1
00:00:00,000 --> 00:00:26,400
Hey everybody, KMO here.

2
00:00:26,400 --> 00:00:30,000
This is episode number 23 of the KMO Show.

3
00:00:30,000 --> 00:00:33,960
And what I have here is a conversation with Tommy Blanchard, somebody that I just spoke

4
00:00:33,960 --> 00:00:36,800
to for the first time today.

5
00:00:36,800 --> 00:00:43,900
He reached out via Substack, and reading from his Substack bio, Tommy is a PhD in neuroscience

6
00:00:43,900 --> 00:00:49,720
with degrees in philosophy and computer science and postdoctoral training at Harvard, writing

7
00:00:49,720 --> 00:00:52,080
about science and philosophy of mind.

8
00:00:52,080 --> 00:00:58,280
He is a semi-pro science fiction author, and his day job is as a data scientist.

9
00:00:58,280 --> 00:01:01,260
And I'm reading this after the fact.

10
00:01:01,260 --> 00:01:05,560
We didn't really talk much about science fiction, certainly not about the writing of science

11
00:01:05,560 --> 00:01:06,920
fiction.

12
00:01:06,920 --> 00:01:10,960
So we're going to mostly talk about artificial intelligence, and in the first hour, we're

13
00:01:10,960 --> 00:01:14,640
going to talk about how it relates to society.

14
00:01:14,640 --> 00:01:17,200
So here we go.

15
00:01:17,200 --> 00:01:19,200
You're listening to the KMO Show.

16
00:01:19,200 --> 00:01:20,200
Let's go.

17
00:01:20,200 --> 00:01:26,160
I'm talking with Tommy Blanchard, somebody that has reached out to me via Substack, and

18
00:01:26,160 --> 00:01:27,560
I'm really enjoying Substack.

19
00:01:27,560 --> 00:01:31,680
So it's good to make a face-to-face, voice-to-voice connection with somebody from that platform.

20
00:01:31,680 --> 00:01:35,080
So Tommy, tell me something about yourself.

21
00:01:35,080 --> 00:01:36,080
Sure.

22
00:01:36,080 --> 00:01:44,120
Yeah, since you mentioned Substack, I am a relative newcomer to Substack.

23
00:01:44,120 --> 00:01:46,040
And I guess, what other this is?

24
00:01:46,040 --> 00:01:50,760
I started almost, I think I'm coming up on four months.

25
00:01:50,760 --> 00:01:53,000
Okay, I'm pretty new.

26
00:01:53,000 --> 00:01:59,360
And this is, I played around with blogging, like back when I was like in undergrad, so

27
00:01:59,360 --> 00:02:06,560
like over a decade ago, like two decades ago at this point.

28
00:02:06,560 --> 00:02:14,440
Sometimes have those moments where you realize, oh, I'm older than I think.

29
00:02:14,440 --> 00:02:19,680
And so I'm relatively new to the whole content creation or whatever you want to call it space

30
00:02:19,680 --> 00:02:21,000
as well.

31
00:02:21,000 --> 00:02:26,760
My background is serving in academia.

32
00:02:26,760 --> 00:02:31,280
So I have a bit of an odd academic background.

33
00:02:31,280 --> 00:02:39,320
I started computer science in undergrad, went through a philosophy master's, and then ended

34
00:02:39,320 --> 00:02:47,280
up in neuroscience for my PhD, stuck around in academia for a while as a postdoc before

35
00:02:47,280 --> 00:02:51,640
moving on to work in data science.

36
00:02:51,640 --> 00:02:58,200
And so now I'm on Substack to kind of explore some of those themes and interests that attracted

37
00:02:58,200 --> 00:03:03,880
me to those academic fields to begin with, but I'm able to explore them with a lot more

38
00:03:03,880 --> 00:03:10,120
freedom than academia often allows.

39
00:03:10,120 --> 00:03:15,640
Well I studied philosophy, went to grad school for philosophy, philosophy of science, philosophy

40
00:03:15,640 --> 00:03:18,240
of mind, but that was back in the 90s.

41
00:03:18,240 --> 00:03:23,280
That was back when notions of using neural networks were sort of theoretical.

42
00:03:23,280 --> 00:03:27,500
People who were into philosophy of mind thought, yeah, maybe this will work out.

43
00:03:27,500 --> 00:03:28,500
Maybe it'll scale up to something.

44
00:03:28,500 --> 00:03:30,640
Other people are like, nah, it hasn't yet.

45
00:03:30,640 --> 00:03:31,640
It probably won't.

46
00:03:31,640 --> 00:03:32,640
Yeah, yeah.

47
00:03:32,640 --> 00:03:33,920
That's wild.

48
00:03:33,920 --> 00:03:40,000
And actually, that brings me to how I found you, was actually, I'm new to this space.

49
00:03:40,000 --> 00:03:42,760
I'm kind of exploring Substack quite a bit.

50
00:03:42,760 --> 00:03:47,360
And I came across your article on AI.

51
00:03:47,360 --> 00:03:53,160
And I thought, oh, hey, this is the kind of perspective that I feel like is often missing

52
00:03:53,160 --> 00:03:55,880
in a lot of discussions of AI.

53
00:03:55,880 --> 00:04:03,160
I felt like, and maybe this was me reading into it, but I felt like you had a much more

54
00:04:03,160 --> 00:04:04,160
balanced perspective.

55
00:04:04,160 --> 00:04:07,480
Like, hey, there's some good, some bad about this whole AI thing.

56
00:04:07,480 --> 00:04:08,480
It's useful for some things.

57
00:04:08,480 --> 00:04:11,360
It's not great at other things.

58
00:04:11,360 --> 00:04:18,360
And that's a nuance that I often find missing in this world, where it's either like, you

59
00:04:18,360 --> 00:04:24,200
have the hype people who are super excited about AI and trying to get venture capitalists

60
00:04:24,200 --> 00:04:29,560
to pour more money in, and then the do-mers about it who either think, oh, it's useless

61
00:04:29,560 --> 00:04:37,680
for anything, or it's going to destroy the world or destroy art, at least.

62
00:04:37,680 --> 00:04:41,920
And so I thought you had a much more kind of balanced take on the whole thing.

63
00:04:41,920 --> 00:04:48,480
So I thought it would be interesting to talk to somebody with that kind of nuanced perspective.

64
00:04:48,480 --> 00:04:54,920
Well, I don't know if you just read the one Substack post, but I have been interested

65
00:04:54,920 --> 00:04:56,320
in AI since the 90s.

66
00:04:56,320 --> 00:04:59,960
But I'm also a visual artist, and I'm a cartoonist.

67
00:04:59,960 --> 00:05:05,400
And before the large language models really hit public consciousness, like six months

68
00:05:05,400 --> 00:05:12,800
before that, OpenAI's Dolly, too, really woke people up, particularly artists who felt

69
00:05:12,800 --> 00:05:13,800
threatened by it.

70
00:05:13,800 --> 00:05:21,120
And in my opinion, more than working artists who felt threatened by it, aspiring artists,

71
00:05:21,120 --> 00:05:25,720
young people who don't want a straight job, who they think, hey, I'm going to be an artist.

72
00:05:25,720 --> 00:05:30,040
Oh, shit, this machine can do something much better than I can do, and it can do it in

73
00:05:30,040 --> 00:05:31,560
seconds.

74
00:05:31,560 --> 00:05:34,240
So there was a lot of people in the art.

75
00:05:34,240 --> 00:05:39,080
I don't want to say the art world, but that brings to mind visions of galleries and such.

76
00:05:39,080 --> 00:05:44,880
You know, young people with ambition and a love for popular media and a wish to participate

77
00:05:44,880 --> 00:05:47,520
in its creation.

78
00:05:47,520 --> 00:05:51,080
They were calling for the machines to be shut down.

79
00:05:51,080 --> 00:05:57,440
They were calling for a regulation, saying that these things can't be trained on any

80
00:05:57,440 --> 00:05:59,840
copyrighted material.

81
00:05:59,840 --> 00:06:05,240
And as an artist myself, I thought, well, that's trying to hold back the tide.

82
00:06:05,240 --> 00:06:07,360
That's not going to fly.

83
00:06:07,360 --> 00:06:08,360
That's completely unrealistic.

84
00:06:08,360 --> 00:06:12,200
And at the same time, as somebody who's been interested in the philosophy of mind and the

85
00:06:12,200 --> 00:06:16,200
philosophy of science, and particularly artificial intelligence, and particularly somebody who's

86
00:06:16,200 --> 00:06:21,760
been aware of the potential of neural networks for decades before they actually came along,

87
00:06:21,760 --> 00:06:30,720
the idea that a neural network is not allowed to reference popular imagery, it just seemed

88
00:06:30,720 --> 00:06:32,040
like an absurd handicap.

89
00:06:32,040 --> 00:06:37,720
I mean, as somebody who wants a new type of intelligence, a new type of consciousness

90
00:06:37,720 --> 00:06:44,360
to manifest itself, saying, yeah, there was a meme that went around.

91
00:06:44,360 --> 00:06:48,160
It showed a box of crayons and it said, if you want to make images start like the rest

92
00:06:48,160 --> 00:06:50,020
of us, pick up a crayon.

93
00:06:50,020 --> 00:06:54,960
But even the kid who picks up a crayon is copying copyrighted material.

94
00:06:54,960 --> 00:06:57,880
They're like superhero comics, they're drawing Spider-Man.

95
00:06:57,880 --> 00:07:01,560
They've got the comic book open next to their drawing pad and they're copying.

96
00:07:01,560 --> 00:07:03,200
And that's how they learn.

97
00:07:03,200 --> 00:07:08,440
And to say that these types of minds are not allowed to learn that way, to me, it seemed

98
00:07:08,440 --> 00:07:15,600
like not just an arbitrary limitation, but one that just completely misunderstands the

99
00:07:15,600 --> 00:07:17,400
nature of the technology.

100
00:07:17,400 --> 00:07:21,360
And you can probably tell, I think about this a lot, I can continue talking, but I won't.

101
00:07:21,360 --> 00:07:24,040
So I'll stop and mute and let you talk.

102
00:07:24,040 --> 00:07:25,040
Yeah.

103
00:07:25,040 --> 00:07:28,480
So I think that's right.

104
00:07:28,480 --> 00:07:34,040
And that's one of the things that I think about a lot, the fact that we're kind of holding

105
00:07:34,040 --> 00:07:40,920
AI to, generative AI, I should say, to a different standard here, if we're saying that we're

106
00:07:40,920 --> 00:07:50,920
not going to allow them to be trained on certain images or text that a human who's learning

107
00:07:50,920 --> 00:07:55,880
to do those same things has access to.

108
00:07:55,880 --> 00:08:00,440
I come at this from kind of two perspectives.

109
00:08:00,440 --> 00:08:05,320
So one is like you, I have an interest in philosophy of mind and I think it's interesting

110
00:08:05,320 --> 00:08:11,640
to look at neural networks, both as a model of certain kinds of cognition, like what can

111
00:08:11,640 --> 00:08:12,960
you do with these things?

112
00:08:12,960 --> 00:08:15,320
What kind of functions can they perform?

113
00:08:15,320 --> 00:08:23,240
And more recently, the large language models as one way of thinking about how language

114
00:08:23,240 --> 00:08:24,240
is learned.

115
00:08:24,240 --> 00:08:30,840
There's a lot of arguments about exactly how closely that resembles human language learning

116
00:08:30,840 --> 00:08:33,640
and concept learning more broadly.

117
00:08:33,640 --> 00:08:36,800
But I think it's at least a really interesting kind of existence proof of like, hey, you

118
00:08:36,800 --> 00:08:43,280
can actually learn a heck of a lot just by being exposed to and doing the right kind

119
00:08:43,280 --> 00:08:47,240
of inferences over just text.

120
00:08:47,240 --> 00:08:49,240
And that's kind of an amazing thing to me.

121
00:08:49,240 --> 00:08:54,040
That's kind of like a mind-extending concept that there's that much information just in

122
00:08:54,040 --> 00:08:56,040
the language that we use.

123
00:08:56,040 --> 00:09:01,400
The other angle I come at this with is I'm a data scientist.

124
00:09:01,400 --> 00:09:02,400
That's my day job.

125
00:09:02,400 --> 00:09:04,080
I work with these things, right?

126
00:09:04,080 --> 00:09:09,900
This isn't just some theoretical thing for me and I'm not using it even as a hobby to

127
00:09:09,900 --> 00:09:13,680
make art or anything like that.

128
00:09:13,680 --> 00:09:22,580
I'm also working with software engineers and doing research on algorithms to put this into

129
00:09:22,580 --> 00:09:24,240
actual products.

130
00:09:24,240 --> 00:09:28,480
And so a lot of the takes out there that I see about like, oh, this stuff isn't useful

131
00:09:28,480 --> 00:09:35,000
for anything except, you know, writing bad poetry or whatever, it's just not true.

132
00:09:35,000 --> 00:09:41,560
Like tons of companies, including my own, have built features into their product that

133
00:09:41,560 --> 00:09:48,640
have made the product better in ways that might not be visible to the users using these

134
00:09:48,640 --> 00:09:49,760
large language models.

135
00:09:49,760 --> 00:09:57,880
And so I think a more nuanced understanding of how these work and the actual use cases

136
00:09:57,880 --> 00:10:07,320
of them just helps to diffuse some of the extreme takes on either side of this whole

137
00:10:07,320 --> 00:10:09,000
thing.

138
00:10:09,000 --> 00:10:12,160
Now, I did a podcast.

139
00:10:12,160 --> 00:10:14,720
I've been podcasting since 2006.

140
00:10:14,720 --> 00:10:17,320
So I've been doing it for a long time.

141
00:10:17,320 --> 00:10:23,120
And for a period, I mean, when I had the largest audience, I was basically articulating a doomer

142
00:10:23,120 --> 00:10:27,160
message not from AI, but from peak oil.

143
00:10:27,160 --> 00:10:32,240
And a lot of people are very attracted to the idea that our civilization is about to

144
00:10:32,240 --> 00:10:33,240
collapse.

145
00:10:33,240 --> 00:10:38,480
I think it's a liberating fantasy for a lot of people.

146
00:10:38,480 --> 00:10:42,240
And I didn't talk much about AI during that period.

147
00:10:42,240 --> 00:10:45,440
Now that I'm talking a lot more about AI, I'm hearing from a lot of people who have

148
00:10:45,440 --> 00:10:50,720
stuck with me, even though I'm not talking about the thing they like anymore.

149
00:10:50,720 --> 00:10:54,960
I guess they're used to the sound of my voice or whatever, and they continue to listen.

150
00:10:54,960 --> 00:11:00,340
And lots of people write to me, not lots of people, but certain people write to me again

151
00:11:00,340 --> 00:11:04,720
and again just to tell me how off-putting the whole notion of AI is, how they will never

152
00:11:04,720 --> 00:11:11,240
participate with it, how it's demeaning to humanity, how everything about it is just

153
00:11:11,240 --> 00:11:14,360
poisoned to what is good about humanity.

154
00:11:14,360 --> 00:11:19,420
And I don't agree with that, but I don't push back against it too hard because it's clearly

155
00:11:19,420 --> 00:11:26,240
not a position that anybody has come to through a process of working with the technology or

156
00:11:26,240 --> 00:11:32,040
studying mental models or different types of neural architecture.

157
00:11:32,040 --> 00:11:35,920
It's an opinion that satisfies some deep need.

158
00:11:35,920 --> 00:11:40,760
And I'm not going to replace that need or offer them, I'm not going to talk them into

159
00:11:40,760 --> 00:11:41,920
not needing it anymore.

160
00:11:41,920 --> 00:11:43,920
So I don't try.

161
00:11:43,920 --> 00:11:48,000
And again, these are topics that I think about a lot and I can continue to monologue, but

162
00:11:48,000 --> 00:11:51,680
once again, I will hit mute and let you take over.

163
00:11:51,680 --> 00:11:52,680
Yeah.

164
00:11:52,680 --> 00:12:03,160
I listened to a couple of your other podcasts with some of the almost reflections about

165
00:12:03,160 --> 00:12:15,520
deumerism on peak oil and I do think this, to segue a little bit on our topic here, I

166
00:12:15,520 --> 00:12:25,720
think there's a very easy to understand operating principle here that like, hey, we kind of

167
00:12:25,720 --> 00:12:26,720
like our narratives, right?

168
00:12:26,720 --> 00:12:30,360
And a lot of those narratives are fairly negative.

169
00:12:30,360 --> 00:12:34,960
There's a well-observed negativity bias that humans tend to have.

170
00:12:34,960 --> 00:12:43,760
We kind of like in some sense, or drawn to is probably the more correct way of saying

171
00:12:43,760 --> 00:12:51,480
it, these narratives of imminent collapse or like something bad happening, we're drawn

172
00:12:51,480 --> 00:12:57,000
to stories of bad things happening in the media.

173
00:12:57,000 --> 00:13:00,600
If it bleeds, it leads.

174
00:13:00,600 --> 00:13:12,500
And I think that is the AI deumerism is the latest iteration of that.

175
00:13:12,500 --> 00:13:22,280
You still see a lot of the economic deumerism and I think there's good reason to be, we

176
00:13:22,280 --> 00:13:28,280
live in a brave new world and there's lots of unknowns, but I think it's worthwhile to

177
00:13:28,280 --> 00:13:31,640
kind of take a step back and say, well, I've kind of been through this before.

178
00:13:31,640 --> 00:13:36,320
Are we sure this time is the time we're going to end the world?

179
00:13:36,320 --> 00:13:42,280
At some point, maybe we'll be right, then that's kind of the end of the cycle.

180
00:13:42,280 --> 00:13:47,320
I don't think there's any particular reason to think this time is special.

181
00:13:47,320 --> 00:13:54,880
I think we should look at AI and what it's capable of and think of safeguards that make

182
00:13:54,880 --> 00:13:58,840
sense, which it turns out, hey, the safeguards that make sense are generally the safeguards

183
00:13:58,840 --> 00:14:03,920
that we have in place to control all of the brilliant humans that we have in the world.

184
00:14:03,920 --> 00:14:08,800
All you need is one rogue human who's really smart that wants to do something really bad

185
00:14:08,800 --> 00:14:09,800
and it'll end the world.

186
00:14:09,800 --> 00:14:13,800
Well, no, because we have all kinds of safeguards to prevent that kind of stuff.

187
00:14:13,800 --> 00:14:16,240
Similarly, a rogue AI that's really smart.

188
00:14:16,240 --> 00:14:17,980
Well, what are they going to do?

189
00:14:17,980 --> 00:14:26,320
We have safeguards protecting anything that a bad actor could be thinking of doing.

190
00:14:26,320 --> 00:14:34,600
And so I think you need to be specific about what you think a rogue AI could do if you

191
00:14:34,600 --> 00:14:37,480
want to make the case that, yeah, this time is going to be different and we're going to

192
00:14:37,480 --> 00:14:41,000
end the world.

193
00:14:41,000 --> 00:14:47,320
People who are interested in AI are mostly talking about a California piece of legislation.

194
00:14:47,320 --> 00:14:54,440
I think it's SB 1047 and it is a proposal for regulating AI.

195
00:14:54,440 --> 00:14:59,400
And big AI companies say they want regulation, but big companies in general like regulation

196
00:14:59,400 --> 00:15:01,480
because it keeps the smaller players out.

197
00:15:01,480 --> 00:15:05,560
It's a barrier to entry to potential competitors.

198
00:15:05,560 --> 00:15:12,040
And I think that open AI is just a paradigm example of how vulnerable the established

199
00:15:12,040 --> 00:15:16,040
players are to upstart competition in this space.

200
00:15:16,040 --> 00:15:24,680
Because by Silicon Valley standards, a modest amount of venture capital seed money, you

201
00:15:24,680 --> 00:15:30,160
can create a product that totally upsets the apple cards of all the established players

202
00:15:30,160 --> 00:15:34,560
to the point where they will attempt to incorporate you into their structure as quickly as possible.

203
00:15:34,560 --> 00:15:39,400
Because the relationship between Microsoft and open AI testifies.

204
00:15:39,400 --> 00:15:45,560
And the fact that Amazon.com has just dumped over $2 billion into Anthropic.

205
00:15:45,560 --> 00:15:46,840
Nobody wants to be left out in the cold.

206
00:15:46,840 --> 00:15:51,600
And Anthropic is an offshoot of open AI.

207
00:15:51,600 --> 00:15:54,840
Basically people who are not satisfied with open AI safety protocols went off and made

208
00:15:54,840 --> 00:16:00,260
their own company and created, in my opinion, a superior product in a very short time.

209
00:16:00,260 --> 00:16:03,640
So I sort of switched tracks.

210
00:16:03,640 --> 00:16:09,600
You were talking about safeguards and SB 0 or 4.0, what is it?

211
00:16:09,600 --> 00:16:13,720
I don't know, the California legislation.

212
00:16:13,720 --> 00:16:15,520
It proposes certain safeguards.

213
00:16:15,520 --> 00:16:21,840
But at the same time, one, lawmakers are not capable, as far as I can tell, of having even

214
00:16:21,840 --> 00:16:25,560
having a conversation on the level that you and I are talking about right now.

215
00:16:25,560 --> 00:16:28,960
And neither one of us is crafting legislation.

216
00:16:28,960 --> 00:16:33,120
And I wouldn't imagine myself capable of crafting legislation that is going to present any

217
00:16:33,120 --> 00:16:36,760
sort of AI related disaster.

218
00:16:36,760 --> 00:16:42,400
I just watched Oppenheimer, the Chris Nolan movie.

219
00:16:42,400 --> 00:16:48,680
And in it, the character of Oppenheimer, who I realize is based on a real person, but he's

220
00:16:48,680 --> 00:16:50,920
pretty much a fictional character in the movie.

221
00:16:50,920 --> 00:16:57,640
But he's saying, look, nobody understands what nuclear weapons are and they won't until

222
00:16:57,640 --> 00:17:00,200
they see them used.

223
00:17:00,200 --> 00:17:05,360
And so in order to save the world, we've got to nuke Japan, basically.

224
00:17:05,360 --> 00:17:12,500
I think that nobody in a position of power who can establish regulations that big companies

225
00:17:12,500 --> 00:17:19,060
have to abide by really has any clue as to how this technology is going to go wrong.

226
00:17:19,060 --> 00:17:20,440
And I don't either.

227
00:17:20,440 --> 00:17:26,200
And you can't really anticipate these sorts of things in such a dynamic and fast moving

228
00:17:26,200 --> 00:17:27,760
realm.

229
00:17:27,760 --> 00:17:34,880
And so we might have to endure a few catastrophic events before we even have any clue as to

230
00:17:34,880 --> 00:17:37,660
which hatches need to be batted down.

231
00:17:37,660 --> 00:17:39,160
So let me get your opinion on that.

232
00:17:39,160 --> 00:17:40,160
I love that.

233
00:17:40,160 --> 00:17:44,480
Yeah, well, are you sure we're not going to get nuked by an AI?

234
00:17:44,480 --> 00:17:48,040
Maybe we'll just have to endure a couple of nukes.

235
00:17:48,040 --> 00:17:56,000
So I think so, first of all, when I say safeguards, I'm much less thinking of regulations and

236
00:17:56,000 --> 00:17:59,200
legislation that specifically targets AI.

237
00:17:59,200 --> 00:18:03,720
I'm much more thinking of safeguards like how would an AI get a nuke?

238
00:18:03,720 --> 00:18:10,120
Well, they could try to get the US nuclear arsenal, I don't know, somehow hack into the

239
00:18:10,120 --> 00:18:15,200
network and get access to the controls for it.

240
00:18:15,200 --> 00:18:17,640
Well, we have safeguards for that, right?

241
00:18:17,640 --> 00:18:23,200
My understanding, and this might be the naive pop culture kind of understanding of it, is

242
00:18:23,200 --> 00:18:28,360
there's two guys that need to like simultaneously turn a key somewhere to launch a nuke.

243
00:18:28,360 --> 00:18:36,860
We have safeguards that can only be overcome with very specific mechanical stuff that needs

244
00:18:36,860 --> 00:18:39,840
to happen.

245
00:18:39,840 --> 00:18:46,240
And so sure, there's a lot of uncertainty about exactly what AIs are going to be capable

246
00:18:46,240 --> 00:18:51,200
of, but these kind of catastrophic events where they just get access to some weapon

247
00:18:51,200 --> 00:18:59,760
of mass destruction, if it was possible to get easy access to a weapon of mass destruction,

248
00:18:59,760 --> 00:19:01,560
bad actors would do it, right?

249
00:19:01,560 --> 00:19:03,640
Terrorists are smart.

250
00:19:03,640 --> 00:19:11,560
And so I think we can be, we can feel relatively safe in the fact that, yeah, a lot of people

251
00:19:11,560 --> 00:19:18,920
have tried to cause catastrophic stuff before, and we've certainly had terrorist attacks,

252
00:19:18,920 --> 00:19:29,240
but nothing that's been, you know, catastrophic at the level of threatening our whole society

253
00:19:29,240 --> 00:19:30,980
or anything like that.

254
00:19:30,980 --> 00:19:40,040
And so I think the level of damage that a misaligned AI could do is limited just because

255
00:19:40,040 --> 00:19:43,800
of those constraints that we already have in place.

256
00:19:43,800 --> 00:19:47,560
Well, to push back on that, I'd say there are constraints.

257
00:19:47,560 --> 00:19:52,120
There are safety precautions in place to protect people's digital identity, and yet people's

258
00:19:52,120 --> 00:19:55,520
digital identity gets stolen all the time.

259
00:19:55,520 --> 00:20:02,600
You know, there are people in Indonesia and, you know, various Southeast Asian countries

260
00:20:02,600 --> 00:20:08,480
who are sitting in front of a huge board full of cell phones, basically just contacting

261
00:20:08,480 --> 00:20:13,160
people whose phone numbers they got, they either stole or bought from somebody and just

262
00:20:13,160 --> 00:20:18,800
saying hi, and, you know, they attached a picture of a pretty woman to their profile.

263
00:20:18,800 --> 00:20:22,120
And you know, I get these things all the time.

264
00:20:22,120 --> 00:20:27,880
And these are just humans, you know, running very simple social scripts, and you do it

265
00:20:27,880 --> 00:20:32,280
often enough, and you know, you find people who are lonely, who are willing to talk and

266
00:20:32,280 --> 00:20:35,600
who will extend trust, and then you fleece them.

267
00:20:35,600 --> 00:20:39,100
And it's a model that works, and it's very, very simple.

268
00:20:39,100 --> 00:20:44,000
And I think one of the things people worry about with AI, particularly generative AI,

269
00:20:44,000 --> 00:20:49,320
is its ability to be persuasive and to gather information about a particular person to become

270
00:20:49,320 --> 00:20:52,000
super persuasive to that person.

271
00:20:52,000 --> 00:20:56,680
And I think, you know, one of the easy to imagine dangers is just what happens when

272
00:20:56,680 --> 00:21:02,160
much smarter AI replaces these very simple scripts in these, you know, these phishing

273
00:21:02,160 --> 00:21:05,160
schemes.

274
00:21:05,160 --> 00:21:09,560
When it comes to nuclear weapons, yeah, the US nuclear arsenal is hardened, it is very

275
00:21:09,560 --> 00:21:12,040
secure.

276
00:21:12,040 --> 00:21:16,980
But if you study the history of the Cold War, we came very close to nuclear exchanges on

277
00:21:16,980 --> 00:21:24,120
many occasions when there were, you know, flocks of birds or, you know, the sun hitting

278
00:21:24,120 --> 00:21:30,840
ice in a particular way, or just, you know, ambiguous signals made people think that there

279
00:21:30,840 --> 00:21:36,840
is a launch, you know, an incoming wave of nuclear weapons that we need to respond to.

280
00:21:36,840 --> 00:21:42,200
And in every instance, it was a human being who said, you know what, I don't believe it.

281
00:21:42,200 --> 00:21:45,240
And I know this could be the end of my career, and this could be the end of my life if I'm

282
00:21:45,240 --> 00:21:48,000
wrong, but no, we're going to stand down.

283
00:21:48,000 --> 00:21:52,880
And if it hadn't been for individual human beings saying, no, I'm not going to follow

284
00:21:52,880 --> 00:21:57,360
the protocol, we would be living in a very different timeline.

285
00:21:57,360 --> 00:22:02,480
And you know, I think the more persuasive AI gets, the more it can spoof those signals,

286
00:22:02,480 --> 00:22:08,680
the more it can engage in social engineering, which is very targeted and which is specific

287
00:22:08,680 --> 00:22:09,680
to individual people.

288
00:22:09,680 --> 00:22:15,760
I think that the hard parts of the system are, you know, the mechanical and the cybernetic

289
00:22:15,760 --> 00:22:18,600
parts and the weak portions are the human portions.

290
00:22:18,600 --> 00:22:21,160
And I think that's where AI is going to find a weakness.

291
00:22:21,160 --> 00:22:26,320
Now, I'm not particularly worried about, you know, AI starting a nuclear war.

292
00:22:26,320 --> 00:22:28,720
It's very Terminator.

293
00:22:28,720 --> 00:22:30,600
That's something that lots of people can imagine.

294
00:22:30,600 --> 00:22:34,280
So the thing that hits you upside the head as a society is not the thing that everybody

295
00:22:34,280 --> 00:22:35,280
was imagining.

296
00:22:35,280 --> 00:22:38,040
It was the thing that nobody thought of until after it happened.

297
00:22:38,040 --> 00:22:41,520
And then after the fact, it seemed obvious in hindsight.

298
00:22:41,520 --> 00:22:45,940
And I just I don't think there's any way to determine what those vulnerabilities are until

299
00:22:45,940 --> 00:22:47,840
they get exploited.

300
00:22:47,840 --> 00:22:51,520
So I think there will be, I'm not sanguine.

301
00:22:51,520 --> 00:22:57,640
At the same time, I just realized that doomerism is a fetish that feeds on itself and is a

302
00:22:57,640 --> 00:23:00,520
cognitive distortion that I reject.

303
00:23:00,520 --> 00:23:02,720
And I reject ideology in general.

304
00:23:02,720 --> 00:23:04,280
I've been caught up by it repeatedly.

305
00:23:04,280 --> 00:23:06,920
I used to be a diet and the world libertarian.

306
00:23:06,920 --> 00:23:07,920
Not anymore.

307
00:23:07,920 --> 00:23:11,040
Used to be a singularitarian before I became a doomer and then it was a doomer.

308
00:23:11,040 --> 00:23:16,440
And now I'm just like, you know what, these are all mind viruses and they're useful in

309
00:23:16,440 --> 00:23:17,480
some instances.

310
00:23:17,480 --> 00:23:24,080
And if you can put them on and take them off, then you can socialize with people and establish

311
00:23:24,080 --> 00:23:27,200
social connections and have satisfying interactions.

312
00:23:27,200 --> 00:23:33,540
But at the same time, if you let these things be rigid and domineering in your psychology,

313
00:23:33,540 --> 00:23:35,800
you're setting yourself up to get played.

314
00:23:35,800 --> 00:23:43,880
And so I'm just very skeptical of Pollyanna, like Hopium, as the doomers would say, Hopium

315
00:23:43,880 --> 00:23:49,080
critics, and I'm also skeptical of doomerism and in politics, particularly in this year

316
00:23:49,080 --> 00:23:54,880
when things are, it's such an intense pitch between Democrats and Republicans.

317
00:23:54,880 --> 00:23:56,760
I'm among the double haters.

318
00:23:56,760 --> 00:23:59,480
I don't like either one of these parties.

319
00:23:59,480 --> 00:24:02,000
I don't like any of the factions that are in charge of these parties.

320
00:24:02,000 --> 00:24:04,160
I have no interest in supporting any of them.

321
00:24:04,160 --> 00:24:07,960
I know that I will have to endure them, whichever one is elected.

322
00:24:07,960 --> 00:24:13,160
And I honestly don't care which one wins this particular election.

323
00:24:13,160 --> 00:24:15,880
I know that I won't like either one.

324
00:24:15,880 --> 00:24:21,760
But at the same time, I also know that they don't really control the society.

325
00:24:21,760 --> 00:24:27,960
The person who's in front of the cameras, behind the lectern or whatnot, it's a tough

326
00:24:27,960 --> 00:24:28,960
job.

327
00:24:28,960 --> 00:24:33,600
Their hair is going to go gray real fast if it's not already.

328
00:24:33,600 --> 00:24:37,740
But they're not in charge.

329
00:24:37,740 --> 00:24:47,920
So I think just living with radical uncertainty is necessary to be both rational and comfortable

330
00:24:47,920 --> 00:24:49,080
in the world that we live in.

331
00:24:49,080 --> 00:24:51,080
So I'll stop there.

332
00:24:51,080 --> 00:24:57,040
Yeah, I'll definitely agree with a lot of that, but not all of it.

333
00:24:57,040 --> 00:25:04,080
So what you left off with there, we have to learn to live with some level of uncertainty.

334
00:25:04,080 --> 00:25:12,400
I'm all for acknowledging the limits of our epistemics.

335
00:25:12,400 --> 00:25:16,680
We can talk about what's going to happen.

336
00:25:16,680 --> 00:25:21,000
Predictions about what's going to happen in 10 or 20 years are always wrong, right?

337
00:25:21,000 --> 00:25:30,600
So I don't think we should ever be too comfortable that we know exactly what's going to happen.

338
00:25:30,600 --> 00:25:34,880
But to go back to some of the specifics of what you were saying about the dangers of

339
00:25:34,880 --> 00:25:40,680
AIs and fishing is, I think, a great example because, yeah, fishing is going to be easier

340
00:25:40,680 --> 00:25:43,280
in a lot of ways.

341
00:25:43,280 --> 00:25:49,600
And that sucks because who's targeted by fishing tends to be older people that are a little

342
00:25:49,600 --> 00:25:52,640
bit less with it.

343
00:25:52,640 --> 00:26:01,000
They are a little bit more easily duped by messages claiming, oh, this is your grandson

344
00:26:01,000 --> 00:26:03,080
or whatever.

345
00:26:03,080 --> 00:26:04,840
And that's sad.

346
00:26:04,840 --> 00:26:09,120
There's going to be more tools for those kinds of people.

347
00:26:09,120 --> 00:26:12,680
There's also going to be more tools to detect and fight that kind of stuff.

348
00:26:12,680 --> 00:26:14,520
So there's sort of an arms race.

349
00:26:14,520 --> 00:26:16,600
And AI is on either side of that, right?

350
00:26:16,600 --> 00:26:20,480
One thing that a lot of people don't realize when we're talking about AI, they're exposed

351
00:26:20,480 --> 00:26:26,240
to chat GPT, the kind of sexy generative AI.

352
00:26:26,240 --> 00:26:33,040
The models that brought about this whole revolution in large language models, they were originally

353
00:26:33,040 --> 00:26:35,400
used in translation.

354
00:26:35,400 --> 00:26:39,560
And there's an encoder and a decoder.

355
00:26:39,560 --> 00:26:46,480
The encoder is the thing that looks at the text in the other language.

356
00:26:46,480 --> 00:26:48,480
So let's say we're going English to French.

357
00:26:48,480 --> 00:26:50,860
It kind of reads the text in English.

358
00:26:50,860 --> 00:26:58,400
And its job is to come up with a really good numeric representation of that text.

359
00:26:58,400 --> 00:27:03,400
Then the decoder's job is to decode that into the French text.

360
00:27:03,400 --> 00:27:06,280
So the generative component is that decoder.

361
00:27:06,280 --> 00:27:12,720
But that encoder, that really robust representation of language, that's actually really important.

362
00:27:12,720 --> 00:27:22,080
And I think if I had to take a guess, it's used much more in industry than the decoder.

363
00:27:22,080 --> 00:27:25,960
Because what it allows you to do is, hey, now we have a bunch of numbers that kind of

364
00:27:25,960 --> 00:27:30,840
represent this language in a really robust way, or this message, or whatever content

365
00:27:30,840 --> 00:27:31,840
it is.

366
00:27:31,840 --> 00:27:38,120
And that can be used with other machine learning models, classifiers, that are able to say,

367
00:27:38,120 --> 00:27:43,200
hey, we think this is phishing, or this is this kind of crappy message that we don't

368
00:27:43,200 --> 00:27:45,680
want to pass along.

369
00:27:45,680 --> 00:27:50,320
The clinician note of a sick patient, not a healthy patient, this patient needs more

370
00:27:50,320 --> 00:27:51,320
resources.

371
00:27:51,320 --> 00:27:57,520
There are all of these things that you can do when you have that robust numeric representation

372
00:27:57,520 --> 00:28:03,440
of text data that actually become really helpful things.

373
00:28:03,440 --> 00:28:08,480
But they don't get the limelight because the decoder, the generative component is so much

374
00:28:08,480 --> 00:28:09,480
sexier.

375
00:28:09,480 --> 00:28:14,040
So I think we're going to see an arms race in that space.

376
00:28:14,040 --> 00:28:15,160
Or maybe we won't, right?

377
00:28:15,160 --> 00:28:17,880
A lot of these arms races kind of happen behind the scenes.

378
00:28:17,880 --> 00:28:23,840
So I think an arms race like that is going to happen.

379
00:28:23,840 --> 00:28:31,580
Now I think you're right that, hey, we don't know necessarily what the biggest vectors

380
00:28:31,580 --> 00:28:35,920
of attack could be for a persuasive AI.

381
00:28:35,920 --> 00:28:40,520
The thing that lets me kind of sleep at night with all of this is, again, I feel like I'm

382
00:28:40,520 --> 00:28:43,880
harping on the same point, but there's a lot of smart people in the world and there's a

383
00:28:43,880 --> 00:28:46,720
lot of bad people in the world.

384
00:28:46,720 --> 00:28:55,160
And so luckily, if there's a way to do bad stuff, it's probably been attempted.

385
00:28:55,160 --> 00:28:57,200
And so we've probably learned from that.

386
00:28:57,200 --> 00:29:02,360
And a point we'll take in that, hey, at various times in the past, we've come close to the

387
00:29:02,360 --> 00:29:08,480
brink, so to speak, of nuclear war or whatever else.

388
00:29:08,480 --> 00:29:11,600
And the important point is we didn't.

389
00:29:11,600 --> 00:29:14,200
We had humans in the loop that made good judgments.

390
00:29:14,200 --> 00:29:17,720
And we continue to have humans in the loops.

391
00:29:17,720 --> 00:29:26,760
And we have to empower them and have the right tools in place and the right information as

392
00:29:26,760 --> 00:29:32,160
our information environment degrades, that becomes all the more important.

393
00:29:32,160 --> 00:29:37,960
If persuasive AIs become common, that's a known thing.

394
00:29:37,960 --> 00:29:42,120
The question people are going to ask themselves whenever they receive some information, if

395
00:29:42,120 --> 00:29:46,740
they're making some really important decision based off of it, is there some possibility

396
00:29:46,740 --> 00:29:53,140
that this is from some persuasive AI that's kind of spoofing this information?

397
00:29:53,140 --> 00:30:02,020
So I'm relatively optimistic about our ability as a society to adapt to these things, both

398
00:30:02,020 --> 00:30:12,160
through improvements in our tools and through our adaptability as humans to change as our

399
00:30:12,160 --> 00:30:17,440
information environment changes.

400
00:30:17,440 --> 00:30:25,800
I'm sure you've seen at least somebody reporting on the Facebook groups that are all just AI

401
00:30:25,800 --> 00:30:30,320
generated imagery that boomers on Facebook cannot tell.

402
00:30:30,320 --> 00:30:32,360
They just see a photograph.

403
00:30:32,360 --> 00:30:34,840
And my mother is 85 and she's like that.

404
00:30:34,840 --> 00:30:40,440
She can't tell an AI generated image from a photograph.

405
00:30:40,440 --> 00:30:44,180
And most people have never interacted with a large language model.

406
00:30:44,180 --> 00:30:50,200
They have no idea how lifelike they can be, how seemingly intelligent and responsive they

407
00:30:50,200 --> 00:30:51,280
are.

408
00:30:51,280 --> 00:30:59,720
So I think the idea that part of the arms race is just in our ability to be discerning

409
00:30:59,720 --> 00:31:01,440
and to detect what's incoming.

410
00:31:01,440 --> 00:31:06,800
Yeah, there'll be a portion of the population that keeps up for a while, but lots of people

411
00:31:06,800 --> 00:31:10,400
are already left behind and they're not going to catch up.

412
00:31:10,400 --> 00:31:15,880
My mother is never going to develop the visual sensitivity to look for six fingers on a hand.

413
00:31:15,880 --> 00:31:21,660
And even the distortions that were common in AI generated imagery a year ago are much

414
00:31:21,660 --> 00:31:23,920
less common now.

415
00:31:23,920 --> 00:31:30,680
And the staggered sort of morphing weird video is improving rapidly.

416
00:31:30,680 --> 00:31:33,880
So some people have already been left behind and won't catch up.

417
00:31:33,880 --> 00:31:39,880
And I'm pretty sure that within a few years, my ability to determine what is real and what

418
00:31:39,880 --> 00:31:42,880
is not will be unreliable.

419
00:31:42,880 --> 00:31:48,640
I'll spot stuff from time to time, but I won't know how much stuff I'm not spotting.

420
00:31:48,640 --> 00:31:51,240
I'll stop there.

421
00:31:51,240 --> 00:31:56,160
So I think that's a good argument to keep your mom away from the nuclear codes.

422
00:31:56,160 --> 00:31:59,600
And I think that's right.

423
00:31:59,600 --> 00:32:06,880
That, hey, there's going to be maybe some kind of bifurcation of people's ability to

424
00:32:06,880 --> 00:32:10,640
navigate this information environment, and it's already happened.

425
00:32:10,640 --> 00:32:19,560
You point out not just AI generated stuff on Facebook, but just random misinformation

426
00:32:19,560 --> 00:32:25,760
or stuff that's right, but kind of has a big slant.

427
00:32:25,760 --> 00:32:31,000
Those aspects of our information environment have always existed.

428
00:32:31,000 --> 00:32:33,240
There have been studies.

429
00:32:33,240 --> 00:32:43,440
I think if we look at present day as a snapshot of what kinds of things we might be worried

430
00:32:43,440 --> 00:32:46,280
about, misinformation, maybe that'll get worse.

431
00:32:46,280 --> 00:32:47,880
It's already so bad.

432
00:32:47,880 --> 00:32:53,240
I think the problem of misinformation is vastly overstated.

433
00:32:53,240 --> 00:32:59,040
I think that studies of misinformation have found it's actually a very small, limited

434
00:32:59,040 --> 00:33:04,080
amount of exposure that people have to it.

435
00:33:04,080 --> 00:33:08,120
Much more common is information presented with some kind of leaning.

436
00:33:08,120 --> 00:33:10,720
That's something that we've had forever.

437
00:33:10,720 --> 00:33:20,080
Everybody's received their information diet digested through some news anchor that has

438
00:33:20,080 --> 00:33:22,680
their personal biases.

439
00:33:22,680 --> 00:33:31,080
You're learning about it from the local news press or the editor back in the days when

440
00:33:31,080 --> 00:33:35,240
you had a much more limited selection of where you got your information from.

441
00:33:35,240 --> 00:33:42,240
Your local paper is the main source of information, and the editor is a racist or something.

442
00:33:42,240 --> 00:33:47,720
You end up with slants to the information that you consume.

443
00:33:47,720 --> 00:33:52,640
We've been dealing with that for a long time.

444
00:33:52,640 --> 00:34:00,880
There's going to be additional wrinkles being introduced to that as we change the culture

445
00:34:00,880 --> 00:34:05,080
and as technology evolves over time.

446
00:34:05,080 --> 00:34:09,240
I don't think it's fundamentally new.

447
00:34:09,240 --> 00:34:16,400
I think it has always been the case that people that consume from a wider set of sources and

448
00:34:16,400 --> 00:34:25,320
who have a better ability to identify not just experts but the right experts for the

449
00:34:25,320 --> 00:34:31,200
right kinds of information and are able to integrate all of that and evaluate information,

450
00:34:31,200 --> 00:34:34,640
that's always going to be an important skill.

451
00:34:34,640 --> 00:34:38,520
But I don't think it's a particularly new one.

452
00:34:38,520 --> 00:34:46,240
Even if we're not able to personally identify AI images, I assume that at some point AI

453
00:34:46,240 --> 00:34:50,760
image generation is going to get good enough that I won't be able to tell, no one will

454
00:34:50,760 --> 00:35:00,360
be able to tell, but we have very good image manipulation techniques already.

455
00:35:00,360 --> 00:35:03,440
What you have to do is rely on the experts.

456
00:35:03,440 --> 00:35:10,920
When an image hits the news networks that it turns out was doctored, journalists catch

457
00:35:10,920 --> 00:35:11,920
it.

458
00:35:11,920 --> 00:35:17,600
We have experts at looking at these kinds of things and trying to track down where they

459
00:35:17,600 --> 00:35:26,680
have come from that these institutions can help protect us from that information.

460
00:35:26,680 --> 00:35:30,880
So I think those institutions have always been important and are just going to continue

461
00:35:30,880 --> 00:35:31,880
to be so.

462
00:35:31,880 --> 00:35:38,600
And the skills of navigating these environments might change a little bit, but I think the

463
00:35:38,600 --> 00:35:45,840
overall game is staying the same.

464
00:35:45,840 --> 00:35:55,800
Sorry, I was looking for the name of a particular company.

465
00:35:55,800 --> 00:36:00,160
Are you familiar with Annie Jacobson?

466
00:36:00,160 --> 00:36:04,520
No, I'm not.

467
00:36:04,520 --> 00:36:08,560
She wrote a book called The Pentagon's Brain.

468
00:36:08,560 --> 00:36:13,000
She's written another book since then, but she's done a lot of research and had access

469
00:36:13,000 --> 00:36:16,480
and talked to high-ranking generals and whatnot.

470
00:36:16,480 --> 00:36:21,440
And regardless of what Silicon Valley wants and regardless of what the general populace

471
00:36:21,440 --> 00:36:25,640
wants and regardless of what any politician might say on the topic of autonomous weapons

472
00:36:25,640 --> 00:36:29,980
systems, the Pentagon wants them.

473
00:36:29,980 --> 00:36:39,200
And there's a company, and this is a very unfortunate trend, but this company is called

474
00:36:39,200 --> 00:36:42,920
Enderil, which is a Tolkien name.

475
00:36:42,920 --> 00:36:50,720
It's the name that Aragorn's sword was granted, you know, was bestowed with after it was reforged.

476
00:36:50,720 --> 00:36:56,160
And this is a company that is dedicated to building autonomous weapons systems, and they

477
00:36:56,160 --> 00:37:01,480
have just signed a big, you know, multi-billion dollar deal, and they're building a huge facility

478
00:37:01,480 --> 00:37:04,520
to create autonomous weapons.

479
00:37:04,520 --> 00:37:07,920
And it doesn't matter what the state of California, you know, writes into law.

480
00:37:07,920 --> 00:37:10,360
This is the Pentagon.

481
00:37:10,360 --> 00:37:13,800
They will get what they want to some extent.

482
00:37:13,800 --> 00:37:18,080
Now the Pentagon wants autonomous ships.

483
00:37:18,080 --> 00:37:21,880
They want lots and lots of small autonomous ships, but Congress doesn't like that.

484
00:37:21,880 --> 00:37:24,680
Congress likes, you know, aircraft carriers.

485
00:37:24,680 --> 00:37:28,120
Congress likes big, expensive things that are built in particular places, you know,

486
00:37:28,120 --> 00:37:30,640
that satisfy particular financial and political interests.

487
00:37:30,640 --> 00:37:34,760
So the Pentagon doesn't get everything they want, but there is no congressional bulwark

488
00:37:34,760 --> 00:37:37,200
against autonomous weapons systems.

489
00:37:37,200 --> 00:37:44,520
And we're seeing in Ukraine, you know, very innovative uses of drones, which are very,

490
00:37:44,520 --> 00:37:48,760
very cheap compared to the tanks, you know, and the other types of traditional military

491
00:37:48,760 --> 00:37:54,720
hardware, which they are, to put it in very sterile terms, neutralizing.

492
00:37:54,720 --> 00:38:01,100
So I think that the general point that I'm pushing back against is that we have experts.

493
00:38:01,100 --> 00:38:02,760
The experts know what's going on.

494
00:38:02,760 --> 00:38:08,220
They will anticipate what's happening, and they have no conflict of interest in terms

495
00:38:08,220 --> 00:38:13,000
of providing what it is we need as individuals and members of this society to live safe and

496
00:38:13,000 --> 00:38:15,160
happy lives.

497
00:38:15,160 --> 00:38:20,000
You know, I'm not appealing to a general sort of mistrust of government, which, you know,

498
00:38:20,000 --> 00:38:24,040
my 25-year-old libertarian self certainly would have pressed on.

499
00:38:24,040 --> 00:38:28,640
I'm just saying that there are—it's a complicated situation, and a lot of times we're just not

500
00:38:28,640 --> 00:38:34,480
going to know where resources need to be deployed and in what manner until things have gone

501
00:38:34,480 --> 00:38:37,560
catastrophically wrong for somebody.

502
00:38:37,560 --> 00:38:43,280
You know, we never had a nuclear war, but people living in Hiroshima and Nagasaki, you

503
00:38:43,280 --> 00:38:45,080
know, they experienced nuclear Armageddon.

504
00:38:45,080 --> 00:38:50,160
For some of them, you know, the last thing they ever saw was that bright flash of light.

505
00:38:50,160 --> 00:38:55,120
You know, some doomsday scenario—I mean, some doomsday scenario will befall all of us

506
00:38:55,120 --> 00:38:56,800
at some point, but it's individual.

507
00:38:56,800 --> 00:38:58,760
It doesn't come all at the same time.

508
00:38:58,760 --> 00:39:03,480
But you know, in limited regions, you know, maybe not in the U.S. because we have these

509
00:39:03,480 --> 00:39:07,160
two big oceans on either side of us and, you know, weak states to the north and south,

510
00:39:07,160 --> 00:39:16,200
but somewhere, pretty much every, like, horrible catastrophic scenario that anybody has imagined

511
00:39:16,200 --> 00:39:22,040
is likely to unfold for somebody, you know, for some local population somewhere.

512
00:39:22,040 --> 00:39:28,760
So this will be my last, you know, volley in the things aren't necessarily rosy because

513
00:39:28,760 --> 00:39:30,240
we have experts looking out for us.

514
00:39:30,240 --> 00:39:37,320
And in fact, you know, I like to quote—what's his name?

515
00:39:37,320 --> 00:39:45,640
Israeli, you know, cabinet member in the Nixon administration won a Nobel Prize, even though

516
00:39:45,640 --> 00:39:49,480
lots of people say he's—wow, who am I thinking of?

517
00:39:49,480 --> 00:39:51,200
Do you know who I'm thinking of?

518
00:39:51,200 --> 00:39:52,200
Sorry, man.

519
00:39:52,200 --> 00:39:54,240
This is before my time.

520
00:39:54,240 --> 00:39:57,160
This is killing me.

521
00:39:57,160 --> 00:40:00,320
I mean, this is a name that should—this is like—this should be on my tongue like,

522
00:40:00,320 --> 00:40:02,320
you know, Ronald Reagan.

523
00:40:02,320 --> 00:40:03,320
Anyway.

524
00:40:03,320 --> 00:40:04,320
Yeah.

525
00:40:04,320 --> 00:40:09,480
I'm also not American, so it's kind of—I'm Canadian, so I get a pass on, like, all things

526
00:40:09,480 --> 00:40:10,760
American history.

527
00:40:10,760 --> 00:40:12,120
Oh, you know this name.

528
00:40:12,120 --> 00:40:15,080
If I were to say it, you'd say, oh, yeah.

529
00:40:15,080 --> 00:40:20,240
Anyway, his definition of an expert is somebody who articulates the needs of power.

530
00:40:20,240 --> 00:40:25,840
You know, so you are not concerned particularly about misinformation, and neither am I.

531
00:40:25,840 --> 00:40:31,120
You know, exposure to a false narrative doesn't necessarily mean somebody's going to adopt

532
00:40:31,120 --> 00:40:32,120
it.

533
00:40:32,120 --> 00:40:35,280
People adopt false narratives because they answer to psychological needs that they have,

534
00:40:35,280 --> 00:40:39,680
and they will seek out the information, you know, or the misinformation, which validates

535
00:40:39,680 --> 00:40:44,200
their positions, which they hold, you know, for other reasons.

536
00:40:44,200 --> 00:40:46,920
But I'll just stop there.

537
00:40:46,920 --> 00:40:48,280
I'll stop there.

538
00:40:48,280 --> 00:40:49,280
That's the beginning.

539
00:40:49,280 --> 00:40:55,080
I see that I'm pointing myself down a road that doesn't resolve quickly, so go ahead.

540
00:40:55,080 --> 00:40:56,080
No problem.

541
00:40:56,080 --> 00:40:59,480
And, you know, this might be kind of annoying, but I kind of want to say—yeah, I think

542
00:40:59,480 --> 00:41:01,840
we're kind of saying the same thing.

543
00:41:01,840 --> 00:41:06,960
We might be kind of taking a slightly, like, you know, the optimist versus the pessimist

544
00:41:06,960 --> 00:41:11,480
version of it, but I think at the end of the day, like, yeah, I don't believe, hey, we

545
00:41:11,480 --> 00:41:16,120
have experts in place, so we should all just kind of, like, sit back.

546
00:41:16,120 --> 00:41:20,920
Nothing bad could ever happen, because at the end of the day, yeah, we need to—there

547
00:41:20,920 --> 00:41:21,920
is nuance.

548
00:41:21,920 --> 00:41:25,360
There are things that we need to be worried about.

549
00:41:25,360 --> 00:41:29,120
There are issues that are going to come up that are unforeseen.

550
00:41:29,120 --> 00:41:38,160
And so we do need to take a robust, nuanced understanding of this as citizens who vote,

551
00:41:38,160 --> 00:41:45,940
as people who can voice opinions about different legislation, and as people who might have

552
00:41:45,940 --> 00:41:51,760
some interaction with experts who do have, you know, speak to power.

553
00:41:51,760 --> 00:41:54,400
And so have some level of influence.

554
00:41:54,400 --> 00:42:00,120
You know, each of us can be playing a small role, and so I think we kind of have a duty

555
00:42:00,120 --> 00:42:04,960
in that sense to try to understand things and understand the nuance.

556
00:42:04,960 --> 00:42:08,960
What I'm pushing back against is the sort of blanket, something definitely is going

557
00:42:08,960 --> 00:42:14,560
to go wrong, and instead pushing for, hey, like, we actually have a lot of, like, important

558
00:42:14,560 --> 00:42:15,760
protections in place.

559
00:42:15,760 --> 00:42:18,640
Let's understand those protections.

560
00:42:18,640 --> 00:42:25,760
And yeah, if something comes up and it's ringing alarm bells, of course, we shouldn't just

561
00:42:25,760 --> 00:42:27,360
assume, yeah, it'll be fine.

562
00:42:27,360 --> 00:42:28,360
Don't worry about it.

563
00:42:28,360 --> 00:42:36,520
Yeah, there's every possibility of misaligned incentives between people making decisions

564
00:42:36,520 --> 00:42:44,960
about autonomous weapons, which isn't something I know a lot about, and the, you know, the

565
00:42:44,960 --> 00:42:52,960
well-being of the population of the United States or abroad.

566
00:42:52,960 --> 00:42:56,360
Will something catastrophic necessarily happen?

567
00:42:56,360 --> 00:42:57,360
I'm not so sure.

568
00:42:57,360 --> 00:43:00,080
Maybe it depends on your definition of catastrophic.

569
00:43:00,080 --> 00:43:02,380
We'll probably make some mistakes.

570
00:43:02,380 --> 00:43:03,380
Will people die?

571
00:43:03,380 --> 00:43:04,380
Hopefully not.

572
00:43:04,380 --> 00:43:07,700
I don't think it's a given that, like, yeah, definitely when mistakes are made, a bunch

573
00:43:07,700 --> 00:43:09,520
of people will die.

574
00:43:09,520 --> 00:43:13,600
With weapons, yeah, this is high-stakes stuff.

575
00:43:13,600 --> 00:43:16,680
Scary stuff is high-stakes stuff.

576
00:43:16,680 --> 00:43:20,520
When there's a gun malfunction, people die.

577
00:43:20,520 --> 00:43:25,440
So if that happens with autonomous weapons, sure, there might be something really bad

578
00:43:25,440 --> 00:43:26,440
that happens.

579
00:43:26,440 --> 00:43:28,280
That's pretty scary, right?

580
00:43:28,280 --> 00:43:29,280
Weapons are really scary.

581
00:43:29,280 --> 00:43:35,080
I don't like, like, this isn't an area I know much about, but yeah, it's not something that

582
00:43:35,080 --> 00:43:41,560
gives me the warm fuzzies thinking about all the new tech and new weapons.

583
00:43:41,560 --> 00:43:44,640
New weapons systems happen all the time, though.

584
00:43:44,640 --> 00:43:47,640
And they're all scary, frankly.

585
00:43:47,640 --> 00:43:50,680
And we wouldn't want something to go wrong with any of them.

586
00:43:50,680 --> 00:43:57,640
Or at least we wouldn't want something to go wrong when we're on this side of them.

587
00:43:57,640 --> 00:44:03,480
So I think you're right that we shouldn't just trust everything with a blanket, but

588
00:44:03,480 --> 00:44:07,240
we also shouldn't have a distrust of everything with a blanket.

589
00:44:07,240 --> 00:44:15,880
We should look at each issue individually and try to better understand what it is that

590
00:44:15,880 --> 00:44:19,440
policies are actually proposing, what are they actually supposed to protect us against,

591
00:44:19,440 --> 00:44:24,840
what are the actual risks of certain developing technologies, and what are the things that

592
00:44:24,840 --> 00:44:32,180
we, as citizens, can kind of vote for and advocate for to try to push things in a direction

593
00:44:32,180 --> 00:44:35,320
that we are more comfortable with.

594
00:44:35,320 --> 00:44:44,920
Well, first, I'm going to watch your face as I say this name, Henry Kissinger.

595
00:44:44,920 --> 00:44:47,320
Yes.

596
00:44:47,320 --> 00:44:56,560
Second, I am a curmudgeon about electoral democracy.

597
00:44:56,560 --> 00:44:59,260
Like, I live in Arkansas.

598
00:44:59,260 --> 00:45:01,160
It does not matter who I vote for.

599
00:45:01,160 --> 00:45:04,160
Trump is going to take Arkansas' six electoral votes, period.

600
00:45:04,160 --> 00:45:09,160
It does not matter at all whether I vote or not or who I vote for.

601
00:45:09,160 --> 00:45:10,160
Doesn't matter.

602
00:45:10,160 --> 00:45:11,160
It doesn't matter who I talk to.

603
00:45:11,160 --> 00:45:13,160
It doesn't matter what I say in public.

604
00:45:13,160 --> 00:45:16,560
Most of the people listening to me are not in Arkansas.

605
00:45:16,560 --> 00:45:21,360
And most of the people that I talk to in daily life, I don't get into political conversations

606
00:45:21,360 --> 00:45:26,520
with because they're not likely to go well.

607
00:45:26,520 --> 00:45:27,520
I want to set that aside.

608
00:45:27,520 --> 00:45:32,520
I mean, just getting into politics is probably not the most productive use of our time.

609
00:45:32,520 --> 00:45:38,920
I would much rather talk about other aspects of artificial intelligence and how it's likely

610
00:45:38,920 --> 00:45:40,120
to impact society.

611
00:45:40,120 --> 00:45:46,160
And I think the thing that most people can relate to is threats to the ability to make

612
00:45:46,160 --> 00:45:47,160
your livelihood.

613
00:45:47,160 --> 00:45:54,160
If an AI does something that you do faster, cheaper, more reliably, and never has any

614
00:45:54,160 --> 00:46:00,040
non-work-related drama come into the workplace because of them, there's going to be a strong

615
00:46:00,040 --> 00:46:01,040
temptation.

616
00:46:01,040 --> 00:46:08,320
So the question is, what sorts of jobs are susceptible to replacement?

617
00:46:08,320 --> 00:46:15,120
And I think one of the things that a lot of AI experts say is that no jobs are susceptible

618
00:46:15,120 --> 00:46:16,680
to being automated.

619
00:46:16,680 --> 00:46:17,680
Tasks.

620
00:46:17,680 --> 00:46:19,920
Tasks are susceptible to being automated.

621
00:46:19,920 --> 00:46:22,560
And jobs typically involve a lot of different tasks.

622
00:46:22,560 --> 00:46:27,480
So the more generalized and the more varied your job is, the safer you are probably, I

623
00:46:27,480 --> 00:46:32,880
used to think, that the more physical your job is, the safer you will be because it is

624
00:46:32,880 --> 00:46:38,440
mostly routine, repetitive, intellectual tasks which are easily automated via AI.

625
00:46:38,440 --> 00:46:44,000
But now humanoid robots are coming on strong after seemingly decades of being stagnant.

626
00:46:44,000 --> 00:46:47,360
Do you remember the ASIMO robot from around 2001, 2002?

627
00:46:47,360 --> 00:46:51,600
It's a Japanese robot made by Honda now.

628
00:46:51,600 --> 00:46:54,520
It was a very impressive little guy.

629
00:46:54,520 --> 00:46:55,520
And it had all the...

630
00:46:55,520 --> 00:46:58,040
Wait, I think I have a picture in my head now.

631
00:46:58,040 --> 00:47:00,600
Is it like a little white robot?

632
00:47:00,600 --> 00:47:01,600
Yeah.

633
00:47:01,600 --> 00:47:02,600
Okay.

634
00:47:02,600 --> 00:47:03,880
Like round head.

635
00:47:03,880 --> 00:47:04,880
Yeah.

636
00:47:04,880 --> 00:47:05,880
Very cute.

637
00:47:05,880 --> 00:47:06,880
Yeah.

638
00:47:06,880 --> 00:47:10,080
And there's some very impressive videos, but there's also, you know, blooper rolls of these

639
00:47:10,080 --> 00:47:11,720
things falling over and they can't get up.

640
00:47:11,720 --> 00:47:15,320
But if you look at the new, you know, the replacement for the Atlas robot from Boston

641
00:47:15,320 --> 00:47:21,640
Dynamics, the most famous video of it, it's lying flat on its back and it gets up very

642
00:47:21,640 --> 00:47:23,880
smoothly, you know, with just its legs.

643
00:47:23,880 --> 00:47:28,560
Like it twists its legs around so that its feet are planted by its hips and then it stands

644
00:47:28,560 --> 00:47:31,920
up in a way that no human ever could.

645
00:47:31,920 --> 00:47:37,000
So I think humanoid robots that, you know, can do what most humans do are much closer

646
00:47:37,000 --> 00:47:39,840
than we had previously thought.

647
00:47:39,840 --> 00:47:44,760
So this brings up the topic of universal basic income.

648
00:47:44,760 --> 00:47:49,200
And I have opinions which I will keep to myself for now and just throw the, you know, the

649
00:47:49,200 --> 00:47:50,200
topic down.

650
00:47:50,200 --> 00:47:52,200
UBI, what do you think?

651
00:47:52,200 --> 00:48:00,600
Well, let me address the premises here first before talking about UBI specifically.

652
00:48:00,600 --> 00:48:10,260
So I think it is easy to overestimate the impact on employment of AI, robotics, et cetera.

653
00:48:10,260 --> 00:48:13,840
We have been through this before, right?

654
00:48:13,840 --> 00:48:14,840
Look at the 1800s.

655
00:48:14,840 --> 00:48:19,220
It's something like 80% of the population was farmers, right?

656
00:48:19,220 --> 00:48:21,520
How many people work in agriculture now?

657
00:48:21,520 --> 00:48:22,520
Way smaller percentage.

658
00:48:22,520 --> 00:48:24,960
Does that mean we lost all of those jobs?

659
00:48:24,960 --> 00:48:26,160
Well, no.

660
00:48:26,160 --> 00:48:30,080
We just like learned how to do it more efficiently with fewer people.

661
00:48:30,080 --> 00:48:32,120
We got better at it.

662
00:48:32,120 --> 00:48:36,840
And that means we could go on and move on to do other things.

663
00:48:36,840 --> 00:48:40,080
That's been the history of technology, right?

664
00:48:40,080 --> 00:48:49,280
Where does the term oh, shoot, I'm blanking on the term.

665
00:48:49,280 --> 00:48:53,080
Luddite, where does the term Luddite come from?

666
00:48:53,080 --> 00:48:56,360
These are people that were against, was it the loom?

667
00:48:56,360 --> 00:49:02,160
Some very basic fiber technology for weaving technology.

668
00:49:02,160 --> 00:49:08,880
I should say fiber technology sounds like I'm talking about fiber optics or something.

669
00:49:08,880 --> 00:49:10,600
So they were against that.

670
00:49:10,600 --> 00:49:11,600
Why?

671
00:49:11,600 --> 00:49:17,440
Well, because it was going to put people out of jobs and they were against that just like

672
00:49:17,440 --> 00:49:20,600
we're against AI today.

673
00:49:20,600 --> 00:49:26,980
So the rate at which industry adopts new technology tends to be very slow.

674
00:49:26,980 --> 00:49:30,480
So I think what we'll see is a very gradual transition.

675
00:49:30,480 --> 00:49:31,560
You already see some of this.

676
00:49:31,560 --> 00:49:37,280
So for example, like tons of companies have already put together, you know, we had crappy

677
00:49:37,280 --> 00:49:38,720
chat bots before.

678
00:49:38,720 --> 00:49:41,280
Hey, LLMs make our chat bots a little bit better.

679
00:49:41,280 --> 00:49:43,320
Let's build new systems based off of those.

680
00:49:43,320 --> 00:49:47,600
So our chat bots have gotten a little better.

681
00:49:47,600 --> 00:49:53,400
The internal stats I've seen from the company I've worked at, these haven't really cut down

682
00:49:53,400 --> 00:49:57,800
the number of questions that go along to support members, but hey, maybe we'll get better at

683
00:49:57,800 --> 00:49:58,800
it.

684
00:49:58,800 --> 00:50:03,800
And hey, maybe we won't need to have so many live humans responding to questions anymore.

685
00:50:03,800 --> 00:50:06,000
That would be great, right?

686
00:50:06,000 --> 00:50:11,880
Like, those people can then go on to do other things that are more productive.

687
00:50:11,880 --> 00:50:21,080
We have fewer people doing the same, to do the same amount of output, right?

688
00:50:21,080 --> 00:50:23,400
And that's how we grow the economy.

689
00:50:23,400 --> 00:50:27,220
That's how we all become better off.

690
00:50:27,220 --> 00:50:35,240
So unemployment is low now, and we have these AI technologies.

691
00:50:35,240 --> 00:50:39,480
I don't see a reason to think that we're going to see a sudden sharp shift.

692
00:50:39,480 --> 00:50:47,600
I think we'll see a slow adoption of AI technologies into different things that make people more

693
00:50:47,600 --> 00:50:54,120
productive and that that will free up resources, grow the economy as those people move on to

694
00:50:54,120 --> 00:51:00,600
do other things that AI is not capable of doing.

695
00:51:00,600 --> 00:51:01,600
Maybe I'm wrong.

696
00:51:01,600 --> 00:51:08,520
Maybe you're right, but like, yeah, actually, if we look 10, 20 years down the line, there's

697
00:51:08,520 --> 00:51:16,160
going to be massive unemployment because anything useful that we can think of the vast majority

698
00:51:16,160 --> 00:51:22,720
of humans doing, we can just do with some combination of robots and AI.

699
00:51:22,720 --> 00:51:25,920
That would be great, right?

700
00:51:25,920 --> 00:51:30,840
Like you said, at that point, yeah, UBI becomes the obvious option.

701
00:51:30,840 --> 00:51:40,360
If you are not able to contribute to an economy because you just don't have skills or abilities

702
00:51:40,360 --> 00:51:50,320
that are over and above what can be done much more cheaply by a robot or AI, yeah, let's

703
00:51:50,320 --> 00:51:54,680
open things up so that you can do things that are maybe not economically valuable, but are

704
00:51:54,680 --> 00:51:58,560
valuable in other ways.

705
00:51:58,560 --> 00:52:03,380
And that expands the things that you're capable of.

706
00:52:03,380 --> 00:52:10,520
Maybe it's you go into art and become more involved in your local community doing art,

707
00:52:10,520 --> 00:52:11,520
right?

708
00:52:11,520 --> 00:52:17,960
I recognize that it's kind of funny to talk about that given the like, oh, yeah, I can

709
00:52:17,960 --> 00:52:18,960
do art now.

710
00:52:18,960 --> 00:52:19,960
Like, why do it?

711
00:52:19,960 --> 00:52:22,680
Well, no, it's a form of expression, right?

712
00:52:22,680 --> 00:52:27,720
If we take away the economic component of it and you're just doing it as a form of expression,

713
00:52:27,720 --> 00:52:33,200
suddenly there's more reason to do art and you can connect with your fellow humans on

714
00:52:33,200 --> 00:52:36,880
pursuing that form of art.

715
00:52:36,880 --> 00:52:42,240
And so if that's kind of the world we live in and if we have the same level or even greater

716
00:52:42,240 --> 00:52:46,720
economic output than we otherwise would have, we can afford to spread around that wealth

717
00:52:46,720 --> 00:52:48,000
more, right?

718
00:52:48,000 --> 00:52:53,520
If we're doing the same amount of, we're building the same amount of stuff as we do now, but

719
00:52:53,520 --> 00:53:01,360
only using 25% of the population, hey, let's spread around that stuff to the full population

720
00:53:01,360 --> 00:53:10,020
using a tool like universal basic income or another like equivalent policy.

721
00:53:10,020 --> 00:53:13,760
So this is a rote conversation.

722
00:53:13,760 --> 00:53:16,640
You've said something that I've heard many times before, so I'm going to give the standard

723
00:53:16,640 --> 00:53:17,640
response to it.

724
00:53:17,640 --> 00:53:21,160
I'm going to try to keep it as short as possible though, because it's not terribly interesting

725
00:53:21,160 --> 00:53:26,240
to me to have the 10th or 15th iteration of this exchange.

726
00:53:26,240 --> 00:53:31,600
But the Industrial Revolution is a bad example.

727
00:53:31,600 --> 00:53:38,000
One, there was something called the enclosure movement in the UK.

728
00:53:38,000 --> 00:53:40,280
It wasn't the UK at the time, it was England.

729
00:53:40,280 --> 00:53:46,880
But there used to be a lot of people who lived agricultural lives and they depended on common

730
00:53:46,880 --> 00:53:51,060
grazing areas, and this is directly related to textile.

731
00:53:51,060 --> 00:53:56,560
Sheep would be grazed, but the sheep belonged to individual families and small communities

732
00:53:56,560 --> 00:53:59,680
and they shared agriculture or they shared grazing lands.

733
00:53:59,680 --> 00:54:07,840
And then the turning the wool into fiber that could be made into cloth and other textiles

734
00:54:07,840 --> 00:54:10,760
was a skilled occupation.

735
00:54:10,760 --> 00:54:17,520
And then more importantly, the weaving of those fibers into cloth was also a skilled occupation.

736
00:54:17,520 --> 00:54:26,440
And the looms that the Luddites were objecting to were taking a skilled job that commanded

737
00:54:26,440 --> 00:54:31,080
a high price and turning it into a job that was not skilled and which could be done by

738
00:54:31,080 --> 00:54:34,680
children, which in fact was often done by children.

739
00:54:34,680 --> 00:54:40,920
Children were a preferred employee in early factories because they could not stand up

740
00:54:40,920 --> 00:54:42,380
for their rights.

741
00:54:42,380 --> 00:54:46,400
You didn't really even have to pay them, you could just bully them into doing the work.

742
00:54:46,400 --> 00:54:53,140
And that transition from agricultural people depending and making use of shared common

743
00:54:53,140 --> 00:54:58,620
lands and then using that to produce a resource that they could then turn into a value-added

744
00:54:58,620 --> 00:55:03,780
product that had a high premium because of the skill involved in it.

745
00:55:03,780 --> 00:55:07,000
The people got forced off the land by the Enclosure Acts.

746
00:55:07,000 --> 00:55:10,000
They couldn't live and work where they had lived and worked for generations.

747
00:55:10,000 --> 00:55:14,040
They had to go into the cities where they weren't established, they didn't have any

748
00:55:14,040 --> 00:55:17,560
status there, forced into slums.

749
00:55:17,560 --> 00:55:24,520
And then their jobs got de-skilled and it didn't all work out for them.

750
00:55:24,520 --> 00:55:29,480
We 100 years, 200 years later can look back and say, look, it all worked out fine, but

751
00:55:29,480 --> 00:55:31,380
it didn't work out fine for them.

752
00:55:31,380 --> 00:55:33,380
They were fucked forever.

753
00:55:33,380 --> 00:55:35,540
They didn't get programming jobs.

754
00:55:35,540 --> 00:55:37,380
They didn't get bus driving jobs.

755
00:55:37,380 --> 00:55:42,000
They lived and died in poverty, in squalor with no representation.

756
00:55:42,000 --> 00:55:49,040
And the Luddites were absolutely correct in their time that they were getting fucked

757
00:55:49,040 --> 00:55:51,040
and they did get fucked.

758
00:55:51,040 --> 00:55:52,960
And it wasn't okay.

759
00:55:52,960 --> 00:55:58,040
It's okay to us now looking back where we can abstract their suffering, but it's not

760
00:55:58,040 --> 00:55:59,040
okay.

761
00:55:59,040 --> 00:56:07,440
In terms of the transition later from mostly manual farming to mechanized farming, pre-mechanized

762
00:56:07,440 --> 00:56:09,480
farming sucks.

763
00:56:09,480 --> 00:56:14,320
The reason we've had slavery for thousands of years is because nobody wants to do it.

764
00:56:14,320 --> 00:56:19,080
So you have to take certain people and say, okay, you go out in the field and you do that

765
00:56:19,080 --> 00:56:23,640
hard shitty work because I don't want to do it.

766
00:56:23,640 --> 00:56:24,640
That's the basis of slavery.

767
00:56:24,640 --> 00:56:28,320
So the change to mechanized farming is a godsend.

768
00:56:28,320 --> 00:56:37,800
At the same time, literally at the same time, we were transitioning from horse-drawn transportation

769
00:56:37,800 --> 00:56:40,800
to mechanized transportation.

770
00:56:40,800 --> 00:56:47,060
And there's a line of reasoning says, look, don't look at the drivers to say what our

771
00:56:47,060 --> 00:56:53,840
fate will be as AI takes over more and more of the job tasks in the society.

772
00:56:53,840 --> 00:56:55,760
Look at the horses.

773
00:56:55,760 --> 00:57:01,200
The number of horses dropped dramatically after the introduction of the automobile because

774
00:57:01,200 --> 00:57:02,740
we didn't need them.

775
00:57:02,740 --> 00:57:11,080
And so you can say human beings have value in and of themselves.

776
00:57:11,080 --> 00:57:12,720
Their rights should be respected.

777
00:57:12,720 --> 00:57:15,840
Their welfare should be considered.

778
00:57:15,840 --> 00:57:20,120
Should be, but in general, that isn't how things have played out historically.

779
00:57:20,120 --> 00:57:25,020
There has been portions of the population that needed a workforce and so had a reason

780
00:57:25,020 --> 00:57:28,880
to provide for at least basic provisioning for their workforce.

781
00:57:28,880 --> 00:57:33,880
But when that workforce becomes unnecessary, this is something Brett Weinstein said on

782
00:57:33,880 --> 00:57:39,960
a recent podcast, if you decide, well, we're going to let the robots do the work and we're

783
00:57:39,960 --> 00:57:42,280
going to take care of people.

784
00:57:42,280 --> 00:57:46,960
And one way to do that is just by giving everybody money so that they can buy the things that

785
00:57:46,960 --> 00:57:47,960
they need.

786
00:57:47,960 --> 00:57:51,960
That'll be a very short lived period because the people who perceive themselves as the

787
00:57:51,960 --> 00:57:57,120
ones creating the value will be resentful to see their value get redistributed to the

788
00:57:57,120 --> 00:58:00,680
useless eaters, to the people who aren't doing anything.

789
00:58:00,680 --> 00:58:09,520
And I guess it comes down to a basic question of where you sit emotionally in terms of the

790
00:58:09,520 --> 00:58:13,280
manipulation of the masses by an elite.

791
00:58:13,280 --> 00:58:20,820
And I think if you look back in time, you always see people justifying the suffering

792
00:58:20,820 --> 00:58:24,520
of others and invoking noble sounding principles in order to do it.

793
00:58:24,520 --> 00:58:26,880
Now in the past, these were typically religious principles.

794
00:58:26,880 --> 00:58:34,960
And in the present, they're going to sound compassionate, they're going to sound noble,

795
00:58:34,960 --> 00:58:39,160
they are going to sound humane when they're coming out of the mouths of the experts.

796
00:58:39,160 --> 00:58:43,440
And remember, the expert is somebody who articulates the needs of power.

797
00:58:43,440 --> 00:58:51,040
But they're going to basically be justifying the fact that we have increasing concentrations

798
00:58:51,040 --> 00:58:56,240
of wealth, power, opportunity, and more and more people who don't really, they're not

799
00:58:56,240 --> 00:58:57,600
needed.

800
00:58:57,600 --> 00:59:01,400
And just because they're human beings doesn't mean we're actually as a society going to

801
00:59:01,400 --> 00:59:07,200
provide for their welfare or the maximization of their potential.

802
00:59:07,200 --> 00:59:12,320
Just look at the streets of San Francisco, Portland, Oregon, Los Angeles.

803
00:59:12,320 --> 00:59:17,320
Like I was in Los Angeles in 2019, I guess was the last time I was there.

804
00:59:17,320 --> 00:59:21,080
I was shocked at the number of tents just everywhere.

805
00:59:21,080 --> 00:59:26,000
Like I drove through Skid Row, I saw the worst of it, but everywhere, everywhere in LA you

806
00:59:26,000 --> 00:59:30,520
go there's tents on the sidewalk, people just living on the street.

807
00:59:30,520 --> 00:59:34,760
And I can't say that, you know, I can't look at that and say, yeah, we're going in the

808
00:59:34,760 --> 00:59:38,800
right direction, that this is a positive development that, you know, clearly these people have

809
00:59:38,800 --> 00:59:43,040
been liberated from the need to engage in, you know, coerced labor.

810
00:59:43,040 --> 00:59:44,040
Good for them.

811
00:59:44,040 --> 00:59:45,920
Not so good for them.

812
00:59:45,920 --> 00:59:47,960
So, yeah, I'll stop.

813
00:59:47,960 --> 00:59:53,960
Obviously this is a rehearsed rant and I said I'd keep it short and I failed.

814
00:59:53,960 --> 00:59:58,600
Yeah, just to quickly react to some of that.

815
00:59:58,600 --> 01:00:14,480
So, mechanization of agriculture versus the leadites and more automation almost of weaving.

816
01:00:14,480 --> 01:00:16,060
Two examples.

817
01:00:16,060 --> 01:00:21,600
The thing they have in common is, hey, we were able to increase productivity with less

818
01:00:21,600 --> 01:00:22,960
labor.

819
01:00:22,960 --> 01:00:27,500
The thing that we they don't have in common is in one case we seem to have an increase

820
01:00:27,500 --> 01:00:33,540
in quality of life and the other a decrease for those in that particular historical period

821
01:00:33,540 --> 01:00:34,880
that had that transition.

822
01:00:34,880 --> 01:00:39,080
Yeah, I think it's totally legitimate to worry about, hey, what's that even if we have some

823
01:00:39,080 --> 01:00:44,080
like long term plan that makes things look good because we'll be more productive as a

824
01:00:44,080 --> 01:00:46,440
society, what does that transition look like?

825
01:00:46,440 --> 01:00:49,160
I think that's a good thing to ask.

826
01:00:49,160 --> 01:00:56,240
Luckily, we have things like lots of rights of workers now.

827
01:00:56,240 --> 01:01:01,980
We don't have quite the same loosey goosey like you can force children to work in factories

828
01:01:01,980 --> 01:01:08,480
kind of stuff that we had back when the leadites were relevant.

829
01:01:08,480 --> 01:01:13,440
But you do still have this worry that, yeah, what about the people that potentially are

830
01:01:13,440 --> 01:01:18,360
going to be losing their livelihoods, something like a skill that they have developed through

831
01:01:18,360 --> 01:01:24,760
their whole life, they have their identity tied to this job and they might learn that's

832
01:01:24,760 --> 01:01:26,120
just not relevant anymore.

833
01:01:26,120 --> 01:01:27,560
You're not creating value.

834
01:01:27,560 --> 01:01:31,560
I think that's a serious risk of psychological harm and I think that's something that we

835
01:01:31,560 --> 01:01:37,120
should, if we get to that point, which again, I do not think is a given, I don't think it's

836
01:01:37,120 --> 01:01:44,560
at all a given that developments in AI inevitably are going to lead to mass unemployment.

837
01:01:44,560 --> 01:01:47,840
I think that's a massive leap.

838
01:01:47,840 --> 01:01:53,460
But if we do get to that, yeah, we have to do some thinking as a society, how do we set

839
01:01:53,460 --> 01:01:57,560
people up to be supported through that transition?

840
01:01:57,560 --> 01:01:59,880
How do we help them deal with that?

841
01:01:59,880 --> 01:02:09,040
Either by, I mean, one thing I'm generally a big advocate for is just investing in human

842
01:02:09,040 --> 01:02:10,040
capital.

843
01:02:10,040 --> 01:02:14,600
So how can we help people retrain for other skills?

844
01:02:14,600 --> 01:02:24,120
Alternatively, how do we help them adapt to some other life, like where they're kind of

845
01:02:24,120 --> 01:02:32,160
redefining their identity to not be something that is economically productive in the same

846
01:02:32,160 --> 01:02:37,260
way that they've conceptualized through their lives, but is useful in some other manner.

847
01:02:37,260 --> 01:02:46,200
That's all very speculative future-facing stuff with regards to, well, horses became

848
01:02:46,200 --> 01:02:50,240
a lot less common as we mechanized agriculture.

849
01:02:50,240 --> 01:02:55,220
I'm not sure that the analogy there holds up super well because the reasons that we

850
01:02:55,220 --> 01:03:02,960
create more humans are not the same reasons that we create more horses.

851
01:03:02,960 --> 01:03:08,360
Please decide to make babies for reasons that I think are completely divorced from, oh,

852
01:03:08,360 --> 01:03:14,400
yeah, this is going to be a new worker for our economy.

853
01:03:14,400 --> 01:03:23,000
Point well taken that there are larger societal forces that can influence the cultural discourse

854
01:03:23,000 --> 01:03:29,340
and therefore influence those discussions happening at the family level.

855
01:03:29,340 --> 01:03:36,840
But I'm not so pessimistic about the result of those being so extreme that, yeah, we're

856
01:03:36,840 --> 01:03:48,840
just going to stop breeding the masses in this hypothetical future.

857
01:03:48,840 --> 01:03:54,760
But is UBI kind of a sustainable long-term policy?

858
01:03:54,760 --> 01:04:02,040
I think that depends a lot on the dynamics of exactly how we set it up, exactly how we

859
01:04:02,040 --> 01:04:03,320
evolve towards it.

860
01:04:03,320 --> 01:04:09,860
And there's enough speculation there that my only claim is, yeah, if somehow we end

861
01:04:09,860 --> 01:04:16,680
up in this future scenario where we have mass unemployment and people just are not being

862
01:04:16,680 --> 01:04:30,200
economically valuable, not because they, well, because AI can do everything, which I'm not

863
01:04:30,200 --> 01:04:32,680
even sure that that's a really plausible scenario.

864
01:04:32,680 --> 01:04:36,640
But if it did somehow happen, yeah, UBI makes sense.

865
01:04:36,640 --> 01:04:43,000
I'm not kind of advocating for it being a likely way for us to go.

866
01:04:43,000 --> 01:04:56,840
Because I think there's a lot of thinking about people's psychology around politics

867
01:04:56,840 --> 01:04:58,480
in general.

868
01:04:58,480 --> 01:05:03,520
I think it is an unlikely policy to ever be adopted in a widespread way.

869
01:05:03,520 --> 01:05:10,320
And I think there's a lot of societal implications that we don't really understand that would

870
01:05:10,320 --> 01:05:18,000
take place if we had a large even majority of the population on something like UBI.

871
01:05:18,000 --> 01:05:25,260
So I'm just saying from a kind of naive perspective, yeah, of course, like, hey, if we have a lot

872
01:05:25,260 --> 01:05:30,120
of stuff and we have a lot of people that aren't able to engage in the economy, it makes

873
01:05:30,120 --> 01:05:32,600
sense to spread that stuff around.

874
01:05:32,600 --> 01:05:35,760
All right.

875
01:05:35,760 --> 01:05:36,760
That was Tommy Blanchard.

876
01:05:36,760 --> 01:05:39,600
We went on to talk for another hour after that.

877
01:05:39,600 --> 01:05:41,680
That will be behind the paywall.

878
01:05:41,680 --> 01:05:47,360
So if you're listening on Patreon or Substack, you should be able to find the paywall portion

879
01:05:47,360 --> 01:05:48,360
pretty easily.

880
01:05:48,360 --> 01:05:52,240
If you're listening on YouTube, well, you'll either need to go to my Substack.

881
01:05:52,240 --> 01:05:55,600
There'll be a link in the description of the video or to my Patreon.

882
01:05:55,600 --> 01:05:59,240
I would suggest Substack, but either way is fine.

883
01:05:59,240 --> 01:06:01,240
Anyway, that was Tommy Blanchard.

884
01:06:01,240 --> 01:06:03,960
If you want to hear more from him, let me know and I'll get him back and we can talk

885
01:06:03,960 --> 01:06:05,320
about science fiction.

886
01:06:05,320 --> 01:06:06,560
All right.

887
01:06:06,560 --> 01:06:08,120
That's all for this episode.

888
01:06:08,120 --> 01:06:12,400
I'm...sometimes I will go on a rant and monologue here at the end.

889
01:06:12,400 --> 01:06:14,800
I'm not going to do that this time.

890
01:06:14,800 --> 01:06:17,440
So thanks for listening.

891
01:06:17,440 --> 01:06:38,800
Talk to you again soon.