02:42:39,000
know, I do think there is a material
difference between what's going on with

2872
02:42:39,030 --> 02:42:42,990
AI as a technology and what's going on
with, building planes and,

2873
02:42:42,990 --> 02:42:44,940
and, and, and building other weapons.

2874
02:42:44,970 --> 02:42:49,350
First of all, the speed at
which those technologies have

2875
02:42:49,350 --> 02:42:53,580
been developed in the past allowed
for the law to kind of catch up.

2876
02:42:53,820 --> 02:42:55,740
It was also in a very
different context in terms of

2877
02:42:55,740 --> 02:42:56,910
the current administration and

2878
02:42:57,120 --> 02:43:00,089
basically like we know
what a. Drone or a plane

2879
02:43:00,089 --> 02:43:00,449
can, and

2880
02:43:00,455 --> 02:43:03,480
can does, and AI is changing all the time.

2881
02:43:03,480 --> 02:43:03,960
That's right.

2882
02:43:03,960 --> 02:43:07,679
And, and, and, we
collectively aren't always sure of

2883
02:43:07,679 --> 02:43:09,029
the capabilities of these things.

2884
02:43:09,029 --> 02:43:12,419
And so, there, there's a
lot of market incentive to position

2885
02:43:12,419 --> 02:43:15,900
these technologies as,
ultra powerful and, magical.

2886
02:43:15,900 --> 02:43:18,269
And the hype, the hype is very
real, but the fact is that they

2887
02:43:18,269 --> 02:43:21,869
make extremely strange non-human
errors with some regularity.

2888
02:43:21,990 --> 02:43:26,310
And so there is a danger in
deploying these simply because of

2889
02:43:26,310 --> 02:43:28,980
the uncertainty and the scale with
which they, with which they operate.

2890
02:43:28,980 --> 02:43:33,210
And I don't think anyone would argue
that the laws are currently designed

2891
02:43:33,210 --> 02:43:35,429
to cover those, those, those cases.

2892
02:43:47,849 --> 02:43:50,910
And there's quite a lot of law
here about mass surveillance.

2893
02:43:50,910 --> 02:43:53,789
There's a lot of history, a
lot of controversial history.

2894
02:43:54,179 --> 02:43:59,429
The entire character of Edward Snowden
exists because of controversies

2895
02:43:59,429 --> 02:44:00,480
around mass surveillance.

2896
02:44:00,839 --> 02:44:05,220
And it all comes down to, I think
you are the one who posted this, the

2897
02:44:05,250 --> 02:44:08,730
National Security Agency, which is part
of the Department of Defense, which

2898
02:44:08,730 --> 02:44:10,050
we have to call the Department of War.

2899
02:44:10,050 --> 02:44:10,830
Now for some reason,

2900
02:44:11,039 --> 02:44:12,300
we don't have to do anything.

2901
02:44:12,535 --> 02:44:14,490
We, we don't, that's true in America.

2902
02:44:14,490 --> 02:44:15,240
We don't have to do anything.

2903
02:44:15,269 --> 02:44:21,240
But the National Security Agency has
basically redefined what a lot of words

2904
02:44:21,240 --> 02:44:26,880
mean out of like colloquial English to
mean, oh, we can just do surveillance.

2905
02:44:27,179 --> 02:44:30,240
And then every so often there's a
scandal when people discover that

2906
02:44:30,240 --> 02:44:31,199
they're just doing surveillance.

2907
02:44:31,380 --> 02:44:33,089
So just set the stage there.

2908
02:44:33,089 --> 02:44:39,119
And I don't wanna rewind you all the way,
but it's been quite a lot of time,

2909
02:44:39,240 --> 02:44:40,955
where this pattern has repeated itself.

2910
02:44:41,595 --> 02:44:42,165
Yeah.

2911
02:44:42,165 --> 02:44:46,245
And, and yeah, it, it sort of depends
on how deep you want to go, but the,

2912
02:44:46,245 --> 02:44:51,825
the sort of short version is obviously
in the post nine 11 world, we, the US

2913
02:44:51,825 --> 02:44:57,825
passed the Patriot Act, which had, some ability for the government

2914
02:44:57,825 --> 02:45:01,845
to engage in surveillance, which was
supposed to be for, protecting

2915
02:45:01,845 --> 02:45:03,915
us against future terrorist threats.

2916
02:45:03,945 --> 02:45:08,475
And the over time that got
interpreted in interesting ways

2917
02:45:08,475 --> 02:45:10,215
and there were some limits on that.

2918
02:45:10,275 --> 02:45:14,595
We also had the FISA court, which is
a special court that is supposed to,

2919
02:45:15,315 --> 02:45:19,125
Review the intelligence
community and their activities.

2920
02:45:19,155 --> 02:45:23,955
But, has been traditionally
a one-sided court, only one side

2921
02:45:23,955 --> 02:45:27,555
gets to plead their case to that
court, and it's all done in secret.

2922
02:45:28,005 --> 02:45:30,975
And so there's a lot of
stuff that, that was not known.

2923
02:45:31,545 --> 02:45:35,265
And then there was one other piece
in all of this, which goes all the

2924
02:45:35,265 --> 02:45:41,535
way back to Ronald Reagan, which is
executive order 12 3 3, which is

2925
02:45:41,535 --> 02:45:47,115
supposedly about setting out the rules
of the road for intelligence collection.

2926
02:45:47,115 --> 02:45:52,755
And so you have these three sets of laws,
well, a few sets of laws

2927
02:45:52,755 --> 02:45:55,935
and an executive order that to the public.

2928
02:45:56,025 --> 02:46:00,045
The parts that you can read seem to
say certain things about what our

2929
02:46:00,045 --> 02:46:03,675
government and the NSA in particular
can do in terms of surveillance.

2930
02:46:04,035 --> 02:46:09,045
Which when read with a plain English
dictionary, the nature of which you

2931
02:46:09,045 --> 02:46:13,695
and I probably have and understand,
we would come away with a belief

2932
02:46:13,695 --> 02:46:19,515
that the NSA's ability to surveil
Americans was, was very limited in fact.

2933
02:46:20,415 --> 02:46:24,435
To the point that they're
supposed to, if they realize

2934
02:46:24,435 --> 02:46:29,325
that they are surveilling a US person,
that they're supposed to immediately

2935
02:46:29,325 --> 02:46:35,115
stop and cry foul and, erase
the data and all of this other stuff.

2936
02:46:35,265 --> 02:46:40,245
And there were rumors for a
while that that was not really

2937
02:46:40,245 --> 02:46:45,015
happening and there were hints and
in particular, Senator Ron Wyden was

2938
02:46:45,015 --> 02:46:50,055
very vocal about, going on
the floor of the Senate and saying,

2939
02:46:50,895 --> 02:46:52,905
Something is not right here.

2940
02:46:52,905 --> 02:46:57,915
And I can't quite tell you what, or
in hearings he would ask intelligence

2941
02:46:57,915 --> 02:47:03,345
officials, are you or are you not
collecting mass data on Americans?

2942
02:47:03,825 --> 02:47:09,135
And they would either deflect
or in some cases outright lie.

2943
02:47:09,315 --> 02:47:15,075
And I believe it was one hearing
in like 2012 with James Clapper,

2944
02:47:15,260 --> 02:47:18,015
who was the director of National
Intelligence at the time.

2945
02:47:18,045 --> 02:47:21,225
He was asked directly on this
point, and he basically said, no,

2946
02:47:21,225 --> 02:47:22,935
we don't collect data on Americans.

2947
02:47:23,325 --> 02:47:24,345
And that was.

2948
02:47:24,945 --> 02:47:31,335
A big part of what inspired Ed Snowden
to leak the data, the, the, the reports

2949
02:47:31,335 --> 02:47:37,365
that he leaked to Glenn Greenwald and,
Bart Gelman and Laura Laura pores as well.

2950
02:47:37,425 --> 02:47:42,765
So from all of that, what we began to
discover was that the NSA has its own

2951
02:47:42,795 --> 02:47:46,244
dictionary that is somewhat different
than the dictionary that you and I use,

2952
02:47:46,305 --> 02:47:50,595
such that they can interpret words in
ways that are different than the plain

2953
02:47:50,595 --> 02:47:58,875
English meaning of them, including words
like target, which feels like kind of

2954
02:47:58,875 --> 02:48:03,795
a key word, but so, a, a sort
of broad understanding of what this is.

2955
02:48:03,795 --> 02:48:07,425
In theory, they're only supposed
to target people who are not US

2956
02:48:07,425 --> 02:48:08,984
persons, I think is the phrase.

2957
02:48:09,885 --> 02:48:18,404
But the way it had been interpreted
over time was that, anything that

2958
02:48:18,675 --> 02:48:23,715
mentions that person, any, anything
that is about a foreign person,

2959
02:48:24,135 --> 02:48:29,324
is now fair game, even if that is
the communications of a US person.

2960
02:48:29,654 --> 02:48:35,414
So if you and I were to text each
other and mention a foreign person,

2961
02:48:35,505 --> 02:48:40,605
that is now fair game for the NSA
to collect and to keep and to store,

2962
02:48:40,635 --> 02:48:42,074
there's a second part of this.

2963
02:48:42,074 --> 02:48:46,560
I mentioned, first the, the
executive order 12 3 3 from Ronald

2964
02:48:46,560 --> 02:48:52,845
Reagan, which effectively allowed
the NSA as as the technology changed

2965
02:48:52,845 --> 02:48:54,105
over time and the internet grew.

2966
02:48:54,105 --> 02:48:59,324
It allowed the NSA to tap into
foreign communications, but that

2967
02:48:59,324 --> 02:49:05,835
included any communications that
maybe left the US on route somewhere.

2968
02:49:06,074 --> 02:49:11,355
So if, if I'm texting you and
a message went from me in California

2969
02:49:11,565 --> 02:49:16,695
through a fiber optic cable that happened
to leave the us, the NSA could put a

2970
02:49:16,695 --> 02:49:20,205
tap in the part once it's outside the
US collect that information, even if

2971
02:49:20,205 --> 02:49:22,275
it was just going to you within the us.

2972
02:49:23,805 --> 02:49:27,885
Then what they could do is keep that
information, even if it was on US

2973
02:49:27,885 --> 02:49:32,685
persons, and they could do specific
searches on that later, sometimes

2974
02:49:32,685 --> 02:49:34,365
referred to as backdoor searches.

2975
02:49:34,605 --> 02:49:37,125
So they collected this information
that we believe they weren't

2976
02:49:37,125 --> 02:49:39,375
supposed to collect in the first
place, but they could keep it.

2977
02:49:39,765 --> 02:49:42,915
And they promised, they sort of pinky
swore that they would keep it private.

2978
02:49:42,915 --> 02:49:48,135
But if they did, a search and found
like that you or I mentioned a foreign

2979
02:49:48,135 --> 02:49:51,615
person, then suddenly it was fair game
for them to do whatever they want with.

2980
02:49:52,355 --> 02:49:56,495
In total that has turned into a
world in which they can basically

2981
02:49:56,495 --> 02:50:01,145
collect any information that ha
happens to touch outside the us.

2982
02:50:01,385 --> 02:50:06,605
And even if it is entirely between two
US persons, if they mention or even

2983
02:50:06,605 --> 02:50:10,895
hint at someone who is not a US person,
suddenly it is fair game to be collected.

2984
02:50:11,225 --> 02:50:15,905
And from that, we've gotten what appears
to be a, a form of mass surveillance of US

2985
02:50:15,905 --> 02:50:21,995
persons by an NSA that claims and publicly
states that they do not spy on US persons.

2986
02:50:22,085 --> 02:50:23,975
How did we get to this point?

2987
02:50:24,035 --> 02:50:24,245
Right?

2988
02:50:24,245 --> 02:50:27,515
This is a lot of incremental baby steps.

2989
02:50:27,725 --> 02:50:29,675
You mentioned James Clapper in 2012.

2990
02:50:29,675 --> 02:50:31,385
That's the Obama administration.

2991
02:50:31,445 --> 02:50:32,465
You mentioned Ronald Reagan.

2992
02:50:32,555 --> 02:50:33,515
That's the eighties.

2993
02:50:33,515 --> 02:50:36,965
Like we're, we're, we're going through
Democrats and Republicans here, right?

2994
02:50:36,965 --> 02:50:37,055
Yes.

2995
02:50:37,055 --> 02:50:40,845
How the, the, the War on Terror
happens in the, in the George W.

2996
02:50:40,845 --> 02:50:43,835
Bush administration nine 11 and
the Patriot Act happens in the

2997
02:50:43,835 --> 02:50:45,005
George W. Bush administration.

2998
02:50:45,035 --> 02:50:47,975
This is a lot of incremental bad things.

2999
02:50:48,065 --> 02:50:48,395
Yes.

3000
02:50:48,545 --> 02:50:52,055
Under presidents of both parties,
under congresses of both parties.

3001
02:50:52,295 --> 02:50:53,285
How did this happen?

3002
02:50:53,699 --> 02:50:58,530
I mean, the, the simplest form of it
is just that nobody, and certainly

3003
02:50:58,530 --> 02:51:01,440
no president, wants to be president
during the time when there's a

3004
02:51:01,440 --> 02:51:05,550
big terrorist attack because that,
that makes them look bad, right?

3005
02:51:05,550 --> 02:51:08,460
I mean, obviously they also
wanna protect Americans, right?

3006
02:51:08,460 --> 02:51:10,410
That's part of, part
of their job, I guess.

3007
02:51:10,770 --> 02:51:15,990
And so, if you have an
intelligence community that is basically

3008
02:51:15,990 --> 02:51:19,650
operating a darkness, because that's
what intelligence communities do, and

3009
02:51:19,650 --> 02:51:22,890
they keep coming to you and saying, you
know, Hey, if we could just get

3010
02:51:22,890 --> 02:51:26,610
access to this information, it'd be really
helpful in preventing a terrorist attack.

3011
02:51:26,880 --> 02:51:29,940
And there may be cases where that's
true, that the intelligence community

3012
02:51:30,000 --> 02:51:34,650
is able to use this information in a
way that works well, but we also are

3013
02:51:34,650 --> 02:51:39,900
in theory, a society of laws and a
constitution that we're supposed to obey.

3014
02:51:40,410 --> 02:51:45,660
But that allowed for the fact that, you
know, administration after administration,

3015
02:51:45,660 --> 02:51:50,190
again, Republican and Democrat, had
lawyers who were very clever and who

3016
02:51:50,190 --> 02:51:54,660
would look through, well, if
we sort of position this this way, or

3017
02:51:54,660 --> 02:51:59,910
we state this this way, or we interpret
this, that way we can get what we want

3018
02:52:00,270 --> 02:52:04,590
and not technically break the law or not
technically violate the Fourth Amendment.

3019
02:52:04,590 --> 02:52:08,460
The assumption was always like, well,
we can sort of bend the law or bend

3020
02:52:08,460 --> 02:52:12,390
our interpretation of the law, and
nobody's really ever going to see

3021
02:52:12,390 --> 02:52:17,039
this, or nobody who cares,
is really ever going to see this,

3022
02:52:17,039 --> 02:52:18,539
and therefore we'll get away with it.

3023
02:52:30,750 --> 02:52:33,330
So now we get to the retaliation piece.

3024
02:52:33,599 --> 02:52:40,349
If retaliation isn't too strong a word,
how did the Pentagon react to,

3025
02:52:40,349 --> 02:52:46,170
Amodei and his company and Anthropic
wanting these two ethics clause clauses

3026
02:52:46,170 --> 02:52:47,789
in their contract with the Pentagon?

3027
02:52:48,509 --> 02:52:52,679
Well, this is the really striking part
of, of the story in on, in terms of

3028
02:52:52,679 --> 02:52:54,570
how the administration is, is acting.

3029
02:52:54,779 --> 02:52:57,210
You would think that they
would say, you know what?

3030
02:52:57,599 --> 02:52:59,820
Anthropic, you've got these red lines.

3031
02:52:59,910 --> 02:53:01,140
We can't accept these.

3032
02:53:01,349 --> 02:53:06,119
We're gonna go to another
company and use their technology.

3033
02:53:06,330 --> 02:53:11,369
We'll have a transition period, where, we'll phase out

3034
02:53:11,369 --> 02:53:17,009
your Claude AI model and do something
else, and we'll go our separate ways.

3035
02:53:17,160 --> 02:53:21,900
Instead, what the Pentagon did and
the administration has embraced, is

3036
02:53:21,900 --> 02:53:26,400
saying that we're gonna label you
what's called a supply chain risk.

3037
02:53:26,490 --> 02:53:30,570
And this is something used against
typically foreign companies, country

3038
02:53:30,750 --> 02:53:32,429
countries, which is seen as our enemies.

3039
02:53:32,429 --> 02:53:32,490
Yeah.

3040
02:53:32,880 --> 02:53:35,550
And they've gone even farther.

3041
02:53:35,820 --> 02:53:40,470
And said that any company that
does business with Anthropic, the

3042
02:53:40,470 --> 02:53:42,119
government will not do business with.

3043
02:53:42,210 --> 02:53:44,939
And that's a potentially
crippling blow to a corporation.

3044
02:53:45,539 --> 02:53:46,560
Are they going through with it?

3045
02:53:47,490 --> 02:53:52,410
Well, we've seen several agencies,
treasury House of Human Services say that

3046
02:53:52,439 --> 02:53:56,609
we're not gonna use Anthropic and we're,
and we're not gonna do business with other

3047
02:53:56,609 --> 02:54:01,769
companies that, that, that do this Now, so far Donald Trump has written this.

3048
02:54:01,890 --> 02:54:02,849
It's only in the tweet.

3049
02:54:03,119 --> 02:54:07,320
They actually haven't formally invoked
this, but the agencies are saying they're

3050
02:54:07,320 --> 02:54:11,130
doing it anyway, and Anthropic has
indicated they're gonna sue to fight that.

3051
02:54:12,449 --> 02:54:16,920
The Financial Times reports that as of
this morning, talks between Anthropic

3052
02:54:16,920 --> 02:54:18,900
and the Pentagon have quietly reopened.

3053
02:54:18,900 --> 02:54:19,410
Is there.

3054
02:54:19,830 --> 02:54:24,960
A version of this that ends well from an
ethical standpoint, or does it seem like

3055
02:54:25,320 --> 02:54:30,269
Amodei will need to relinquish his ideals
for the sake of his company's profits?

3056
02:54:31,199 --> 02:54:31,320
Well,

3057
02:54:31,320 --> 02:54:35,130
or even, or even e existence if
the retaliation is too strong.

3058
02:54:35,640 --> 02:54:35,789
Yeah.

3059
02:54:35,789 --> 02:54:40,560
Amodei, he, he, he put out an internal
memo that, was very strong.

3060
02:54:40,560 --> 02:54:43,470
He said that the reason why
they're being punished is because

3061
02:54:43,470 --> 02:54:45,960
they don't cater to Donald Trump.

3062
02:54:45,990 --> 02:54:50,580
They don't do what he called
dictator like,

3063
02:54:51,000 --> 02:54:54,720
Obedience to, to Trump.

3064
02:54:54,900 --> 02:54:59,099
They don't give money like the, one of the top executives

3065
02:54:59,099 --> 02:55:01,140
of OpenAI gave to a Trump pack.

3066
02:55:01,410 --> 02:55:03,779
And he said, that's
what's behind all this.

3067
02:55:03,990 --> 02:55:07,140
So it's a little surprising that
they're going back to the table now.

3068
02:55:07,349 --> 02:55:11,130
But I suspect if that's happening,
it's because the Pentagon's realizing

3069
02:55:11,130 --> 02:55:15,269
it's maybe tougher than they thought
to disengage Claude from their

3070
02:55:15,269 --> 02:55:17,460
activities, especially during a war.

3071
02:55:17,580 --> 02:55:19,140
And they need Anthropic's help.

3072
02:55:19,410 --> 02:55:25,529
But to your larger question, there
actually isn't a, a way to,

3073
02:55:25,650 --> 02:55:28,170
spin this as something we're all as happy.

3074
02:55:28,320 --> 02:55:32,310
A couple years ago, we were in a
position where all the AI companies,

3075
02:55:32,460 --> 02:55:36,119
most of the people in the government
and Congress were saying, you know what?

3076
02:55:36,300 --> 02:55:42,119
We have to come across with some sort of
regulation, some sort of plan to figure

3077
02:55:42,119 --> 02:55:44,310
out how we're going to deal with AI's.

3078
02:55:44,609 --> 02:55:46,140
Very scary powers.

3079
02:55:46,380 --> 02:55:49,199
There's great opportunities,
but also very scary powers.

3080
02:55:49,439 --> 02:55:50,939
They talked about regulation.

3081
02:55:51,060 --> 02:55:55,140
They talked about international
bodies that might eliminate ai,

3082
02:55:55,140 --> 02:55:57,179
AI from certain kinds of warfare.

3083
02:55:57,359 --> 02:55:58,349
That's all gone now.

3084
02:55:58,590 --> 02:56:01,590
We're all about
competition with China.

3085
02:56:01,695 --> 02:56:05,429
In, in ai, the companies are all
about competition with each other.

3086
02:56:05,699 --> 02:56:10,199
So, I see a rather
bleak outcome from that.

3087
02:56:10,199 --> 02:56:12,480
This, flap is a symbol of,

3088
02:56:14,609 --> 02:56:16,260
I said in the intro that.

3089
02:56:16,860 --> 02:56:19,800
This standoff between AI and the Pentagon.

3090
02:56:20,130 --> 02:56:24,330
It's obviously specific to that company
and that branch of government, but

3091
02:56:24,330 --> 02:56:29,550
it might have overtones for the safe
deployment of AI in all kinds of sectors

3092
02:56:30,060 --> 02:56:32,880
of American life, world life and society.

3093
02:56:33,270 --> 02:56:34,110
Do you see that?

3094
02:56:34,710 --> 02:56:35,550
Absolutely.

3095
02:56:35,550 --> 02:56:40,830
I mean, I, I, I think, it is,
yeah, because I, I just said this is an

3096
02:56:40,830 --> 02:56:47,790
indicator that when it comes to what we
once thought very recently was,

3097
02:56:47,850 --> 02:56:54,780 an important effort to, go, internationally, even to figure

3098
02:56:54,780 --> 02:56:59,790
out how, how this technology could not be
used to the detriment of human humankind.

3099
02:57:00,120 --> 02:57:02,160
That's gone by the wayside.

3100
02:57:02,280 --> 02:57:06,810
And now it's, accelerate,
accelerate, accelerate until,

3101
02:57:06,810 --> 02:57:10,350 AI is, beyond
the point, we can control it.

3102
02:57:16,980 --> 02:57:19,350
And finally, section D corporate power.

3103
02:57:24,120 --> 02:57:27,030
I'm wondering why, you can tell me
why the left hates technology so much.

3104
02:57:27,660 --> 02:57:31,950
Well, it's, have you ever had a, a, a
personal consumer grade printer before?

3105
02:57:33,005 --> 02:57:39,450
That's, that's my recent, oh, my
brother, brother, the manufacturer of

3106
02:57:39,450 --> 02:57:41,100
the only good consumer grade printer.

3107
02:57:41,220 --> 02:57:47,790
No, it's, it's, there has been,
I think in the recent round as Bryan

3108
02:57:47,820 --> 02:57:50,610
Befriend of, both of us, Bryan
Merchant has pointed out in the recent

3109
02:57:50,610 --> 02:57:57,720
round of fundraising for Anthropic,
there has been a sort of meme that has

3110
02:57:57,990 --> 02:58:04,410
proliferated throughout the internet that, opposition to or support for AI can

3111
02:58:04,410 --> 02:58:06,060
be something you can predicate along.

3112
02:58:06,060 --> 02:58:06,900
Left right.

3113
02:58:07,275 --> 02:58:09,615
Grounds and baseline.

3114
02:58:09,615 --> 02:58:13,365
That is not an argument I agree with
because I see people of all stripes,

3115
02:58:13,365 --> 02:58:17,324
even people who politically are not on my
side at all say that they're grossed out

3116
02:58:17,324 --> 02:58:22,755
and disgusted by generative AI and the
things that it's creates and also because

3117
02:58:22,755 --> 02:58:27,135
like there are dumb leftists that also
like this stuff, like it's that tweet.

3118
02:58:27,195 --> 02:58:28,335
I do not support a woman.

3119
02:58:28,335 --> 02:58:29,685
Some of you bitches are very dumb.

3120
02:58:29,775 --> 02:58:34,395
It's true of people across the political
spectrum, but also it's something I don't

3121
02:58:34,395 --> 02:58:40,365
believe because there is technology out
there that I do enjoy and like using and

3122
02:58:40,365 --> 02:58:43,305
think could be a net benefit for society.

3123
02:58:43,725 --> 02:58:50,595
It's just that when you, when people start
talking about technology as a discreet

3124
02:58:50,595 --> 02:58:56,925
object, frequently what they mean is the
most extractive and most repressive and

3125
02:58:56,925 --> 02:58:59,324
oppressive technology you could have.

3126
02:58:59,625 --> 02:59:04,185
In this case, we are referring to
generative AI and AI assistance in

3127
02:59:04,185 --> 02:59:10,275
companies like Anthropic and Open ai, who
are all, I think I can say on this show.

3128
02:59:10,275 --> 02:59:11,744
I think all those companies are evil.

3129
02:59:12,450 --> 02:59:13,830
Absolutely a hundred percent.

3130
02:59:13,890 --> 02:59:17,789
I just recently wrote as much about
open ai, but we, we feel the,

3131
02:59:17,789 --> 02:59:19,410
the same way about the rest of them.

3132
02:59:19,770 --> 02:59:20,730
It's fascinating to me.

3133
02:59:20,730 --> 02:59:24,240
Like there's many things like a pick
up on, in, in what you just said and,

3134
02:59:24,240 --> 02:59:28,230
We, we will get to a lot of
it, but even the term technology and

3135
02:59:28,289 --> 02:59:33,150
how it is like defined, like for a long
time, if you talked about technology,

3136
02:59:33,480 --> 02:59:37,289
you basically meant like internet
technology, digital technology, right?

3137
02:59:37,560 --> 02:59:40,320
Any other form of technology
was not considered technology.

3138
02:59:40,320 --> 02:59:43,980
It did not like fit under the term
of what we're talking about when we

3139
02:59:43,980 --> 02:59:47,250
generally talk about technology because
this is what the industry is focused on.

3140
02:59:47,250 --> 02:59:48,570
This is what the industry is pushing.

3141
02:59:48,660 --> 02:59:52,440
And now it feels like, especially in
these recent kind of like discussions,

3142
02:59:52,770 --> 02:59:58,170
the kind of scope of that term is
narrowed even further around generative

3143
02:59:58,170 --> 03:00:01,680
AI and like what is being pushed in
this moment and like, it's such a

3144
03:00:01,680 --> 03:00:05,500
reductive way to talk about technology
and, and what is going on, right.

3145
03:00:06,050 --> 03:00:08,720
Yeah, let's expand the
definition of what technology mean.

3146
03:00:08,720 --> 03:00:09,710
What does innovation mean?

3147
03:00:09,740 --> 03:00:13,550
I think mRNA vaccines are an
incredible piece of technology.

3148
03:00:13,700 --> 03:00:13,850
Love 'em.

3149
03:00:13,850 --> 03:00:14,360
Huge.

3150
03:00:14,600 --> 03:00:15,440
A huge innovation.

3151
03:00:15,440 --> 03:00:15,710
I have a

3152
03:00:15,710 --> 03:00:15,980
bunch

3153
03:00:15,980 --> 03:00:17,455
of me, yeah.

3154
03:00:17,840 --> 03:00:18,800
I wanna get jabbed.

3155
03:00:19,040 --> 03:00:20,360
Let me get jabbed even more.

3156
03:00:20,360 --> 03:00:25,430
I think high speed rail,
the kind that is not being developed.

3157
03:00:25,430 --> 03:00:25,670
Boom.

3158
03:00:25,910 --> 03:00:27,590
Big fans of trains on this podcast.

3159
03:00:27,650 --> 03:00:27,800
Yeah,

3160
03:00:27,950 --> 03:00:29,360
I love trains.

3161
03:00:29,415 --> 03:00:30,110
Absolutely love that.

3162
03:00:30,110 --> 03:00:33,860
Oh my God, if I could get to Chicago
from New York by train, if I could go

3163
03:00:33,860 --> 03:00:36,680
to Toronto by train, I would be golden.

3164
03:00:36,800 --> 03:00:38,030
I would be so happy.

3165
03:00:38,210 --> 03:00:40,640
How is there not a high speed
train between Toronto and

3166
03:00:40,640 --> 03:00:41,960
Montreal, like this baffle?

3167
03:00:41,960 --> 03:00:41,990
Oh

3168
03:00:42,020 --> 03:00:42,320
God.

3169
03:00:42,380 --> 03:00:42,680
3170
03:00:42,680 --> 03:00:44,360
that doesn't make any sense.

3171
03:00:44,450 --> 03:00:44,720
No.

3172
03:00:44,725 --> 03:00:44,935
At all.

3173
03:00:44,935 --> 03:00:49,700
Like it, that's a perfect corridor for
high speed train and high speed transit.

3174
03:00:50,030 --> 03:00:52,700
There's also little things that
I've been using in my life.

3175
03:00:52,700 --> 03:00:55,220
I asked, I'm sure you're aware
of Chris person from Aftermath.

3176
03:00:55,250 --> 03:00:55,490
Heard of

3177
03:00:55,490 --> 03:00:55,550
him.

3178
03:00:55,550 --> 03:00:56,570
I've, yeah, I know.

3179
03:00:56,570 --> 03:00:56,960
Recent guest

3180
03:00:56,960 --> 03:00:57,260
on the show.

3181
03:00:57,260 --> 03:00:57,320
Yeah.

3182
03:00:58,280 --> 03:01:00,350
A totally normal, regular person.

3183
03:01:00,355 --> 03:01:01,610
Chris person.

3184
03:01:01,880 --> 03:01:03,800
We're recording this
episode on his birthday.

3185
03:01:03,800 --> 03:01:04,550
Happy birthday, Chris.

3186
03:01:04,550 --> 03:01:05,030
Oh, happy birthday.

3187
03:01:05,030 --> 03:01:05,090
Yeah.

3188
03:01:06,315 --> 03:01:07,175
And, it'll

3189
03:01:07,175 --> 03:01:08,735
be belated by the time
it airs, but you know.

3190
03:01:08,765 --> 03:01:09,410
Yeah, of course.

3191
03:01:09,415 --> 03:01:09,735
But you know.

3192
03:01:10,289 --> 03:01:12,360
I think anytime someone says
it's your birthday, it should

3193
03:01:12,360 --> 03:01:13,170
count as your birthday.

3194
03:01:13,200 --> 03:01:14,610
You should just feel
like it's your birthday.

3195
03:01:14,615 --> 03:01:14,845
Totally.

3196
03:01:15,240 --> 03:01:15,420
Totally.

3197
03:01:15,420 --> 03:01:15,570
Because

3198
03:01:15,570 --> 03:01:16,050
you're allowed.

3199
03:01:16,140 --> 03:01:16,289
Yeah.

3200
03:01:16,295 --> 03:01:18,240
And we should also
celebrate half birthdays.

3201
03:01:18,240 --> 03:01:19,080
That is my thing.

3202
03:01:19,170 --> 03:01:19,380
I, yeah.

3203
03:01:19,620 --> 03:01:19,980
Yes.

3204
03:01:20,400 --> 03:01:22,500
You mentioned
that to me years ago.

3205
03:01:22,650 --> 03:01:22,710
Yeah.

3206
03:01:22,770 --> 03:01:25,289
And at the time I think I had
lower self-esteem, but now

3207
03:01:25,289 --> 03:01:26,640
I'm like, yeah, absolutely.

3208
03:01:27,600 --> 03:01:30,840
I think you should just have
a party whenever you want for

3209
03:01:30,840 --> 03:01:33,870
whatever reason, and everyone
should act like it's your birthday.

3210
03:01:33,960 --> 03:01:38,430
I think party fall is also a meaningful
piece of technology in my life.

3211
03:01:38,760 --> 03:01:42,600
There's little things and there's
big things like this, but the idea

3212
03:01:42,600 --> 03:01:46,260
that in order to be in support of
technology, you have to be in support

3213
03:01:46,260 --> 03:01:50,760
of the financial interest of Silicon
Valley is what really frustrates me.

3214
03:01:51,060 --> 03:01:55,950
Because when you say technology is
AI is generative ai, that is what you

3215
03:01:55,950 --> 03:02:00,600
essentially are saying, that there was a
vaccine that just lost funding and will

3216
03:02:00,600 --> 03:02:05,010
now have to be indefinitely shelved for
the herpes simplex virus, which would've

3217
03:02:05,010 --> 03:02:09,720
allowed people to be vaccinated against
any, derivative virus that is like

3218
03:02:09,750 --> 03:02:13,770
illness is caused by herpes, which
includes shingles, which I know a lot

3219
03:02:13,770 --> 03:02:16,920
of people of my generation have because
they were encouraged to get chickenpox by

3220
03:02:16,920 --> 03:02:18,870
their parents who didn't know any better.

3221
03:02:19,590 --> 03:02:21,150
That is technology.

3222
03:02:21,150 --> 03:02:26,789
I would so much rather our
government in our economy devote,

3223
03:02:26,789 --> 03:02:28,380
like, orient itself around.

3224
03:02:28,680 --> 03:02:31,800
Instead, we are in this place
where, as Chris probably

3225
03:02:31,800 --> 03:02:33,870
mentioned, RAMP is their food roof.

3226
03:02:34,050 --> 03:02:35,490
No one can buy a hard drive anymore.

3227
03:02:35,949 --> 03:02:39,430
You're not gonna get your fun
consumer, luxury electronics like

3228
03:02:39,430 --> 03:02:41,770
a new switch or a new PlayStation.

3229
03:02:42,039 --> 03:02:42,970
Your the internet worse.

3230
03:02:42,970 --> 03:02:43,180
Don't even

3231
03:02:43,180 --> 03:02:45,580
try to get that steam machine
that Chris is so excited

3232
03:02:45,580 --> 03:02:46,449
about, not even coming.

3233
03:02:46,449 --> 03:02:46,720
Oh God

3234
03:02:47,725 --> 03:02:48,225
so bad.

3235
03:02:48,445 --> 03:02:55,060
And it's not gonna happen now because the
entire world's economy is oriented around

3236
03:02:55,090 --> 03:03:00,310
the finances of two or three companies
that are not even making a product that

3237
03:03:00,340 --> 03:03:05,020
functions like it makes me, it is very
much an emperor's new clothes moment.

3238
03:03:05,020 --> 03:03:09,070
Like the Emperor Sam Elman is out
here nude as hell, and I feel like

3239
03:03:09,070 --> 03:03:10,900
I'm being forced to tolerate him.

3240
03:03:11,380 --> 03:03:14,800
It's wild to me too because like, I just
came across this study that was like

3241
03:03:14,890 --> 03:03:18,550
published last year by OpenAI itself,
where they like basically admitted that

3242
03:03:18,550 --> 03:03:20,199
hallucinations are part of the product.

3243
03:03:20,199 --> 03:03:21,580
Like they're not going to come out.

3244
03:03:21,580 --> 03:03:24,460
So it's like the whole notion was
these things are gonna get better,

3245
03:03:24,460 --> 03:03:27,640
we're gonna like get it out of the
product at some point, but here's

3246
03:03:27,640 --> 03:03:31,539
OpenAI, like the leader in the charge
for generative AI being like, yeah.

3247
03:03:31,539 --> 03:03:34,630
That thing where like, you
know, the technology that we make

3248
03:03:34,630 --> 03:03:37,630
that we say is so intelligent and
like basically human level smart.

3249
03:03:37,930 --> 03:03:38,140
Yeah.

3250
03:03:38,140 --> 03:03:41,230
It's just gonna, gonna make
things that are wrong all the

3251
03:03:41,230 --> 03:03:42,730
time because that's built in there

3252
03:03:42,760 --> 03:03:42,850
thick.

3253
03:03:43,360 --> 03:03:43,570
Yeah.

3254
03:03:43,660 --> 03:03:43,930
Great.

3255
03:03:44,350 --> 03:03:44,710
Yeah.

3256
03:03:44,710 --> 03:03:47,770
I feel like we keep
coming across these cases.

3257
03:03:47,800 --> 03:03:51,880
The blogs that I wrote that you
read and that you invited me on for

3258
03:03:51,880 --> 03:03:53,080
the left doesn't hate technology.

3259
03:03:53,080 --> 03:03:54,940
They hate being, we hate Being Exploited,

3260
03:03:55,060 --> 03:03:55,360
which is

3261
03:03:55,365 --> 03:03:55,505
in

3262
03:03:55,505 --> 03:03:56,440
the show notes for listeners.

3263
03:03:56,470 --> 03:03:56,860
Yeah.

3264
03:03:57,130 --> 03:04:02,500
It was re a response to a, a really
dumb blog by some random person that

3265
03:04:02,500 --> 03:04:04,060
I just kept seeing on the timeline.

3266
03:04:04,240 --> 03:04:04,360
Yeah.

3267
03:04:04,360 --> 03:04:07,119
And you know what, sometimes
you read something and it makes

3268
03:04:07,119 --> 03:04:10,240
you so angry that you write 1500
words in about half an hour.

3269
03:04:10,539 --> 03:04:16,780
And that is what happened to me, where
the insistence that the potential

3270
03:04:16,780 --> 03:04:22,900
for good that this technology could
maybe possibly do in the future is the

3271
03:04:22,900 --> 03:04:26,470
reason why we have to invest in it now.

3272
03:04:26,470 --> 03:04:30,970
And we were foolish if you're a leftist
for not investing in it, that is just.

3273
03:04:32,090 --> 03:04:36,470
Speculative at best, like that is
something you could speculate about

3274
03:04:36,470 --> 03:04:42,170
the abilities for AI to cure cancer,
that there's a, a speculative case you

3275
03:04:42,170 --> 03:04:44,180
could possibly make in the present.

3276
03:04:44,180 --> 03:04:49,160
Right now, AI has caused a school shooting
where the casualties were children.

3277
03:04:49,490 --> 03:04:54,529
There were like 8, 8, 12 year olds died,
where multiple children, again, have

3278
03:04:54,529 --> 03:04:57,020
taken their own lives because of this.

3279
03:04:57,020 --> 03:05:02,359
It's, I used to say that Q Anon really
scared me 'cause I'd never seen something

3280
03:05:02,359 --> 03:05:06,050
drive people to psychosis like with
no previous history of psychosis.

3281
03:05:06,289 --> 03:05:11,779
So quickly, AI drives those same
kinds of people even faster.

3282
03:05:12,140 --> 03:05:17,090
And it's like you get your own personal
Q anon when you really get deep into it.

3283
03:05:17,330 --> 03:05:21,410
And there's examples and examples and
examples of it not even being able to

3284
03:05:21,410 --> 03:05:23,420
do its core functionality correctly.

3285
03:05:23,720 --> 03:05:27,830
I know that OpenAI has said it's patch,
the Rs and strawberries thing, but I

3286
03:05:27,830 --> 03:05:33,050
saw a video video of someone a month
ago asking OpenAI, asking ChatGPT how

3287
03:05:33,050 --> 03:05:36,800
many Rs there are in strawberry and it
not being able to answer it correctly.

3288
03:05:37,160 --> 03:05:41,000
So I you have to be, the word I
keep coming back to is credulous.

3289
03:05:41,240 --> 03:05:44,779
You have to be a credulous
buffoon in order to take these

3290
03:05:44,779 --> 03:05:49,490
things and then project outward
optimistically, I, I believe.

3291
03:05:49,845 --> 03:05:54,375
The, the idea of fully
automated luxury communism is

3292
03:05:54,375 --> 03:05:57,765
something that, something that I
really wished would come to pass.

3293
03:05:58,245 --> 03:06:02,205
But the more we can see the way that
this technology interacts with capital

3294
03:06:02,325 --> 03:06:06,555
interacts with corrupt governance, the
more it becomes clear that it's just.

3295
03:06:06,945 --> 03:06:10,605
Corporations milking the government
for all the money they can get

3296
03:06:10,875 --> 03:06:14,565
and then not delivering a product
as they have done previously.

3297
03:06:14,744 --> 03:06:15,375
In the last 10,

3298
03:06:27,930 --> 03:06:30,195
in the fallout of the Quick
Department of Defense deal with

3299
03:06:30,195 --> 03:06:34,545
OpenAI, protestors showed up outside
OpenAI headquarters and users began

3300
03:06:34,545 --> 03:06:36,435
deleting chat GPT from their phones.

3301
03:06:36,465 --> 03:06:39,855
And then Claude saw a massive uptick
in downloads and subscriptions,

3302
03:06:39,945 --> 03:06:43,335
surpassing chat GPT for the first time
in the app store to become the number

3303
03:06:43,335 --> 03:06:45,404
one app across all phone interfaces.

3304
03:06:45,645 --> 03:06:49,485
This has, of course, sent OpenAI
scrambling to try to fix the PR disaster

3305
03:06:49,485 --> 03:06:52,875
of this Department of Defense contract,
and I think it's a powerful indicator of

3306
03:06:52,875 --> 03:06:54,945
the current public sentiment around ai.

3307
03:06:54,945 --> 03:06:56,895
Generally, people are curious.

3308
03:06:56,895 --> 03:06:58,244
Many people are using it.

3309
03:06:58,244 --> 03:07:01,875
Some use it daily, either because their
corporate jobs make them or because

3310
03:07:01,875 --> 03:07:03,465
they find the chat tools helpful.

3311
03:07:03,525 --> 03:07:06,945
Some people have developed a concern
in codependency on these apps using

3312
03:07:06,945 --> 03:07:10,574
them instead of real human interaction
as a form of companionship in ways

3313
03:07:10,574 --> 03:07:13,574
that we're just beginning to see the
consequences of which include numerous.

3314
03:07:13,715 --> 03:07:16,625
AI encouraged suicides and self-harm.

3315
03:07:16,625 --> 03:07:20,615
And so people, even those who use the
technology, are rightfully concerned

3316
03:07:20,644 --> 03:07:24,664
about the safety of AI from the safety
of consumers to the potential for AI

3317
03:07:24,664 --> 03:07:29,465
to be used at mass scale to undermine
democratic values or take human life On

3318
03:07:29,465 --> 03:07:34,685
top of that, we're watching our lawmakers
do absolutely fucking nothing about it.

3319
03:07:34,865 --> 03:07:39,035
And so absent actual laws, reforms,
regulations, or any action at all

3320
03:07:39,154 --> 03:07:42,635
from the people we've put in charge of
regulating these things, we are turning

3321
03:07:42,635 --> 03:07:46,804
to the AI companies themselves for
assurances that they aren't up to no good.

3322
03:07:46,955 --> 03:07:50,705
And Anthropic, which was founded
in 2020 by former open AI employees

3323
03:07:50,705 --> 03:07:54,484
who were deeply concerned with open
AI's, recklessness, provides a lot of

3324
03:07:54,484 --> 03:07:56,404
assurances of its ethics on its website.

3325
03:07:56,404 --> 03:07:59,615
And through its founder, Dario
Amodei, and his many blog posts.

3326
03:07:59,705 --> 03:08:02,465
And now through its apparent
stance against the Pentagon, its

3327
03:08:02,465 --> 03:08:05,675
willingness to put its neck on the
line to honor its commitments while

3328
03:08:05,675 --> 03:08:08,795
OpenAI is happy to do whatever the
government wants or so it seems.

3329
03:08:08,914 --> 03:08:11,824
And the public is voting with its
attention, its dollars and its

3330
03:08:11,824 --> 03:08:15,875
data by deleting chat GPT in favor
of Anthropics Claude Chatbot.

3331
03:08:16,115 --> 03:08:18,304
But I think it's deeply
misplaced to put so much trust

3332
03:08:18,304 --> 03:08:20,135
in any of these AI companies.

3333
03:08:20,135 --> 03:08:21,664
And let me explain why I think this.

3334
03:08:21,845 --> 03:08:24,904
If there's one thing I've learned as I've
delved into the world of AI development,

3335
03:08:24,904 --> 03:08:28,445
it's that the people doing the developing
don't just see it as their job.

3336
03:08:28,474 --> 03:08:31,564
It makes up a part of
their entire worldview.

3337
03:08:31,685 --> 03:08:35,224
So to understand the AI world, which
is deeply important for everyday

3338
03:08:35,224 --> 03:08:39,035
consumers and everyday members of civil
society who wanna hold their lawmakers

3339
03:08:39,035 --> 03:08:43,265
accountable to understand the AI world,
you have to understand the AI worldview,

3340
03:08:43,595 --> 03:08:47,855
the book Empire of AI by Karen Howe,
is the best thing period that I've

3341
03:08:47,855 --> 03:08:49,085
read over probably the last year.

3342
03:08:49,085 --> 03:08:53,164
And it lays out the AI world in
a really accessible, readable

3343
03:08:53,164 --> 03:08:56,855
way that is a genuinely
entertaining, if horrifying read.

3344
03:08:56,945 --> 03:08:58,324
And I think everyone.

3345
03:08:58,460 --> 03:08:59,900
Should read this book genuinely.

3346
03:08:59,900 --> 03:09:01,340
I'm not affiliated in any way with it.

3347
03:09:01,340 --> 03:09:05,720
I just think everyone should read it as
a very brief and incomplete overview.

3348
03:09:05,960 --> 03:09:09,170
The key ideas that you need to know
to understand how AI fits into the

3349
03:09:09,170 --> 03:09:13,010
worldview of its creators includes
understanding the spectrum of AI

3350
03:09:13,010 --> 03:09:15,140
ethics that AI developers fall onto.

3351
03:09:15,290 --> 03:09:18,770
You have accelerationist on one end
people who think that artificial

3352
03:09:18,770 --> 03:09:21,950
intelligence is like the key to
human development and flourishing,

3353
03:09:21,950 --> 03:09:25,430
and that it should be developed as
quickly as possible unencumbered

3354
03:09:25,430 --> 03:09:27,170
by pesky government regulations.

3355
03:09:27,200 --> 03:09:29,900
And then on the other end of the
spectrum, you have the doomers who

3356
03:09:29,900 --> 03:09:33,530
believe that AI development will result
in the destruction of humanity, and

3357
03:09:33,530 --> 03:09:36,830
therefore its development must be
carefully restrained within that camp.

3358
03:09:36,860 --> 03:09:40,670
Many of its adherence are guided
by effective altruism or attempting

3359
03:09:40,670 --> 03:09:43,790
to do the most good with their
technological innovations.

3360
03:09:43,820 --> 03:09:47,510
However, taking a step back from all
of that, it's important to keep in

3361
03:09:47,510 --> 03:09:52,100
mind the kind of accepted ideas that
everyone on this spectrum, no matter

3362
03:09:52,100 --> 03:09:55,820
where they fall, has that, is that
whether you are an accelerationist

3363
03:09:55,850 --> 03:09:59,930
or a dor, you kind of have to be
convinced in general of the power of ai.

3364
03:10:00,200 --> 03:10:03,650
Many people falling anywhere
on the spectrum deeply believe

3365
03:10:03,650 --> 03:10:05,540
in the possibility of a GI.

3366
03:10:05,540 --> 03:10:07,310
Artificial general intelligence.

3367
03:10:07,490 --> 03:10:10,910
If you didn't genuinely believe that
AI is going to revolutionize the world,

3368
03:10:11,030 --> 03:10:14,060
then you wouldn't be so passionate
as to fall anywhere on the spectrum.

3369
03:10:14,270 --> 03:10:17,690
The problem is that no one can
actually define a GI, not specifically.

3370
03:10:17,840 --> 03:10:19,880
Most people just say, we'll
know it when we see it.

3371
03:10:19,910 --> 03:10:23,270
Or they give vague definitions
like it's a type of AI that is

3372
03:10:23,270 --> 03:10:25,070
sentient that can think for itself.

3373
03:10:25,100 --> 03:10:29,060
But large language models, the type
of AI that's taken over popularity

3374
03:10:29,060 --> 03:10:31,190
because there's more than one
type of artificial intelligence.

3375
03:10:31,190 --> 03:10:35,270
The LLM models are just guessing
the next word based on the

3376
03:10:35,270 --> 03:10:36,350
data it's been trained on.

3377
03:10:36,380 --> 03:10:39,170
That is not sentience no
matter how much data you feed.

3378
03:10:39,170 --> 03:10:39,350
It.

3379
03:10:39,590 --> 03:10:42,890
Also, the latest AI models have
already been trained on basically

3380
03:10:42,890 --> 03:10:47,150
all available human created language
that's been digitized in existence.

3381
03:10:47,150 --> 03:10:49,790
And I don't think anyone is
arguing that we've reached a GI.

3382
03:10:49,790 --> 03:10:51,290
So like where are they
gonna get that data?

3383
03:10:51,590 --> 03:10:54,710
So the generally accepted
hypothesis in the AI community.

3384
03:10:54,745 --> 03:10:58,945
That a GI is the clear outcome of all of
this, and therefore, we either must let it

3385
03:10:58,945 --> 03:11:03,175
flourish and it will benefit all mankind
or tamp it down with a serious guardrails,

3386
03:11:03,175 --> 03:11:05,065
or it will extinguish all mankind.

3387
03:11:05,245 --> 03:11:09,685
All of that rests on the overall belief
in the power of ai and that a GI is

3388
03:11:09,685 --> 03:11:13,945
possible, and that whole thesis is
unproven, which is I think, an important

3389
03:11:13,945 --> 03:11:17,395
high level point to understand because
much of the AI landscape we're dealing

3390
03:11:17,395 --> 03:11:21,625
with right now is the result of the
hubris and ego of a few mostly men.

3391
03:11:21,835 --> 03:11:25,405
And in practice, AI can definitely
do impressive things, but the

3392
03:11:25,405 --> 03:11:29,065
level of false advertising that the
company's trying to sell us on, the

3393
03:11:29,065 --> 03:11:33,385
AI version of whatever products they
offer is through the fucking roof.

3394
03:11:33,385 --> 03:11:38,035
For example, last November, the Verge did
an experiment where it asked Microsoft

3395
03:11:38,035 --> 03:11:43,285
copilot to do the exact things Microsoft
claimed it could do in the copilot

3396
03:11:43,285 --> 03:11:45,715
commercials with embarrassing results.

3397
03:11:45,955 --> 03:11:48,085
Can you get me a link
to one that's in stock?

3398
03:11:48,565 --> 03:11:52,945
Here's the link to Best Buy,
link to Best buy.com/eleven.

3399
03:11:52,945 --> 03:11:55,375
Best Buy slash Sure website.

3400
03:11:55,725 --> 03:12:00,315
And it's foremost definitely in stock
and it's for most definitely in stock and

3401
03:12:00,315 --> 03:12:03,855
it's available for pickup or delivery,
and it's for most definitely in stock.

3402
03:12:04,635 --> 03:12:06,345
Okay, thanks Copilot.

3403
03:12:06,465 --> 03:12:10,665
So there is a vast assumption happening
based on a theory that the end result

3404
03:12:10,665 --> 03:12:13,485
of AI will mostly definitely be a GI.

3405
03:12:13,695 --> 03:12:15,195
And that is just a theory at this point.

3406
03:12:15,195 --> 03:12:18,195
And a lot of this worldview has
resulted in what Karen Howe describes

3407
03:12:18,195 --> 03:12:21,825
in her book, empire of AI as
basically a self-fulfilling prophecy.

3408
03:12:21,975 --> 03:12:27,195
Because for open ai, especially in its
fearmongering over the future of a GI, the

3409
03:12:27,195 --> 03:12:32,205
dangers of it and especially the dangers
of a GI, if China develops it first, which

3410
03:12:32,205 --> 03:12:34,305
is a favorite fearmongering tactics for.

3411
03:12:34,305 --> 03:12:35,890
Any new technology.

3412
03:12:35,980 --> 03:12:37,539
What if the comms get it first?

3413
03:12:37,660 --> 03:12:43,330
In all of this, fretting OpenAI itself
has accelerated the rate of AI growth

3414
03:12:43,449 --> 03:12:45,789
beyond anyone's wildest expectations.

3415
03:12:45,789 --> 03:12:46,840
Even a decade ago.

3416
03:12:46,990 --> 03:12:50,679
Its demands for data centers for high
powered chips for more and more and more

3417
03:12:50,679 --> 03:12:52,720
data to feed the AI training models.

3418
03:12:52,900 --> 03:12:57,160
All of it far surpasses anything
China was ever doing, but it has

3419
03:12:57,160 --> 03:13:01,000
shown the world what's possible
and how to do it and instigated an

3420
03:13:01,000 --> 03:13:04,509
arms race for the first company to
reach the promised land of a GI.

3421
03:13:04,750 --> 03:13:09,279
If not for open ai, especially under
Sam Altman, we would not have the AI

3422
03:13:09,279 --> 03:13:13,480
technology we have today and the move
fast break things mentality that he

3423
03:13:13,480 --> 03:13:17,109
brought to the AI world has meant that
for all the hand wringing over the

3424
03:13:17,109 --> 03:13:21,580
potential damage a GI could cause to
humanity, it's become a race to the

3425
03:13:21,580 --> 03:13:23,650
bottom when it comes to AI development.

3426
03:13:23,679 --> 03:13:28,359
Who can do it the fastest, no matter
the human cost spurred on by open AI

3427
03:13:28,449 --> 03:13:33,130
itself and all of this is over a GIA
technology that is just a theory.

3428
03:13:33,160 --> 03:13:35,500
Meanwhile, despite all that
advancement, we can't even get a

3429
03:13:35,500 --> 03:13:37,539
computer to do basic tasks on command.

3430
03:13:37,570 --> 03:13:41,529
AI has not delivered on the wide
scale productivity it has promised

3431
03:13:41,650 --> 03:13:44,679
it's not delivered on the profits,
it's promised it's not delivered on

3432
03:13:44,679 --> 03:13:46,660
the human flourishing it's promised.

3433
03:13:46,929 --> 03:13:49,869
All it has delivered so
far is shareholder value.

3434
03:13:49,990 --> 03:13:54,789
A lot of speculation and endless,
endless AI ads shoving this

3435
03:13:54,789 --> 03:13:56,590
technology down our throats.

3436
03:13:56,769 --> 03:13:57,849
Whether we like it or not.

3437
03:13:57,849 --> 03:14:01,815
I don't know if you watch the Super Bowl
literally every other ad. Was an AI ad on

3438
03:14:01,815 --> 03:14:06,795
top of the human costs of ai, including
suicides spurred on by chatbots, but also

3439
03:14:06,795 --> 03:14:11,985
the incredibly exploitative practices
used in data harvesting, stealing IP

3440
03:14:12,105 --> 03:14:15,765
and paying people in the poorest parts
of the globe, pennies to sift through

3441
03:14:15,765 --> 03:14:20,655
deeply traumatizing material to try to
help train this AI and the incredibly

3442
03:14:20,655 --> 03:14:22,845
high environmental cost of all of it.

3443
03:14:23,085 --> 03:14:26,565
That's why Karen Howe named
her book Empire of ai.

3444
03:14:26,715 --> 03:14:31,185
Her central thesis is that the AI boom is
similar in many ways to all the other past

3445
03:14:31,185 --> 03:14:35,385
technological improvements that mainly
led to increased wealth stratification

3446
03:14:35,535 --> 03:14:39,195
and depended on the exploited
labor of marginalized populations.

3447
03:14:39,315 --> 03:14:43,125
Whether you're talking about the
cotton gin or ChatGPT, this is why we

3448
03:14:43,125 --> 03:14:45,705
cannot pick winners in the AI race.

3449
03:14:45,705 --> 03:14:48,705
We can't look at these companies
in black and white and think

3450
03:14:48,735 --> 03:14:50,865
anthropic good, open AI bad.

3451
03:14:51,015 --> 03:14:54,885
We should be looking at all of it with
skepticism because the founders of these

3452
03:14:54,945 --> 03:14:58,965
companies, even ones like Anthropic with
a lot of lofty, ostensibly charitable

3453
03:14:58,965 --> 03:15:03,195
and good goals, are coming from the
same worldview that created the problems

3454
03:15:03,195 --> 03:15:04,785
of the world we currently exist in.

3455
03:15:18,465 --> 03:15:21,105
On that point, I was interested in
asking you because you brought it up

3456
03:15:21,105 --> 03:15:24,765
and you mentioned it in your piece,
but I was also a big techno optimist

3457
03:15:24,945 --> 03:15:28,485
for a while, and I know that you've
written about being one as well.

3458
03:15:28,815 --> 03:15:32,565
I was wondering if you could talk to me
about your journey from being a techno

3459
03:15:32,565 --> 03:15:36,315
optimist, being very optimistic about
what this technology was going to deliver

3460
03:15:36,435 --> 03:15:38,865
to us and and shifting away from that.

3461
03:15:39,585 --> 03:15:42,165
I'm so glad you asked me that because
I feel like in the piece, it's not

3462
03:15:42,165 --> 03:15:45,705
the point of it and I, I didn't get
a chance to fully explain my journey.

3463
03:15:45,975 --> 03:15:49,485
So my parents are both like old
hippies, so I always like the.

3464
03:15:49,794 --> 03:15:53,695
The household was always going
to be a left wing household.

3465
03:15:53,724 --> 03:15:57,025
My parents didn't want to indoctrinate
us, but they allowed a lot of

3466
03:15:57,025 --> 03:16:00,505
space for us to ask questions like,
why does everybody need money?

3467
03:16:00,744 --> 03:16:02,695
Why do, why are some people poor?

3468
03:16:02,814 --> 03:16:03,925
And stuff like that.

3469
03:16:03,925 --> 03:16:08,365
They were generally in support of me and
my brother becoming politically active.

3470
03:16:08,394 --> 03:16:12,085
And especially for me, like during,
I mean, none of us knew what was

3471
03:16:12,085 --> 03:16:14,664
gonna happen with Obama, and that
was my freshman year of college.

3472
03:16:14,664 --> 03:16:18,324
And they had had people from the
campaign come out to Oberlin College

3473
03:16:18,595 --> 03:16:20,875
and put us on buses and say, let's go.

3474
03:16:20,875 --> 03:16:23,214
You're gonna go canvas in this
neighborhood and let's go do it.

3475
03:16:23,425 --> 03:16:27,294
And I, I participated in that, it was impossible not

3476
03:16:27,294 --> 03:16:30,054
to get swept up in that emotion.

3477
03:16:30,054 --> 03:16:34,585
And my parents who are more left wing than
that, they also wanted a black president.

3478
03:16:34,585 --> 03:16:34,824
Right?

3479
03:16:34,824 --> 03:16:37,794
Even if he didn't align a
hundred percent on their views and

3480
03:16:37,794 --> 03:16:42,775
their values, it is like, my dad
grew up, in the Jim Crow South.

3481
03:16:42,865 --> 03:16:45,925
Like I called him
crying when it happened.

3482
03:16:46,015 --> 03:16:51,625
It is a, it was a moment that
feels emotionally resonant, even though

3483
03:16:51,625 --> 03:16:53,724
I now hate Barack Obama with my life.

3484
03:16:54,835 --> 03:16:59,574
But at the time, like adding on
to that, my dad's career is in it.

3485
03:17:00,115 --> 03:17:04,855
So my mother got a, has a tenureship
at a college and he, up until a few

3486
03:17:04,855 --> 03:17:08,755
years ago, before when he retired,
worked in it as a PC specialist.

3487
03:17:08,994 --> 03:17:15,894
So for me and for him, hand in hand
with our liberation from a oppressive

3488
03:17:15,894 --> 03:17:19,794
government was an excitement
for the communication networks

3489
03:17:19,794 --> 03:17:21,324
that the internet could provide.

3490
03:17:21,625 --> 03:17:28,404
We got a modem before anyone else I knew,
and I was just a sponge for information.

3491
03:17:28,404 --> 03:17:31,914
The idea that I could talk to people
from other countries who live totally

3492
03:17:31,914 --> 03:17:36,265
different lives for me, I completely
ate all that up and I, I made friends

3493
03:17:36,265 --> 03:17:40,945
as an awkward teenager kind of for the
first time with the aid of the internet.

3494
03:17:41,595 --> 03:17:48,045
And it introduced me to more joy,
more discovery, more books, more

3495
03:17:48,045 --> 03:17:50,475
movies, more kinds of people.

3496
03:17:50,895 --> 03:17:54,975
I was deep in the fucking like
internet piracy game too, so I

3497
03:17:54,975 --> 03:18:00,525
was voraciously like downloading
and more information, more art.

3498
03:18:00,795 --> 03:18:06,440
To me, that is what technology meant was
just a window into, I downloaded a tour

3499
03:18:06,525 --> 03:18:09,585
of Stranger Things when I was in college
because it wasn't available anywhere.

3500
03:18:09,705 --> 03:18:13,185
And so I felt like I was doing some
kind of magic spell that allowed

3501
03:18:13,185 --> 03:18:16,005
me to have access to culture
that had otherwise disappeared.

3502
03:18:16,335 --> 03:18:21,405
I remember, being like in Eastern
Europe and downloading a torrent of

3503
03:18:21,405 --> 03:18:25,375
House of Cards to watch like the season
or second season or something like that,

3504
03:18:25,375 --> 03:18:26,520
that my God, like this was That's amazing.

3505
03:18:26,520 --> 03:18:28,065
This back in 2013 or something.

3506
03:18:28,065 --> 03:18:28,305
Yeah.

3507
03:18:28,455 --> 03:18:30,015
God, I meant strange days.

3508
03:18:30,015 --> 03:18:31,065
Not stranger things, but it's fine.

3509
03:18:31,245 --> 03:18:33,045
Katherine Bigelow.

3510
03:18:33,045 --> 03:18:34,425
Way better than those two weirdos.

3511
03:18:34,425 --> 03:18:35,415
Netflix, but yeah.

3512
03:18:35,445 --> 03:18:41,175
I remember freshman year of college I
was downloading fan subs of the anime,

3513
03:18:41,175 --> 03:18:47,625
MOA magica in my dorm room every Sunday
hungover before I went to the dining hall.

3514
03:18:47,865 --> 03:18:52,515
And I was just like, this to me,
like I, and what was so cute is the

3515
03:18:52,515 --> 03:18:57,945
fan subs like left in one commercial
every episode because they thought

3516
03:18:57,945 --> 03:18:59,265
the commercial was so funny.

3517
03:18:59,265 --> 03:19:04,035
It was for a drink called Morning Rescue
and it would show like a sad office worker

3518
03:19:04,155 --> 03:19:10,245
exhausted and then rest, like emergency, like 9 1 1 firefighter people going

3519
03:19:10,245 --> 03:19:12,435
into the window singing morning rescue.

3520
03:19:12,435 --> 03:19:14,745
It was, it was just cute.

3521
03:19:14,745 --> 03:19:16,305
Like people did that, right?

3522
03:19:16,305 --> 03:19:16,365
Yeah.

3523
03:19:16,365 --> 03:19:19,755
Like a human being made that choice,
I felt connected to those people.

3524
03:19:19,845 --> 03:19:23,745
I felt like I was participating in a
special community by watching that show.

3525
03:19:23,865 --> 03:19:25,725
The internet is what made it possible.

3526
03:19:25,845 --> 03:19:27,645
And it was like a little
transgressive too.

3527
03:19:27,645 --> 03:19:29,835
I'm stealing, Ooh, I'm a Biden pirate.

3528
03:19:30,345 --> 03:19:33,915
By the time I graduated from
college though, and it had become

3529
03:19:33,915 --> 03:19:36,165
clear, the major corporations.

3530
03:19:36,495 --> 03:19:42,165
Understood what that optimism was
and wanted this workforce that was

3531
03:19:42,165 --> 03:19:48,705
now really super jobless and around
to buy in hook line and sinker to

3532
03:19:48,705 --> 03:19:50,535
what they were offering as employees.

3533
03:19:50,625 --> 03:19:55,515
So you got all these tech companies
saying, we're gonna change the world.

3534
03:19:55,545 --> 03:20:02,865
And I, I did want to believe that
companies like Uber and Lyft and Airbnb

3535
03:20:03,135 --> 03:20:09,255
and WeWork could affect the market
in ways that made things more fair.

3536
03:20:09,735 --> 03:20:14,475
But as my cynicism with Obama
grew, so did my cynicism with what

3537
03:20:14,475 --> 03:20:16,095
these corporations were saying.

3538
03:20:16,395 --> 03:20:21,015
I think it was around the time Lyft
was offering loans for, it was Uber

3539
03:20:21,015 --> 03:20:23,325
offering loans for drivers to buy cars.

3540
03:20:23,685 --> 03:20:27,765
Where I was like, okay, so
now you, you're, that just

3541
03:20:27,765 --> 03:20:28,995
sounds like script to me.

3542
03:20:28,995 --> 03:20:32,295
That sounds like you're, but you're
buying, so at the company store, you're

3543
03:20:32,355 --> 03:20:37,725
relying on this company to buy a car for
you on a loan that they finance for you,

3544
03:20:38,085 --> 03:20:42,975
and you have to pay that loan back with
the wages you earn from driving that car.

3545
03:20:43,125 --> 03:20:44,265
Tell me how that makes sense.

3546
03:20:44,295 --> 03:20:45,005
And people were frustrat me.

3547
03:20:45,005 --> 03:20:45,400
And if I remember like

3548
03:20:45,675 --> 03:20:47,535
the interest rates were like ridiculous.

3549
03:20:47,625 --> 03:20:47,985
Yeah.

3550
03:20:48,045 --> 03:20:51,765
It, if I remember correctly as well,
I remember like arguing with someone

3551
03:20:51,765 --> 03:20:56,295
my own age about this and being like,
yes, it's good that these people have

3552
03:20:56,295 --> 03:21:00,165
access to transportation, but you have
to understand like that, that now it

3553
03:21:00,165 --> 03:21:01,755
makes Lyft in charge of their whole life.

3554
03:21:01,755 --> 03:21:03,615
They might as well move to a company town.

3555
03:21:03,945 --> 03:21:07,005
And people were like, no, but they're,
they're providing things for you.

3556
03:21:07,005 --> 03:21:07,545
You blah, blah blah.

3557
03:21:07,725 --> 03:21:08,955
Taxis are bad.

3558
03:21:08,955 --> 03:21:11,355
And in some places it's
really difficult to get one.

3559
03:21:11,355 --> 03:21:12,825
And it's like, that is true.

3560
03:21:13,035 --> 03:21:16,125
And I do think some things about the
hotel and taxi lobby needed to be

3561
03:21:16,125 --> 03:21:19,965
broken up, but it doesn't mean just
handing control to another corporation.

3562
03:21:20,730 --> 03:21:23,669
There needs to be a diversified
market if we wanna still

3563
03:21:23,669 --> 03:21:25,230
believe that capitalism works.

3564
03:21:25,439 --> 03:21:29,189
And this is when I started really
doubting that capitalism worked.

3565
03:21:29,400 --> 03:21:33,720
I think it was becoming so invested
in Silicon Valley and all the things

3566
03:21:33,720 --> 03:21:35,310
they said they could do for us.

3567
03:21:35,580 --> 03:21:39,734
And then seeing how they just consolidated
money and power for themselves, where

3568
03:21:39,734 --> 03:21:42,419
I was really just, I became a cynic.

3569
03:21:42,750 --> 03:21:46,320
I became pessimistic
about all of these things.

3570
03:21:46,380 --> 03:21:49,859
And I became angry because they lied.

3571
03:21:49,859 --> 03:21:54,839
And there was no consequence That was
like the scales had followed from my eyes.

3572
03:21:54,839 --> 03:22:00,634
I can no longer take anything Tech
corporation says in good faith because

3573
03:22:00,779 --> 03:22:04,529
I look at the humble beginnings of
something like Amazon as a bookstore,

3574
03:22:04,710 --> 03:22:08,910
and I see Jeff Bezos renting out
of town for his stupid wedding now.

3575
03:22:09,089 --> 03:22:11,279
And like that's what they all wanna do.

3576
03:22:11,369 --> 03:22:13,740
Line go up until you're the
richest man in the world.

3577
03:22:14,640 --> 03:22:15,089
Totally.

3578
03:22:15,269 --> 03:22:19,830
And it's fascinating to me to hear
your story because I feel like it

3579
03:22:20,130 --> 03:22:23,310
mirrors so closely my own as well.

3580
03:22:23,640 --> 03:22:25,410
Obviously I love the early internet too.

3581
03:22:25,619 --> 03:22:27,330
I love going onto all these forums.

3582
03:22:27,359 --> 03:22:31,980
I love being able to rip music and
Get movies and shows and

3583
03:22:31,980 --> 03:22:33,449
like all this kind of stuff, right?

3584
03:22:33,720 --> 03:22:38,460
And there was the piece of like the 2010s
and seeing the discourse around AI and

3585
03:22:38,460 --> 03:22:40,439
stuff that, that really radicalized me.

3586
03:22:40,439 --> 03:22:44,250
And like, the kind of fully
automated communism stuff of like how

3587
03:22:44,250 --> 03:22:47,250
it's going to destroy all the jobs,
but then we can use the technology

3588
03:22:47,250 --> 03:22:48,839
to, to free us and blah blah blah.

3589
03:22:49,199 --> 03:22:51,210
And how it actually just meant
like algorithm management

3590
03:22:51,210 --> 03:22:52,320
and more pressure on workers.

3591
03:22:52,560 --> 03:22:56,880
But like the piece that even kind
of came before that was Uber really.

3592
03:22:56,884 --> 03:22:58,109
And and the gig economy.

3593
03:22:58,140 --> 03:22:58,439
Right.

3594
03:22:58,439 --> 03:23:00,269
And that was like, for me it was like.

3595
03:23:00,735 --> 03:23:03,615
No, this is like, this is like
terrible and it's being sold

3596
03:23:03,615 --> 03:23:05,175
as like liberation and freedom.

3597
03:23:05,235 --> 03:23:07,845
And this is just crazy to me.

3598
03:23:08,175 --> 03:23:11,655
I was observing it, but then also
reading the work of like, people

3599
03:23:11,655 --> 03:23:12,945
who did great analysis of this.

3600
03:23:12,945 --> 03:23:15,885
Hubert Horan of
course, who is like,

3601
03:23:15,885 --> 03:23:18,915
one of the like original Uber
critics who, who did this so well.

3602
03:23:18,915 --> 03:23:23,385
But yeah, it was just like, that was like
a real turning point for me and a, and a

3603
03:23:23,385 --> 03:23:27,435
real moment of like, wake up to what is
really going on here and like how this

3604
03:23:27,435 --> 03:23:32,085
industry works and the kind
of extraction and control and whatnot

3605
03:23:32,085 --> 03:23:35,775
that is so that this whole industry
is wrapped up in, and, and I'm sure

3606
03:23:35,775 --> 03:23:38,715
like, the Snowden revelations
and stuff played into that as well.

3607
03:23:38,715 --> 03:23:41,445
It was like, 'cause
there's so many discourses around

3608
03:23:41,445 --> 03:23:44,235
technology that are like, oh, it's
so liberatory and blah, blah, blah.

3609
03:23:44,235 --> 03:23:47,295
And then you look at it and it's like,
no, this is like private companies

3610
03:23:47,295 --> 03:23:50,625
building out a surveillance network that
then the state can piggyback off of.

3611
03:23:50,745 --> 03:23:52,995
And we're supposed to act like the
government is the bad guy, but the

3612
03:23:52,995 --> 03:23:54,225
tech companies are the good guys.

3613
03:23:54,225 --> 03:23:55,335
What, what is going on?

3614
03:23:55,365 --> 03:23:55,695
You know?

3615
03:23:55,695 --> 03:23:56,355
Yeah, exactly.

3616
03:23:56,355 --> 03:23:58,545
And it's also
liberatory for whom, right?

3617
03:23:58,545 --> 03:24:03,015
Is it liberatory for the workers in
Africa that have to look at extremely

3618
03:24:03,015 --> 03:24:07,185
triggering material all day long,
in order to train these algorithms?

3619
03:24:07,185 --> 03:24:12,195
Is it liberatory for any nation in Africa
where they p they mine the minerals that

3620
03:24:12,195 --> 03:24:15,945
are necessary for building computers
And then these countries are not asked

3621
03:24:15,945 --> 03:24:17,175
to share in the profits whatsoever.

3622
03:24:17,385 --> 03:24:21,045
Is it liberatory for the small towns where
these data centers are being built, where

3623
03:24:21,045 --> 03:24:22,605
there are water and errors being poisoned?

3624
03:24:22,785 --> 03:24:24,315
Who is it liberatory for?

3625
03:24:24,525 --> 03:24:26,985
Then you understand what
class struggle is, right?

3626
03:24:27,135 --> 03:24:30,735
You understand there's workers and
bosses and you're one or the other.

3627
03:24:30,915 --> 03:24:35,295
And there if you, I think being on
the autism spectrum can help a little

3628
03:24:35,295 --> 03:24:39,465
bit here where it's just like, if you
understand binary thinking in that

3629
03:24:39,465 --> 03:24:43,305
absolute if that, if you understand that
nuance is there, but it doesn't matter.

3630
03:24:43,695 --> 03:24:47,355
Really, then you really see the
entire world for what it is.

3631
03:24:47,625 --> 03:24:51,164
There is a small class of people
that are at the top and they

3632
03:24:51,164 --> 03:24:54,824
want to change the world so that
everything funnels towards them.

3633
03:24:54,914 --> 03:24:56,595
And we are their serfs.

3634
03:24:56,775 --> 03:24:57,975
We are in a fiefdom.

3635
03:24:57,975 --> 03:24:59,115
We live underneath them.

3636
03:25:11,475 --> 03:25:15,824
I will soon be introducing legislation,
calling for a moratorium on the

3637
03:25:15,824 --> 03:25:17,805
construction of new data centers.

3638
03:25:18,555 --> 03:25:24,570
Now as a result, I've been called a
Luddite anti-innovation, anti-progress

3639
03:25:24,980 --> 03:25:27,405
proin among many other things.

3640
03:25:27,914 --> 03:25:30,975
So why am I doing that?

3641
03:25:31,215 --> 03:25:35,535
Why am I calling for a moratorium on
the construction of new data centers?

3642
03:25:36,135 --> 03:25:43,395
Bottom line, we are at the beginning
of the most profound technological

3643
03:25:43,395 --> 03:25:45,525
revolution in world history.

3644
03:25:46,244 --> 03:25:46,965
That's the truth.

3645
03:25:47,655 --> 03:25:52,815
This is a revolution which will bring
unimaginable changes to our world.

3646
03:25:53,505 --> 03:25:59,175
This is a revolution which will impact
our economy with massive job displacement.

3647
03:25:59,775 --> 03:26:02,775
It will threaten our
democratic institutions.

3648
03:26:03,225 --> 03:26:08,685
It will impact our emotional wellbeing and
what it even means to be a human being.

3649
03:26:09,285 --> 03:26:12,375
It will impact how we
educate and raise our kids.

3650
03:26:12,885 --> 03:26:15,765
It will impact the nature of warfare.

3651
03:26:16,185 --> 03:26:23,324
Something we are seeing, right now
in Iran further and frighteningly some

3652
03:26:23,324 --> 03:26:28,935
very knowledgeable people, fear that,
well, that what was once seen as science

3653
03:26:28,935 --> 03:26:36,285
fiction could soon become a reality, and
that is that super intelligent I AI could

3654
03:26:36,285 --> 03:26:41,085
become smarter than human beings could
become independent of human control.

3655
03:26:41,415 --> 03:26:45,945
And pose an existential threat
to the entire human race.

3656
03:26:46,455 --> 03:26:51,975
In other words, human beings could
actually lose control over the planet.

3657
03:26:53,325 --> 03:26:57,945
And in the midst of all of that,
all of this transformative change,

3658
03:26:58,845 --> 03:27:04,425
what I have to tell you is that the
United States Congress hasn't a clue,

3659
03:27:05,025 --> 03:27:09,825
not a clue as to how to respond to
these revolutionary technologies

3660
03:27:10,065 --> 03:27:12,225
and protect the American people.

3661
03:27:12,825 --> 03:27:15,825
And it's not only not having a clue,
they're out busy raising money all

3662
03:27:15,825 --> 03:27:20,295
day long from AI and their Super
pacs, which is a whole other problem.

3663
03:27:21,705 --> 03:27:26,355
As many of the AI revolution
is being pushed by the wealthiest

3664
03:27:26,355 --> 03:27:31,905
people in our country, including Elon
Musk, Jeff Bezos, Larry Ellison, mark

3665
03:27:31,905 --> 03:27:33,885
Zuckerberg, Peter Thiel, and others.

3666
03:27:34,575 --> 03:27:37,905
All of these people are
multi-billionaires, who, if they are

3667
03:27:37,905 --> 03:27:44,325
successful at AI, will become even richer
and more powerful than they are today.

3668
03:27:45,435 --> 03:27:52,935
What I wanna do now is not tell you
my fears regarding AI and robotics.

3669
03:27:53,655 --> 03:27:56,265
I want you to actually hear from them.

3670
03:27:57,015 --> 03:28:02,055
The billionaires who are pushing
these technologies listen

3671
03:28:02,145 --> 03:28:04,425
carefully to what they are saying.

3672
03:28:05,235 --> 03:28:12,105
Elon Musk, wealthiest person
alive, stated that quote, AI and

3673
03:28:12,105 --> 03:28:16,185
robots will replace all jobs.

3674
03:28:17,055 --> 03:28:20,955
All jobs working will be optional.

3675
03:28:21,045 --> 03:28:23,955
End of quote, Dario Amodei,

3676
03:28:23,955 --> 03:28:32,235
Amodei, the CEO of Anthropic predicted
that quote, AI could displace half

3677
03:28:32,595 --> 03:28:37,245
of all entry level white collar
jobs in the next one to five years.

3678
03:28:37,949 --> 03:28:45,060
And that quote, humanity is about to
be handed almost unimaginable power.

3679
03:28:45,779 --> 03:28:51,150
And it is deeply unclear whether our
social, political, and technological

3680
03:28:51,150 --> 03:28:54,630
systems possess the maturity to wield it.

3681
03:28:54,720 --> 03:28:56,490
End quote, that's Amodei.

3682
03:28:57,359 --> 03:29:02,490
According to Demis Hassabis, the
head of Google DeepMind,

3683
03:29:03,390 --> 03:29:04,859
this is Google DeepMind.

3684
03:29:05,609 --> 03:29:10,890
The AI revolution will be 10
times bigger than the Industrial

3685
03:29:10,890 --> 03:29:13,980
revolution and 10 times faster.

3686
03:29:14,609 --> 03:29:15,449
Alright, you got that.

3687
03:29:16,170 --> 03:29:21,779
That means it will have a hundred
times greater impact on society

3688
03:29:22,080 --> 03:29:24,029
than the industrial Revolution had.

3689
03:29:25,230 --> 03:29:30,269
Jeff Bezos, the fourth richest person
in the world, has been pushing his

3690
03:29:30,269 --> 03:29:35,880
staff for years to think big and
envision what it would take for

3691
03:29:35,880 --> 03:29:42,029
Amazon, which he owns to fully automate
its operations and replace at least

3692
03:29:42,300 --> 03:29:47,160
600,000 warehouse workers with robots.

3693
03:29:47,250 --> 03:29:49,500
600,000 jobs gone.

3694
03:29:49,890 --> 03:29:51,210
Robots doing the work.

3695
03:29:51,929 --> 03:29:57,179
Bill Gates, also one of the wealthiest
people on earth, predicted that humans

3696
03:29:57,179 --> 03:30:02,820
quote won't be needed for most things,
end quote, such as manufacturing

3697
03:30:02,820 --> 03:30:07,710
products, delivering packages, or
growing food over the next decade.

3698
03:30:08,130 --> 03:30:15,150
Due to artificial intelligence, Mustafa
Suleyman, the CEO of Microsoft AI said

3699
03:30:15,150 --> 03:30:20,970
most white collar work, quote, will
be fully automated by an AI when the

3700
03:30:20,970 --> 03:30:23,519
ne within the next 12 to 18 months.

3701
03:30:23,519 --> 03:30:29,640
End quote, Jim Foley, the CEO of Ford
predicted that AI will eliminate,

3702
03:30:29,640 --> 03:30:34,080
quote, nearly half, literally half
of all white college jobs in the US

3703
03:30:34,080 --> 03:30:36,060
End quote, within the next decade.

3704
03:30:37,215 --> 03:30:38,175
I want you to hear this one.

3705
03:30:39,465 --> 03:30:44,085
Larry Ellison, also one of the
richest people on earth and a major

3706
03:30:44,085 --> 03:30:50,475
investor in ai, said that there will
be an artificial intelligence powered

3707
03:30:50,835 --> 03:30:56,685
surveillance state where quote,
citizens will be on their best behavior

3708
03:30:57,285 --> 03:31:02,595
because we're constantly recording and
reporting everything that is going on.

3709
03:31:04,245 --> 03:31:10,725
Dr. Geoffrey Hinton considered to be the
godfather of AI believes there is a quote,

3710
03:31:11,415 --> 03:31:16,515
10% to 20% chance for AI to wipe us out.

3711
03:31:18,705 --> 03:31:24,735
Mark Zuckerberg, the fifth richest
person in the world, is building a

3712
03:31:24,735 --> 03:31:31,095
data center in the state of Louisiana,
a data center that is the size of

3713
03:31:31,095 --> 03:31:38,025
Manhattan and will use three times the
quantity of electricity that the entire

3714
03:31:38,025 --> 03:31:41,115
city of New Orleans uses every year.

3715
03:31:42,855 --> 03:31:50,085
Alright, now, for many years now, leading
experts have called for regulation and

3716
03:31:50,085 --> 03:31:56,265
reasonable pauses to the development of
artificial intelligence to ensure the

3717
03:31:56,265 --> 03:31:58,755
safety, the very safety of humanity.

3718
03:31:59,685 --> 03:32:03,765
Let's go back to our good friend,
Elon Musk, he said back in

3719
03:32:03,765 --> 03:32:07,575
2018, quote, this is Elon Musk.

3720
03:32:07,995 --> 03:32:09,225
Mark my words.

3721
03:32:09,795 --> 03:32:12,645
AI is far more dangerous than nukes.

3722
03:32:13,575 --> 03:32:16,665
So why do we have no regulatory oversight?

3723
03:32:17,325 --> 03:32:18,375
This is insane.

3724
03:32:18,375 --> 03:32:27,045
End quote, Elon Musk, in March of 2023,
over 1000 business leaders in the big

3725
03:32:27,045 --> 03:32:30,015
tech industry, prominent scientists.

3726
03:32:30,630 --> 03:32:38,820
AI researchers and academics co-signed
an open letter entitled Pause Giant

3727
03:32:39,210 --> 03:32:47,070
AI Experiments and quote stating,
we must ask ourselves, should we

3728
03:32:47,070 --> 03:32:52,380
let machines flood our information
channels with propaganda and untruth?

3729
03:32:53,279 --> 03:32:58,140
Should we automate away all the
jobs, including the fulfilling ones?

3730
03:32:58,949 --> 03:33:03,720
Should we develop non-human minds
that might eventually outnumber

3731
03:33:03,869 --> 03:33:06,750
outsmart obsolete and replace us?

3732
03:33:07,710 --> 03:33:12,840
Should we risk control, loss
of control of our civilization?

3733
03:33:13,980 --> 03:33:17,369
Should we risk loss of
control of our civilization?

3734
03:33:17,910 --> 03:33:23,730
Such decisions must not be
delegated to unelected tech leaders.

3735
03:33:24,330 --> 03:33:29,550
Therefore, we call on all AI labs to
immediately pause for at least six months.

3736
03:33:29,550 --> 03:33:34,500
The training of AI systems
more, more powerful than GPT-4.

3737
03:33:35,429 --> 03:33:39,990
This pause should be public and
verifiable and include all key actors.

3738
03:33:40,769 --> 03:33:47,189
If such a pause cannot be enacted
quickly, governments should step in and

3739
03:33:47,189 --> 03:33:49,529
institute a moratorium end of quote.

3740
03:33:50,910 --> 03:33:59,310
That is what some of the leaders
in the AI industry have said, and

3741
03:33:59,310 --> 03:34:04,470
clearly where we are right now is
that there has not been any pause.

3742
03:34:05,250 --> 03:34:08,009
There has been massive amounts
of competition between one

3743
03:34:08,009 --> 03:34:11,099
company and the other, between
the United States and China.

3744
03:34:11,759 --> 03:34:19,050
So bottom line is that, in my view, to
protect our workers from losing their

3745
03:34:19,050 --> 03:34:23,220
jobs, to protect human beings from
attacks on their mental health, to

3746
03:34:23,220 --> 03:34:26,939
protect our kids, to protect the safety.

3747
03:34:27,570 --> 03:34:28,650
Of human life.

3748
03:34:29,310 --> 03:34:32,100
Yeah, we need a moratorium
on data centers.

3749
03:34:32,580 --> 03:34:33,990
We need to take a deep breath.

3750
03:34:34,650 --> 03:34:40,500
We need to make sure that AI and
robotics work for all of us, not

3751
03:34:40,500 --> 03:34:42,630
just a handful of billionaires.

3752
03:34:43,320 --> 03:34:43,685
Thanks very much.

3753
03:35:03,189 --> 03:35:04,210
That's going to be it for today.

3754
03:35:04,210 --> 03:35:05,859
As always, keep the comments coming in.

3755
03:35:05,859 --> 03:35:08,619
You can record and rerecord
a voice message by tapping

3756
03:35:08,619 --> 03:35:09,609
the link in the show notes.

3757
03:35:09,910 --> 03:35:13,660
You can reach us on signal at the
handle Best of left zero one or

3758
03:35:13,660 --> 03:35:16,600
simply email me to j@bestofleft.com.

3759
03:35:16,900 --> 03:35:19,869
The additional sections of the show
included clips from Confronting

3760
03:35:19,869 --> 03:35:25,210
Capitalism, Pitchfork Economics, your
Undivided Attention sound side, the

3761
03:35:25,210 --> 03:35:30,460
Brian Lara Show on the media today
explained Decoder Tech Won't Save us,

3762
03:35:30,670 --> 03:35:35,320
why America and Senator Bernie Sanders,
further details are in the show notes.

3763
03:35:35,590 --> 03:35:36,699
Thanks everyone for listening.

3764
03:35:36,820 --> 03:35:38,830
Thanks to Dion and Aaron for
their production work for the

3765
03:35:38,830 --> 03:35:40,539
show and co-starring in Solved.

3766
03:35:40,960 --> 03:35:43,480
Thanks to Amanda for all of her
work behind the scenes, including

3767
03:35:43,480 --> 03:35:44,680
her co-starring Unsolved.

3768
03:35:45,010 --> 03:35:48,160
Thanks to our editors and thanks to
those who already support the show by

3769
03:35:48,160 --> 03:35:51,369
becoming a member or purchasing guest
memberships, you'll find that link

3770
03:35:51,369 --> 03:35:54,189
in the show notes along with the link
to join our Discord community where

3771
03:35:54,189 --> 03:35:55,810
you can also continue the discussion.

3772
03:35:56,050 --> 03:35:58,750
And don't forget to follow us on
all the social media platforms.

3773
03:35:58,900 --> 03:36:01,689
We are on Blue Sky, but we're
also making a move to video on

3774
03:36:01,720 --> 03:36:03,699
Instagram with our new show solved.

3775
03:36:03,850 --> 03:36:05,109
So please support us there.

3776
03:36:05,619 --> 03:36:08,020
So coming to you from far outside
the conventional wisdom of

3777
03:36:08,020 --> 03:36:11,140
Washington dc My name is Jay, and
this has been the Best of the Left

3778
03:36:11,140 --> 03:36:12,910
podcast coming to you twice weekly.

3779
03:36:13,060 --> 03:36:15,250
Thanks entirely to the members
and donors to the show.

3780
03:36:15,580 --> 03:36:17,145
From best of the left.com.