1
00:00:02,480 --> 00:00:10,680
Speaker 1: Bloomberg Audio Studios, Podcasts, Radio News.

2
00:00:17,920 --> 00:00:21,840
Speaker 2: Hello and welcome to another episode of the Odd Lots Podcast.

3
00:00:21,920 --> 00:00:24,240
Speaker 1: I'm Joe Wisenthal and I'm Tracy Alloway.

4
00:00:24,520 --> 00:00:27,680
Speaker 2: Tracy, I feel like AI is a great thing for

5
00:00:28,000 --> 00:00:30,280
Speaker 2: anyone who wants to have an opinion on anything. It's

6
00:00:30,280 --> 00:00:34,080
Speaker 2: like this blank canvas out there in which any idea

7
00:00:34,159 --> 00:00:37,760
Speaker 2: you have it. It's just a great moment for pontificators

8
00:00:37,800 --> 00:00:38,280
Speaker 2: in general.

9
00:00:38,400 --> 00:00:40,680
Speaker 1: Well, not only can you hang a bunch of different

10
00:00:40,680 --> 00:00:43,640
Speaker 1: opinions on it, but it can generate those opinions for you.

11
00:00:45,080 --> 00:00:46,120
Speaker 3: Yeah, you can, that's right.

12
00:00:46,159 --> 00:00:48,560
Speaker 2: Like you can just go to chat GPT and say

13
00:00:48,800 --> 00:00:51,360
Speaker 2: which jobs are going to be lost thanks to you,

14
00:00:51,400 --> 00:00:54,960
Speaker 2: and it'll like spew some answer forward based on the

15
00:00:55,040 --> 00:00:58,440
Speaker 2: collective wisdom of trillions of words that people have typed

16
00:00:58,440 --> 00:00:59,160
Speaker 2: over the years.

17
00:00:59,360 --> 00:01:03,400
Speaker 1: I do think, though, if we're talking about one opinion,

18
00:01:03,600 --> 00:01:06,920
Speaker 1: in particular, the dominant opinion at this point in time,

19
00:01:07,200 --> 00:01:10,360
Speaker 1: it does feel like there's a lot of nervousness about

20
00:01:10,400 --> 00:01:13,319
Speaker 1: this new technology and what exactly it means for the economy,

21
00:01:13,400 --> 00:01:16,520
Speaker 1: what specifically it means for jobs. And so you see

22
00:01:16,520 --> 00:01:19,520
Speaker 1: all these headlines that AI is going to lead to

23
00:01:19,560 --> 00:01:22,560
Speaker 1: a bunch of job losses, that it's going to basically

24
00:01:22,600 --> 00:01:26,280
Speaker 1: be a new technological revolution that plays out very similarly

25
00:01:26,520 --> 00:01:30,200
Speaker 1: to the computer revolution that led to the destruction of

26
00:01:30,240 --> 00:01:33,039
Speaker 1: a bunch of sort of middle office jobs, or the

27
00:01:33,120 --> 00:01:38,160
Speaker 1: industrial revolution that led to a loss of skilled artisan jobs.

28
00:01:38,520 --> 00:01:42,520
Speaker 1: And we've seen some hints of that, to be fair,

29
00:01:42,680 --> 00:01:45,560
Speaker 1: So I'm thinking back to last year. I think it

30
00:01:45,680 --> 00:01:50,120
Speaker 1: was in the summer, maybe in June, and the Challenger

31
00:01:50,240 --> 00:01:53,280
Speaker 1: Jobs Report came out and for the first time ever,

32
00:01:53,320 --> 00:01:57,000
Speaker 1: they included a line about job losses stemming from AI.

33
00:01:57,840 --> 00:02:00,520
Speaker 2: Yeah, although I'm just going to say right here then,

34
00:02:00,560 --> 00:02:03,559
Speaker 2: I think when a company lays off workers and says

35
00:02:03,560 --> 00:02:06,400
Speaker 2: it's due to AI, I still have this assumption that

36
00:02:06,480 --> 00:02:08,760
Speaker 2: it's like, we're doing badly, so we're gonna put a

37
00:02:08,760 --> 00:02:11,320
Speaker 2: positive spin on it by making it seem as though

38
00:02:11,320 --> 00:02:14,880
Speaker 2: our layoffs are the result of some internal productivity breakthrough

39
00:02:14,919 --> 00:02:17,280
Speaker 2: that we're getting from a chatbot. So, like, I don't

40
00:02:17,400 --> 00:02:19,840
Speaker 2: quite believe it, but I think.

41
00:02:19,639 --> 00:02:23,280
Speaker 1: That's totally fair. That's totally fair. But I think clearly

42
00:02:23,320 --> 00:02:25,600
Speaker 1: this is something people are paying attention to. You are

43
00:02:25,760 --> 00:02:28,040
Speaker 1: starting to see some of the economic reports sort of

44
00:02:28,240 --> 00:02:31,760
Speaker 1: break this down. At least the Challenger report if not

45
00:02:32,000 --> 00:02:34,840
Speaker 1: like the BLS and things like that. So there is

46
00:02:35,000 --> 00:02:39,120
Speaker 1: this idea hovering over the economy at the moment, which is, Okay,

47
00:02:39,200 --> 00:02:42,440
Speaker 1: maybe AI will be great for productivity, we'll get that boost,

48
00:02:42,800 --> 00:02:44,720
Speaker 1: but what does it mean for jobs? Right?

49
00:02:44,880 --> 00:02:48,480
Speaker 2: Basically everyone in any realm loses their job and is

50
00:02:48,480 --> 00:02:51,160
Speaker 2: on the UBI drip, and only Sam Altman is the

51
00:02:51,240 --> 00:02:54,040
Speaker 2: last person who is employed, is it? But like, I

52
00:02:54,040 --> 00:02:56,120
Speaker 2: don't know, I get freaked out, Like it's pretty good,

53
00:02:56,280 --> 00:03:00,000
Speaker 2: Like there are many I use AI almost chatbots all

54
00:03:00,080 --> 00:03:01,640
Speaker 2: the time in my work, and it's like, well, maybe

55
00:03:01,639 --> 00:03:04,360
Speaker 2: it could one day be a better host than myself

56
00:03:04,680 --> 00:03:08,400
Speaker 2: for a podcast. It seems possible to me. I am anxious.

57
00:03:08,520 --> 00:03:11,920
Speaker 2: Of course, people also like to project onto their perceived

58
00:03:11,919 --> 00:03:14,920
Speaker 2: ideological enemies. That's like, oh, all you English majors are

59
00:03:14,919 --> 00:03:17,040
Speaker 2: going to lose your jobs hahaha, And then the English

60
00:03:17,040 --> 00:03:19,800
Speaker 2: majors all go, all you coders are going to lose

61
00:03:19,880 --> 00:03:22,120
Speaker 2: your jobs, and you're going to need English majors there.

62
00:03:22,160 --> 00:03:23,359
Speaker 3: It's just an endless thing.

63
00:03:23,400 --> 00:03:25,400
Speaker 2: And actually I think I tune most of it out

64
00:03:25,440 --> 00:03:28,639
Speaker 2: because it's so ambiguous in my view where this technology

65
00:03:28,680 --> 00:03:30,919
Speaker 2: is going that there are very few people I even

66
00:03:30,960 --> 00:03:34,480
Speaker 2: want to hear from on the topic, because I think

67
00:03:34,480 --> 00:03:37,520
Speaker 2: it's just so there's so much extreme uncertainty.

68
00:03:37,040 --> 00:03:41,280
Speaker 1: Still extreme uncertainty. As you mentioned, people kind of harness

69
00:03:41,440 --> 00:03:46,280
Speaker 1: it to further their own biases or arguments. But you're right,

70
00:03:46,400 --> 00:03:48,360
Speaker 1: there are people who are good on this topic, and

71
00:03:48,360 --> 00:03:49,600
Speaker 1: we're about to speak to one of them.

72
00:03:50,080 --> 00:03:50,960
Speaker 3: That's exactly right.

73
00:03:51,000 --> 00:03:54,800
Speaker 2: So last month, there was this really interesting headline that

74
00:03:54,880 --> 00:03:58,640
Speaker 2: I saw in a Nuema magazine and it sort of

75
00:03:58,680 --> 00:04:02,240
Speaker 2: felt like this sort of like provocative, maybe clickbaity type

76
00:04:02,280 --> 00:04:05,840
Speaker 2: headline that said AI could actually help rebuild the middle class,

77
00:04:05,920 --> 00:04:08,280
Speaker 2: which is very counterintuitive, very the opposite of what we're

78
00:04:08,280 --> 00:04:12,160
Speaker 2: talking about. But then I noticed who the author of

79
00:04:12,200 --> 00:04:16,040
Speaker 2: the piece was, and it's someone whose work is very

80
00:04:16,080 --> 00:04:19,839
Speaker 2: strongly associated with forces in the past and forces in

81
00:04:19,920 --> 00:04:23,520
Speaker 2: technology that have been destructive to the middle class and

82
00:04:23,560 --> 00:04:27,120
Speaker 2: have caused great labor market upheaval. And so if someone

83
00:04:27,160 --> 00:04:30,479
Speaker 2: who has sort of been watching this exact topic, the

84
00:04:30,600 --> 00:04:35,320
Speaker 2: intersection of labor market upheaval and technological change, is saying,

85
00:04:35,600 --> 00:04:38,160
Speaker 2: actually this could be good, and this person is a

86
00:04:38,200 --> 00:04:40,800
Speaker 2: track record in this area, I'm like, Okay, this is

87
00:04:40,839 --> 00:04:43,400
Speaker 2: an argument maybe I'll pay more attention to than the

88
00:04:43,440 --> 00:04:44,880
Speaker 2: random person doing a Twitter threat.

89
00:04:45,240 --> 00:04:48,320
Speaker 1: I'm into it. As you mentioned, we're speaking to someone

90
00:04:48,320 --> 00:04:51,480
Speaker 1: who is an expert on this particular topic and specifically

91
00:04:51,560 --> 00:04:54,520
Speaker 1: has has written a lot and researched a lot about

92
00:04:54,520 --> 00:04:58,280
Speaker 1: previous labor market shocks, including the China Shock. So competition

93
00:04:58,400 --> 00:05:02,000
Speaker 1: from China in the realms of manufacturing in the sort

94
00:05:02,000 --> 00:05:05,239
Speaker 1: of nineteen nineties, early two thousands. So I'm very excited

95
00:05:05,240 --> 00:05:08,520
Speaker 1: for this conversation. I am interested to hear an argument

96
00:05:08,600 --> 00:05:11,200
Speaker 1: that's not just AI is terrible and it's going to

97
00:05:11,240 --> 00:05:12,320
Speaker 1: take all of our jobs.

98
00:05:12,560 --> 00:05:15,480
Speaker 2: Absolutely well, I'm really excited. We do, in fact have

99
00:05:15,600 --> 00:05:17,719
Speaker 2: the perfect guest. We are going to be speaking with,

100
00:05:17,800 --> 00:05:21,240
Speaker 2: David Otter. He's a professor of economics at MIT and

101
00:05:21,320 --> 00:05:25,320
Speaker 2: co director of the MIT Shaping the Future of Work Initiative,

102
00:05:25,880 --> 00:05:29,560
Speaker 2: and he's really known for his work on the China

103
00:05:29,640 --> 00:05:34,080
Speaker 2: Shock and the devastating impact that China's boom in tradeable goods,

104
00:05:34,320 --> 00:05:38,520
Speaker 2: particularly after its essension to the WTO, had on various

105
00:05:38,520 --> 00:05:42,159
Speaker 2: communities within the United States that were sort of dependent

106
00:05:42,400 --> 00:05:46,400
Speaker 2: on a sort of regional manufacturing. So, David, thank you

107
00:05:46,480 --> 00:05:48,159
Speaker 2: so much for coming on odd LATS.

108
00:05:48,800 --> 00:05:50,599
Speaker 4: Thank you so much. Joe and Tracy for inviting me.

109
00:05:50,680 --> 00:05:51,960
Speaker 4: I'll try not to be clickbaity.

110
00:05:52,800 --> 00:05:54,320
Speaker 3: That's okay. It's okay.

111
00:05:54,320 --> 00:05:56,880
Speaker 2: It's okay to be clickbait if it delivers. And the

112
00:05:56,920 --> 00:05:58,839
Speaker 2: other thing about this article, by the way, is that

113
00:05:58,920 --> 00:06:01,640
Speaker 2: it wasn't like a paragraph thought piece like this is

114
00:06:01,760 --> 00:06:04,960
Speaker 2: clearly some serious work which we obviously appreciated and made

115
00:06:05,000 --> 00:06:07,599
Speaker 2: me take it seriously. But before we get into this

116
00:06:07,839 --> 00:06:11,359
Speaker 2: or even the China shock or general work or AI

117
00:06:11,400 --> 00:06:14,760
Speaker 2: in general, what is your like, what do you tell

118
00:06:14,839 --> 00:06:17,040
Speaker 2: us like what has been the thrust of your career

119
00:06:17,120 --> 00:06:17,680
Speaker 2: over time?

120
00:06:17,839 --> 00:06:17,919
Speaker 1: Like?

121
00:06:18,000 --> 00:06:21,120
Speaker 2: What is what is sort of the main interest of

122
00:06:21,160 --> 00:06:25,680
Speaker 2: yours that spans from the effects of globalization to now AI,

123
00:06:25,960 --> 00:06:26,480
Speaker 2: et cetera.

124
00:06:26,920 --> 00:06:30,640
Speaker 4: My focus has always been on forces that shape opportunity,

125
00:06:30,839 --> 00:06:35,000
Speaker 4: particularly for workers without for your college degrees, the majority

126
00:06:35,000 --> 00:06:37,040
Speaker 4: of workers in the United States and of course elsewhere,

127
00:06:37,600 --> 00:06:41,200
Speaker 4: and that have been so buffeted by computerization, by globalization,

128
00:06:41,640 --> 00:06:44,680
Speaker 4: by changes and institutions, including the unionization, the fall the

129
00:06:44,680 --> 00:06:47,320
Speaker 4: mid and wages the United States. And so that is

130
00:06:47,360 --> 00:06:50,440
Speaker 4: the common focus of my work, and that has included,

131
00:06:50,560 --> 00:06:52,680
Speaker 4: you know, a lot of work on technological change, computerization,

132
00:06:53,200 --> 00:06:55,880
Speaker 4: the China trade shock, and many other angles to that.

133
00:06:55,880 --> 00:06:57,480
Speaker 4: But that kind of unifies that's you know, I think

134
00:06:57,480 --> 00:06:59,279
Speaker 4: the labor market is the most important thing in the world.

135
00:06:59,640 --> 00:07:01,359
Speaker 4: I think that's where people drive most of their income,

136
00:07:01,400 --> 00:07:05,120
Speaker 4: spend most of their time, derive identity from, and so

137
00:07:05,480 --> 00:07:08,840
Speaker 4: things that affect the quality of jobs, the opportunities that

138
00:07:08,880 --> 00:07:12,760
Speaker 4: people have are just quintessentially important and are going to

139
00:07:12,800 --> 00:07:15,520
Speaker 4: shape the structure of their lives, you know, more than

140
00:07:15,520 --> 00:07:18,200
Speaker 4: the quality of entertainment, more than the ease of transportation,

141
00:07:18,360 --> 00:07:20,680
Speaker 4: more than you know, what fashion is available. This is

142
00:07:20,680 --> 00:07:21,560
Speaker 4: really a big ee.

143
00:07:21,840 --> 00:07:24,800
Speaker 1: So in the spirit of this discussion, I asked to

144
00:07:24,880 --> 00:07:30,480
Speaker 1: chat GPT to poke intellectual and logical holes in this article.

145
00:07:30,920 --> 00:07:33,720
Speaker 1: So let's just start there. Number one. No, I'm joking.

146
00:07:33,760 --> 00:07:36,080
Speaker 1: I did actually do that, and some of them, some

147
00:07:36,160 --> 00:07:38,400
Speaker 1: of them are quite good, and I will get to

148
00:07:38,440 --> 00:07:41,240
Speaker 1: them later. But maybe just to begin with, could you

149
00:07:41,280 --> 00:07:45,560
Speaker 1: talk about the current discourse on AI and why there

150
00:07:45,680 --> 00:07:50,200
Speaker 1: seems to be this distrust of new technology. What is

151
00:07:50,240 --> 00:07:52,880
Speaker 1: that predicated on. I mean I kind of referred to

152
00:07:52,920 --> 00:07:56,160
Speaker 1: it in the intro, but there is past history, obviously

153
00:07:56,240 --> 00:08:00,160
Speaker 1: with major technological advances and booms that have led to

154
00:08:00,240 --> 00:08:03,480
Speaker 1: certain outcomes in the labor market. How does that inform

155
00:08:03,560 --> 00:08:04,520
Speaker 1: the current discussion.

156
00:08:04,800 --> 00:08:08,360
Speaker 4: Sure, so people are understandably very concerned about all of

157
00:08:08,400 --> 00:08:11,440
Speaker 4: these technological forces because they are disruptive and they create

158
00:08:11,480 --> 00:08:13,920
Speaker 4: winners and losers. There's no sense in which everyone is

159
00:08:13,960 --> 00:08:17,280
Speaker 4: better off because of a new technology. So you mentioned

160
00:08:17,680 --> 00:08:20,360
Speaker 4: the industrial era, and the Luddites rose up against the

161
00:08:20,360 --> 00:08:25,040
Speaker 4: introduction power looms and smashed them, and they're often derided historically,

162
00:08:25,080 --> 00:08:29,400
Speaker 4: but they were right. The industry revolution, the mechanization of

163
00:08:29,520 --> 00:08:33,120
Speaker 4: weaving wiped out the careers of artisans and made their

164
00:08:33,160 --> 00:08:37,000
Speaker 4: work non tenable, and wages didn't rise for decades into

165
00:08:37,000 --> 00:08:40,160
Speaker 4: the Industrial Revolution. So that was very displacing. Ultimately at

166
00:08:40,240 --> 00:08:42,160
Speaker 4: raised living stairs, but it took a long time and

167
00:08:42,200 --> 00:08:46,199
Speaker 4: the beneficiaries were not workers. The computer revolution has raised productivity,

168
00:08:46,520 --> 00:08:50,160
Speaker 4: but it's been very unequal and polarizing. It's automated a

169
00:08:50,200 --> 00:08:53,160
Speaker 4: lot of middle skill, middle class work in factories and

170
00:08:53,160 --> 00:08:56,199
Speaker 4: in offices. It's been great for professionals, but for many

171
00:08:56,200 --> 00:08:58,560
Speaker 4: other people it's just meant that because they can no

172
00:08:58,600 --> 00:09:00,800
Speaker 4: longer do. Those middle skill jobs are often found in

173
00:09:01,040 --> 00:09:05,040
Speaker 4: food service, cleaning, security, entertainment, recreation, and those are valuable,

174
00:09:05,440 --> 00:09:09,400
Speaker 4: laudable activities, but they don't pay well. Because they don't

175
00:09:09,520 --> 00:09:13,319
Speaker 4: use specialized expertise in training, so most people can do

176
00:09:13,400 --> 00:09:16,080
Speaker 4: that work almost right away, so it tends to be

177
00:09:16,160 --> 00:09:19,480
Speaker 4: low paid. So I think there's many reasons to take

178
00:09:19,520 --> 00:09:22,560
Speaker 4: this very very seriously and think carefully about what the

179
00:09:22,600 --> 00:09:23,600
Speaker 4: implications are.

180
00:09:23,920 --> 00:09:26,200
Speaker 2: Before we even get to AI talked to us more

181
00:09:26,200 --> 00:09:30,000
Speaker 2: about the computer revolution, because, like I said, I saw

182
00:09:30,040 --> 00:09:31,680
Speaker 2: your piece and I'm like, uh, and I first thought

183
00:09:31,800 --> 00:09:34,280
Speaker 2: China Shock and your work on that. It's like, okay,

184
00:09:34,320 --> 00:09:36,599
Speaker 2: this is interesting, but actually, just like I feel like

185
00:09:36,640 --> 00:09:40,239
Speaker 2: there actually has not been a lot of general conversation

186
00:09:40,360 --> 00:09:44,640
Speaker 2: about the sort of unequalizing effects of the computer revolution,

187
00:09:44,760 --> 00:09:47,720
Speaker 2: like how did that happen? What does the research show

188
00:09:47,720 --> 00:09:50,720
Speaker 2: about the timing of the introduction of the computers, And

189
00:09:50,760 --> 00:09:53,240
Speaker 2: then this sort of like I don't know, maybe Barbelle

190
00:09:53,320 --> 00:09:55,679
Speaker 2: or fragmentation of what happened to workers.

191
00:09:56,360 --> 00:09:59,560
Speaker 4: So you know, this really begins in the nineteen eighties

192
00:09:59,600 --> 00:10:04,480
Speaker 4: and it can continues through the over at least thirty

193
00:10:04,480 --> 00:10:07,120
Speaker 4: five years. And you know, a very simple way to

194
00:10:07,360 --> 00:10:09,600
Speaker 4: boil it down and say, look, what are computers useful for.

195
00:10:10,000 --> 00:10:12,840
Speaker 4: They're useful for following rules and procedures, right, they don't

196
00:10:13,000 --> 00:10:16,800
Speaker 4: think they're not creative. They're not problem solvers. They don't improvise,

197
00:10:17,280 --> 00:10:21,200
Speaker 4: they follow codified rules and procedures. But that describes a

198
00:10:21,200 --> 00:10:23,840
Speaker 4: lot of middle skill work. Right, whether you're in an

199
00:10:23,840 --> 00:10:26,520
Speaker 4: office or you're doing repetitive assembly work. The ability to

200
00:10:26,840 --> 00:10:30,720
Speaker 4: accurately carry out codified procedures is a valuable skill. It

201
00:10:30,800 --> 00:10:34,000
Speaker 4: requires off in literacy and numeracy and training, and so

202
00:10:34,160 --> 00:10:37,360
Speaker 4: the ability to automate that was a really big deal,

203
00:10:37,960 --> 00:10:41,640
Speaker 4: and that had the effect of displacing many people who

204
00:10:41,640 --> 00:10:43,880
Speaker 4: were doing what I would call these mass expertise jobs

205
00:10:43,880 --> 00:10:46,880
Speaker 4: where they were following codified procedures. Right, it takes education

206
00:10:47,400 --> 00:10:51,000
Speaker 4: to be a typist or a bookkeeper or someone who

207
00:10:51,040 --> 00:10:54,760
Speaker 4: does filing an organization, keeps track of accounts, and so

208
00:10:54,840 --> 00:10:56,920
Speaker 4: the fact that a lot of that work takes real skill.

209
00:10:57,040 --> 00:10:59,600
Speaker 4: To do high quality work on assembly line, you have

210
00:10:59,600 --> 00:11:01,760
Speaker 4: to understan the tools, who have to understand the product

211
00:11:01,760 --> 00:11:03,680
Speaker 4: and so on. So the fact that that work could

212
00:11:03,679 --> 00:11:07,559
Speaker 4: be automated was not unambiguously good. It was good for

213
00:11:07,840 --> 00:11:09,920
Speaker 4: you know, it was good for productivity, it was good

214
00:11:09,920 --> 00:11:11,800
Speaker 4: for consumers, it was good for firms. But for the

215
00:11:11,840 --> 00:11:15,000
Speaker 4: workers who had invested their careers in those activities, that

216
00:11:15,120 --> 00:11:17,719
Speaker 4: was definitely a negative. And on the other hand, if

217
00:11:17,720 --> 00:11:20,360
Speaker 4: you were a professional or you know, a manager or

218
00:11:20,400 --> 00:11:26,119
Speaker 4: a designer, researcher, doctor, having access to information and quick calculation,

219
00:11:26,720 --> 00:11:29,240
Speaker 4: that's not your main job. Those are just inputs into

220
00:11:29,240 --> 00:11:32,600
Speaker 4: your decision making. So computers were very complimentary to people

221
00:11:32,640 --> 00:11:34,720
Speaker 4: who are decision makers, which is really the bulk of

222
00:11:34,720 --> 00:11:38,280
Speaker 4: the professions making high stakes decisions about you know, important

223
00:11:38,280 --> 00:11:40,280
Speaker 4: one off cases. You know how to care for a

224
00:11:40,280 --> 00:11:43,040
Speaker 4: cancer patient, or you know how to design a building,

225
00:11:43,320 --> 00:11:46,600
Speaker 4: or how to do a marketing plan. Right, computerizations extremely

226
00:11:46,600 --> 00:11:48,960
Speaker 4: helpful for that doesn't displace your main job, it just

227
00:11:48,960 --> 00:11:51,200
Speaker 4: makes you more efficient add it. But for people who

228
00:11:51,520 --> 00:11:53,920
Speaker 4: did not have the opportunity to get degrees and move

229
00:11:54,000 --> 00:11:57,160
Speaker 4: upward into that work, what remained was a lot of

230
00:11:57,520 --> 00:12:00,560
Speaker 4: work that's very hard to automate. You to mention these

231
00:12:00,600 --> 00:12:02,960
Speaker 4: a lot of these hands on manual jobs, so you know,

232
00:12:03,040 --> 00:12:07,160
Speaker 4: food service and cleaning, could be a transportation and many

233
00:12:07,240 --> 00:12:09,680
Speaker 4: of those jobs. Not all of those jobs are open

234
00:12:09,800 --> 00:12:13,880
Speaker 4: to many many people. They don't require much training or experience,

235
00:12:14,520 --> 00:12:17,200
Speaker 4: and you don't get a great deal better at them

236
00:12:17,240 --> 00:12:20,360
Speaker 4: over time. And so because of that, because they're non

237
00:12:20,360 --> 00:12:22,680
Speaker 4: expert work, they tend to be low paid in all

238
00:12:22,800 --> 00:12:25,240
Speaker 4: industrialized countries. Now, I want to be clear that not

239
00:12:25,320 --> 00:12:27,600
Speaker 4: all hands on work is low paid or low skilled

240
00:12:27,600 --> 00:12:29,679
Speaker 4: in any sense. Right, if you're a plumber, electrician, you're

241
00:12:29,679 --> 00:12:31,800
Speaker 4: working the skilled trades, right, if you do skilled repair.

242
00:12:32,200 --> 00:12:35,040
Speaker 4: There are many, many skilled hands on jobs, but the

243
00:12:35,160 --> 00:12:37,719
Speaker 4: ones that have grown so much as the middle has

244
00:12:37,760 --> 00:12:40,120
Speaker 4: hollowed out, have been much more of these personal service

245
00:12:40,160 --> 00:12:56,559
Speaker 4: occupations that have low training and expertise requirements.

246
00:12:58,760 --> 00:13:01,280
Speaker 1: The thing this reminds me of, and I cannot, for

247
00:13:01,320 --> 00:13:04,000
Speaker 1: the life of me remember which guest this was, but

248
00:13:04,040 --> 00:13:07,720
Speaker 1: a previous all Lots guests described this as remember the

249
00:13:07,760 --> 00:13:11,120
Speaker 1: scene from The Producers where Matthew Broderick is like an

250
00:13:11,160 --> 00:13:14,480
Speaker 1: actuary or something working in an office and they're all

251
00:13:14,520 --> 00:13:15,520
Speaker 1: toiling away.

252
00:13:15,400 --> 00:13:17,439
Speaker 2: And then Stuart Butterfield.

253
00:13:19,120 --> 00:13:22,280
Speaker 1: That's right, and then all of those people eventually get

254
00:13:22,320 --> 00:13:25,400
Speaker 1: replaced by an Excel spreadsheet, right, Like that's the function

255
00:13:25,800 --> 00:13:28,960
Speaker 1: that became Excel. So, David, I want to kind of

256
00:13:28,960 --> 00:13:30,520
Speaker 1: press you on this point because I think it's a

257
00:13:30,559 --> 00:13:33,679
Speaker 1: really interesting one, and I think it's essential to understanding

258
00:13:33,679 --> 00:13:38,520
Speaker 1: your overall argument. But you make the distinction between information

259
00:13:39,360 --> 00:13:43,000
Speaker 1: and decision making. So the idea that people can have

260
00:13:43,080 --> 00:13:46,400
Speaker 1: access to a lot of information. In fact, plenty of

261
00:13:46,400 --> 00:13:49,679
Speaker 1: people would argue that people are drowning in information at

262
00:13:49,679 --> 00:13:49,920
Speaker 1: the moment.

263
00:13:50,520 --> 00:13:50,880
Speaker 4: Information.

264
00:13:51,040 --> 00:13:54,920
Speaker 1: Yeah, but they're not necessarily using that to make the

265
00:13:54,960 --> 00:13:57,959
Speaker 1: best decisions. Decision making is sort of a separate skill.

266
00:13:58,000 --> 00:13:59,959
Speaker 1: Can you talk a little bit more about that aspect

267
00:14:00,040 --> 00:14:00,680
Speaker 1: of your argument?

268
00:14:01,160 --> 00:14:03,280
Speaker 4: Absolutely so. Let me so, I want to draw a

269
00:14:03,320 --> 00:14:06,280
Speaker 4: sharp line between AI and traditional computing, which is what

270
00:14:06,280 --> 00:14:09,720
Speaker 4: we've been discussing, because they're quite different. But before you

271
00:14:09,720 --> 00:14:11,360
Speaker 4: do that, let me kind of make a kind of

272
00:14:11,360 --> 00:14:13,400
Speaker 4: a meta argument that I think is useful our discussion.

273
00:14:13,840 --> 00:14:16,319
Speaker 4: So the concern we should be having is not about

274
00:14:16,320 --> 00:14:19,200
Speaker 4: the quantity of jobs. We are not running out of jobs.

275
00:14:19,400 --> 00:14:21,600
Speaker 4: And in fact, you know, all the Western world right

276
00:14:21,640 --> 00:14:24,960
Speaker 4: now is in full or overemployment, and even during the

277
00:14:24,960 --> 00:14:27,360
Speaker 4: whole computer revolutions on, we didn't run out of jobs.

278
00:14:27,400 --> 00:14:29,480
Speaker 4: It's not the quantity that matters. In fact, we're all

279
00:14:29,520 --> 00:14:32,640
Speaker 4: facing a demographic crunch. It's the quality. Right A world

280
00:14:32,640 --> 00:14:34,840
Speaker 4: in which everyone's waiting tables is very different from the

281
00:14:34,880 --> 00:14:37,880
Speaker 4: world in which everyone is doing medical care. And so

282
00:14:38,240 --> 00:14:41,600
Speaker 4: what matters is not simply whether there is work, but

283
00:14:41,680 --> 00:14:45,640
Speaker 4: whether it's expert work that requires real skills. If it's

284
00:14:45,680 --> 00:14:49,320
Speaker 4: non expert work, work that anyone can do with no

285
00:14:49,400 --> 00:14:52,640
Speaker 4: training your certification, unfortunately it will be low paid. On

286
00:14:52,640 --> 00:14:56,080
Speaker 4: the other hand, if it's work that requires specialized knowledge

287
00:14:56,080 --> 00:14:59,160
Speaker 4: and that is made more productive by uses of tools,

288
00:14:59,160 --> 00:15:01,640
Speaker 4: and computers are to and AI as a tool, then

289
00:15:02,120 --> 00:15:04,920
Speaker 4: that's good for labor, that's good for earnings, that's good

290
00:15:04,920 --> 00:15:06,680
Speaker 4: for the quality of careers. And so we should be

291
00:15:06,760 --> 00:15:09,360
Speaker 4: thinking about expertise. Just to give you, like a very

292
00:15:09,840 --> 00:15:12,680
Speaker 4: stylized example, you know, think of the job of crossing

293
00:15:12,720 --> 00:15:16,560
Speaker 4: guard and air traffic controller. These are basically the same job.

294
00:15:17,000 --> 00:15:20,120
Speaker 4: The job is to prevent things from crashing into other things, right,

295
00:15:20,280 --> 00:15:23,560
Speaker 4: airplanes from crashing to airplanes, cars from crashing into children

296
00:15:23,800 --> 00:15:26,440
Speaker 4: on their way to school. But air traffic controllers United

297
00:15:26,440 --> 00:15:28,000
Speaker 4: States are paid, you know, four and a half times

298
00:15:28,000 --> 00:15:31,040
Speaker 4: as much as crossing guards, and the reason is expertise.

299
00:15:31,320 --> 00:15:34,280
Speaker 4: Almost anyone can become a crossing guard in the United

300
00:15:34,280 --> 00:15:37,280
Speaker 4: States with no trainer certification, whereas to become an air

301
00:15:37,280 --> 00:15:41,680
Speaker 4: traffic controller requires years of school and thousands of hours

302
00:15:41,680 --> 00:15:44,720
Speaker 4: of practice. And so even though those jobs do the

303
00:15:44,760 --> 00:15:48,480
Speaker 4: same thing, because of the difference in skill requirements, they

304
00:15:48,520 --> 00:15:51,480
Speaker 4: pay very different wage levels, and so we want to

305
00:15:51,520 --> 00:15:56,320
Speaker 4: have jobs where expertise is valuable, not just where physical

306
00:15:56,360 --> 00:15:59,880
Speaker 4: presence is the primary requirement. So that's what we should

307
00:15:59,880 --> 00:16:02,280
Speaker 4: be thinking about. Having said that, let me talk about

308
00:16:02,440 --> 00:16:06,120
Speaker 4: how AI relates to that. So, you know, a traditional computerization,

309
00:16:06,200 --> 00:16:10,000
Speaker 4: as we've been talking about, is really about automating well

310
00:16:10,040 --> 00:16:13,000
Speaker 4: understood procedures and rules, right, what we call formal knowledge.

311
00:16:13,040 --> 00:16:14,960
Speaker 4: You know how to do math, how to reproduce a

312
00:16:14,960 --> 00:16:18,240
Speaker 4: document or check for spelling errors. And it's very limited

313
00:16:18,280 --> 00:16:23,640
Speaker 4: because it cannot do what people do fairly effortlessly, which

314
00:16:23,720 --> 00:16:26,560
Speaker 4: is learn from kind of tacit knowledge. Tacit knowledge is

315
00:16:26,960 --> 00:16:29,480
Speaker 4: all the things that you implicitly understand that you infer

316
00:16:29,520 --> 00:16:32,080
Speaker 4: from your environment, but you never formalize. Right. So, you

317
00:16:32,120 --> 00:16:34,560
Speaker 4: know how to ride a bicycle, but you couldn't explain

318
00:16:34,640 --> 00:16:37,160
Speaker 4: how it's done. Right. You couldn't sit up and explain

319
00:16:37,240 --> 00:16:39,840
Speaker 4: that you know the gyroscopic physics of a bicycle. You

320
00:16:39,920 --> 00:16:41,680
Speaker 4: know how to make a funny joke, but you don't

321
00:16:41,680 --> 00:16:43,840
Speaker 4: know the rules for making a funny joke. You know

322
00:16:43,880 --> 00:16:46,120
Speaker 4: how to recognize the face of someone after you haven't

323
00:16:46,160 --> 00:16:48,920
Speaker 4: seen them for thirty years, right, But that's actually a

324
00:16:48,920 --> 00:16:52,320
Speaker 4: hard problem, and we do it, but we do it

325
00:16:52,400 --> 00:16:55,000
Speaker 4: based on some tacit understanding. And this has always been

326
00:16:55,040 --> 00:16:58,520
Speaker 4: a barrier to computerization because we couldn't code up the

327
00:16:58,560 --> 00:17:01,480
Speaker 4: things that we understood only acidly. We had to understand

328
00:17:01,480 --> 00:17:07,240
Speaker 4: them explicitly, informally. So AI overcomes that barrier. AI essentially

329
00:17:07,640 --> 00:17:11,359
Speaker 4: infers tacit information from large bodies of data. It learns

330
00:17:11,359 --> 00:17:14,760
Speaker 4: the associations between you know, words and phrases and sentences

331
00:17:14,800 --> 00:17:18,520
Speaker 4: between pictures and words. It can look at a scan

332
00:17:18,800 --> 00:17:22,520
Speaker 4: of a patient's lungs and make predictions or you know,

333
00:17:22,640 --> 00:17:25,600
Speaker 4: guesses about whether that patient has an endema or other

334
00:17:26,119 --> 00:17:28,919
Speaker 4: medical disorders. It does that not because someone has written

335
00:17:28,920 --> 00:17:31,919
Speaker 4: a program that says, these things tell you whether you have,

336
00:17:32,000 --> 00:17:34,640
Speaker 4: you know, a lung issue. It's because it learns from

337
00:17:34,640 --> 00:17:37,399
Speaker 4: the patterns it's trained on that data, and so that

338
00:17:37,440 --> 00:17:40,680
Speaker 4: gives it a really different set of capabilities. It gives

339
00:17:40,720 --> 00:17:43,560
Speaker 4: it the ability to do what a lot of us do,

340
00:17:44,000 --> 00:17:45,679
Speaker 4: or at least to supplement a lot of what we do,

341
00:17:45,720 --> 00:17:49,840
Speaker 4: which is to sort of make decisions based on lots

342
00:17:49,840 --> 00:17:53,360
Speaker 4: and lots of inputs and educated guesses. Right, So let's

343
00:17:53,359 --> 00:17:55,879
Speaker 4: say you know, you're a medical doctor, right, when you

344
00:17:55,920 --> 00:17:59,320
Speaker 4: see a patient, you're not simply essentially reading from your

345
00:17:59,320 --> 00:18:02,720
Speaker 4: textbook in your mind about what to do. You understand

346
00:18:02,960 --> 00:18:06,639
Speaker 4: bodily systems, you understand the biology and so on, but

347
00:18:06,720 --> 00:18:08,959
Speaker 4: then you've had lots and lots of experience. So when

348
00:18:09,000 --> 00:18:11,439
Speaker 4: you've see an individual patient, you're going to make a

349
00:18:11,440 --> 00:18:14,639
Speaker 4: decision based on a kind of translation from this formal

350
00:18:14,640 --> 00:18:17,679
Speaker 4: body of knowledge plus all the experience you've had to

351
00:18:17,720 --> 00:18:19,879
Speaker 4: make a good judgment. And the stakes are really high

352
00:18:20,000 --> 00:18:22,359
Speaker 4: because obviously if there was just a simple rule book

353
00:18:22,359 --> 00:18:25,119
Speaker 4: for it, you wouldn't need a doctor. You need a

354
00:18:25,160 --> 00:18:27,360
Speaker 4: person who can make a judgment about how to care

355
00:18:27,400 --> 00:18:29,280
Speaker 4: for this patient and their individual needs.

356
00:18:29,440 --> 00:18:29,960
Speaker 3: It's so funny.

357
00:18:30,000 --> 00:18:32,679
Speaker 2: I was just talking to Tracy in a different context,

358
00:18:32,680 --> 00:18:34,760
Speaker 2: and I was like, I was talking about the TV

359
00:18:34,840 --> 00:18:38,160
Speaker 2: show House, which I'm really into, and like, you know,

360
00:18:38,200 --> 00:18:41,119
Speaker 2: even though it's probably hyper dramatized, this idea of like

361
00:18:41,240 --> 00:18:44,360
Speaker 2: how still today like doctors don't really know a lot

362
00:18:44,400 --> 00:18:46,520
Speaker 2: and they have to like they debate, well, what's actually

363
00:18:46,640 --> 00:18:48,840
Speaker 2: going on here? And of course the show has some

364
00:18:49,000 --> 00:18:52,760
Speaker 2: very entertaining depictions of what those debates among doctors of

365
00:18:52,920 --> 00:18:55,280
Speaker 2: what's really going wrong with the patient and what's the

366
00:18:55,440 --> 00:18:58,320
Speaker 2: proper treatment. So I guess you can go from there

367
00:18:58,359 --> 00:19:01,439
Speaker 2: and just say, well, House was like the most brilliant

368
00:19:01,440 --> 00:19:03,800
Speaker 2: and he had seen thousands of patients over the course

369
00:19:03,840 --> 00:19:06,119
Speaker 2: of several seasons of that show, and so he had

370
00:19:06,160 --> 00:19:10,560
Speaker 2: the best like intuitions. But basically it sounds like, thanks

371
00:19:10,600 --> 00:19:16,000
Speaker 2: to AI, someone can harness those same intuitions without having

372
00:19:16,080 --> 00:19:19,040
Speaker 2: seen thousands of patients before, like doctor House did.

373
00:19:19,680 --> 00:19:21,879
Speaker 4: I think that's a nice way to put it, is

374
00:19:21,920 --> 00:19:25,639
Speaker 4: that what AI can do is provide kind of guidance

375
00:19:25,760 --> 00:19:28,439
Speaker 4: and guardrails for decision making. So what do I mean

376
00:19:28,440 --> 00:19:30,720
Speaker 4: by guidance and guardrails. By guidance, I mean, you know,

377
00:19:30,840 --> 00:19:34,960
Speaker 4: had you considered this set of possibilities, these potential diagnoses,

378
00:19:35,040 --> 00:19:37,760
Speaker 4: guardrails were like, you know, don't prescribe these two drugs together.

379
00:19:37,800 --> 00:19:42,200
Speaker 4: They negatively interact and in decision making work. Having that

380
00:19:42,280 --> 00:19:46,040
Speaker 4: kind of access to support, to a form of expertise,

381
00:19:46,119 --> 00:19:48,000
Speaker 4: not that you should one hundred percent rely upon it,

382
00:19:48,320 --> 00:19:51,760
Speaker 4: but that you can supplement your own judgment is potentially

383
00:19:52,000 --> 00:19:54,480
Speaker 4: very useful. So you know, let me give you a

384
00:19:54,600 --> 00:19:57,680
Speaker 4: concrete example, sticking with medicine. So the job of nurse

385
00:19:57,720 --> 00:20:01,399
Speaker 4: practitioner is pretty prominent right now, they're several hundred thousand

386
00:20:01,400 --> 00:20:03,400
Speaker 4: the United States. They make quite a good living, about

387
00:20:03,400 --> 00:20:04,960
Speaker 4: one hundred and thirty two thousand dollars a year of

388
00:20:05,000 --> 00:20:08,000
Speaker 4: the median. And they barely existed twenty years ago. And

389
00:20:08,200 --> 00:20:11,640
Speaker 4: nurse practitioners are nurses with an additional master's degree who

390
00:20:11,680 --> 00:20:16,520
Speaker 4: could do diagnosing, prescribing, treating things that were only done

391
00:20:16,880 --> 00:20:21,920
Speaker 4: by medical doctors decades earlier. And this new occupation has

392
00:20:21,920 --> 00:20:24,600
Speaker 4: come into existence, and it's terrific for patients in the

393
00:20:24,680 --> 00:20:26,600
Speaker 4: way in that it saves them time, it saves the

394
00:20:26,640 --> 00:20:29,600
Speaker 4: healthcare system money, it creates a good job, and it

395
00:20:29,640 --> 00:20:31,840
Speaker 4: does a very important task. Now, this is not a

396
00:20:31,880 --> 00:20:36,600
Speaker 4: technological creation, socially, the result of nurses recognizing they were

397
00:20:36,680 --> 00:20:41,080
Speaker 4: under used, fighting for a larger role, developing a training

398
00:20:41,080 --> 00:20:44,480
Speaker 4: and certification program, and eventually against over the dead body

399
00:20:44,520 --> 00:20:48,359
Speaker 4: of the American Medical Association, effectively carving out this new role.

400
00:20:48,920 --> 00:20:51,680
Speaker 4: So it's not because of technology. However, at this point,

401
00:20:51,760 --> 00:20:55,840
Speaker 4: nurse practitioners are heavily supported by technology. Right, So electronic

402
00:20:55,960 --> 00:20:58,919
Speaker 4: medical records right provide all the information all that you

403
00:20:58,920 --> 00:21:01,240
Speaker 4: would need or some of the information you would need

404
00:21:01,280 --> 00:21:05,080
Speaker 4: for good decision making, as do extensive diagnostic tests, as

405
00:21:05,119 --> 00:21:08,560
Speaker 4: does software that looks for drug interactions, among other things.

406
00:21:08,920 --> 00:21:11,560
Speaker 4: And it's easy to imagine that as we roll the

407
00:21:11,560 --> 00:21:15,880
Speaker 4: clock forward, the set of tools that will support decision

408
00:21:15,880 --> 00:21:19,560
Speaker 4: making by nurse practitioners will improve dramatically. And as it

409
00:21:19,600 --> 00:21:22,359
Speaker 4: does so, it will allow them to do more of

410
00:21:22,400 --> 00:21:26,720
Speaker 4: the tasks that are currently kind of controlled by more

411
00:21:26,720 --> 00:21:29,680
Speaker 4: expensive professionals. And why is that a good thing? You

412
00:21:29,760 --> 00:21:30,919
Speaker 4: might say, Well, it's not a good thing if your

413
00:21:31,000 --> 00:21:33,800
Speaker 4: doctor necessarily, But we live in a world in which

414
00:21:34,160 --> 00:21:38,119
Speaker 4: a lot of the bottlenecks are expensive decision makers, people

415
00:21:38,160 --> 00:21:42,199
Speaker 4: who are the MBAs and the lawyers and the medical

416
00:21:42,240 --> 00:21:45,360
Speaker 4: doctors and the architects and the engineers, and they all

417
00:21:45,400 --> 00:21:47,560
Speaker 4: do valid work and they deserve what they earn, and

418
00:21:47,600 --> 00:21:49,639
Speaker 4: I'm not disputing that, but it would be great to

419
00:21:49,640 --> 00:21:51,760
Speaker 4: be able to create more people who could do that

420
00:21:51,880 --> 00:21:55,800
Speaker 4: work without them being quite so expensive and the advantage

421
00:21:55,800 --> 00:21:59,080
Speaker 4: of that, So, if an AI can enable more people

422
00:21:59,119 --> 00:22:02,679
Speaker 4: to do good decision making work, it actually can open

423
00:22:02,800 --> 00:22:05,919
Speaker 4: up opportunity for people who are not the elite. Right,

424
00:22:06,000 --> 00:22:07,879
Speaker 4: we have tons and tons of healthcare that needs to

425
00:22:07,880 --> 00:22:10,400
Speaker 4: be done right. It doesn't all need to be done

426
00:22:10,400 --> 00:22:13,359
Speaker 4: by medical doctors, or we have lots of software coding

427
00:22:13,400 --> 00:22:15,199
Speaker 4: that needs to be done. It doesn't all need to

428
00:22:15,200 --> 00:22:19,159
Speaker 4: be done by people from top universities with Bachelors of

429
00:22:19,160 --> 00:22:22,879
Speaker 4: Science degrees in computer science. We have tons of design

430
00:22:22,920 --> 00:22:25,760
Speaker 4: that needs to be done, tons of care, tons of

431
00:22:25,840 --> 00:22:31,040
Speaker 4: legal work. Right, So the potential for an AI is

432
00:22:31,119 --> 00:22:34,960
Speaker 4: to enable people who have training and judgment to go

433
00:22:35,200 --> 00:22:39,120
Speaker 4: further with those skills. So it's not to make them unnecessary,

434
00:22:39,680 --> 00:22:44,879
Speaker 4: but simply to extend their range by supporting decision making. So,

435
00:22:45,080 --> 00:22:49,480
Speaker 4: just to give you another super concrete analogy, take YouTube. Right.

436
00:22:49,520 --> 00:22:53,399
Speaker 4: So YouTube is used all the time by people in

437
00:22:53,440 --> 00:22:56,200
Speaker 4: the trades among other groups to try to figure out

438
00:22:56,240 --> 00:22:59,560
Speaker 4: how to do a specific repair or diagnose the problem

439
00:22:59,600 --> 00:23:01,960
Speaker 4: that they have seen before. Now I'm gonna say, well,

440
00:23:02,000 --> 00:23:05,040
Speaker 4: who is YouTube really for. Well, it's not for the

441
00:23:05,080 --> 00:23:07,280
Speaker 4: frontier experts they already know how to do these things,

442
00:23:07,880 --> 00:23:11,000
Speaker 4: nor is it necessarily for the rank amateur. Right. You

443
00:23:11,000 --> 00:23:12,680
Speaker 4: don't want to go to YouTube and say, well, how

444
00:23:12,680 --> 00:23:16,439
Speaker 4: do I install and wire in a brand new central

445
00:23:16,440 --> 00:23:19,120
Speaker 4: house air conditioning? I've never done anything like that before. Right,

446
00:23:19,119 --> 00:23:21,040
Speaker 4: If you went Toto YouTube for that, you would quickly

447
00:23:21,080 --> 00:23:23,720
Speaker 4: get yourself into trouble because if you don't have some

448
00:23:23,800 --> 00:23:27,280
Speaker 4: foundational skills, that could be a problem. On the other hand,

449
00:23:27,880 --> 00:23:30,240
Speaker 4: if you were handy and you had some experience with

450
00:23:30,240 --> 00:23:32,920
Speaker 4: electrical work, some experience with plumbing, some experience with carpentry,

451
00:23:33,080 --> 00:23:36,159
Speaker 4: but you've never done an ac installation before, well, now

452
00:23:36,600 --> 00:23:38,680
Speaker 4: you could go to YouTube and that would get you further.

453
00:23:38,760 --> 00:23:41,040
Speaker 4: So you could think of YouTube as kind of like

454
00:23:41,040 --> 00:23:44,120
Speaker 4: a mini AI that provides guidance and guardrails.

455
00:23:44,280 --> 00:23:48,040
Speaker 2: I feel like Tracy has watched many youtubes in the

456
00:23:48,119 --> 00:23:50,200
Speaker 2: last year to fix your Connecticut house.

457
00:23:50,400 --> 00:23:54,160
Speaker 1: This example hits home so hard, and I'll give you

458
00:23:54,200 --> 00:23:57,239
Speaker 1: a specific anecdote, which is my husband and I are

459
00:23:57,280 --> 00:23:59,680
Speaker 1: currently building a shed and we're trying to put a

460
00:24:00,200 --> 00:24:02,440
Speaker 1: roof on it. And we thought like, okay, we put

461
00:24:02,440 --> 00:24:04,560
Speaker 1: the plywood on the roof, and then we get some

462
00:24:04,760 --> 00:24:08,240
Speaker 1: joyst tape. We put the joystape down over the edges,

463
00:24:08,320 --> 00:24:11,399
Speaker 1: and then we put on the shingles, and we watched many,

464
00:24:11,440 --> 00:24:13,920
Speaker 1: many YouTube videos on how to do this. It turns

465
00:24:13,920 --> 00:24:16,720
Speaker 1: out that you can't use joyste tape when it's less

466
00:24:16,760 --> 00:24:21,560
Speaker 1: than fifty degrees fahrenheit outside, which it was, which, of course,

467
00:24:21,640 --> 00:24:24,359
Speaker 1: none of the YouTube videos that are filmed down in

468
00:24:24,400 --> 00:24:27,720
Speaker 1: Florida or wherever actually mentioned. And then secondly, it turns

469
00:24:27,760 --> 00:24:30,200
Speaker 1: out that the ability of the joyst tape to actually

470
00:24:30,240 --> 00:24:34,600
Speaker 1: adhere to the plywood varies enormously depending on what plywood

471
00:24:34,760 --> 00:24:37,560
Speaker 1: you're using. So there are all these subtleties and nuances

472
00:24:37,600 --> 00:24:41,359
Speaker 1: that you don't necessarily get from a ten minute YouTube video.

473
00:24:41,560 --> 00:24:44,720
Speaker 1: Maybe that's not that surprising. But on this note, so

474
00:24:44,960 --> 00:24:48,119
Speaker 1: you mentioned training, and you've spoken a lot at this

475
00:24:48,200 --> 00:24:51,400
Speaker 1: point about the idea of AI being able to provide

476
00:24:51,760 --> 00:24:57,520
Speaker 1: guardrails and context around decision making that maybe yeah, can

477
00:24:57,640 --> 00:25:01,399
Speaker 1: resolve the bottleneck of expensive decision makers, as you put it,

478
00:25:01,480 --> 00:25:04,919
Speaker 1: by creating more of them or allowing more people to

479
00:25:05,160 --> 00:25:09,480
Speaker 1: tap that function. I guess my big question is how

480
00:25:09,640 --> 00:25:13,040
Speaker 1: much of this is just going to be Well, we

481
00:25:13,119 --> 00:25:16,479
Speaker 1: add a new layer of training that people have to

482
00:25:16,520 --> 00:25:18,720
Speaker 1: do so you can use AI, but you still have

483
00:25:18,800 --> 00:25:21,399
Speaker 1: to know how to use AI. You still have to

484
00:25:21,520 --> 00:25:24,879
Speaker 1: understand the result that it's spitting out and interpret that.

485
00:25:24,960 --> 00:25:27,959
Speaker 1: You still have to know how to actually apply and

486
00:25:28,359 --> 00:25:32,320
Speaker 1: use that result. Are we basically just replacing one skill

487
00:25:32,400 --> 00:25:33,680
Speaker 1: set with another.

488
00:25:34,280 --> 00:25:37,720
Speaker 4: It's a good question. We want it to require skills,

489
00:25:37,840 --> 00:25:40,640
Speaker 4: right if everyone is expert. No one is expert, right,

490
00:25:41,200 --> 00:25:45,399
Speaker 4: it's important. The question is whether it can be the

491
00:25:45,440 --> 00:25:48,520
Speaker 4: acquisition of expertise or whether it just gets in the way.

492
00:25:48,560 --> 00:25:51,360
Speaker 4: Another thing you have to certify on. We now have

493
00:25:51,480 --> 00:25:54,760
Speaker 4: you know a bunch of evidence on AI and specific

494
00:25:54,800 --> 00:25:57,320
Speaker 4: applications and where it works well and where it doesn't.

495
00:25:57,480 --> 00:26:00,320
Speaker 4: So for example, you know some students of mine, ked

496
00:26:00,320 --> 00:26:02,520
Speaker 4: Noy and Whitney Zang published a paper in Science last

497
00:26:02,600 --> 00:26:05,600
Speaker 4: year where they gave chat gipt three and a half

498
00:26:05,640 --> 00:26:09,200
Speaker 4: to people who were doing advertising writing and marketing plans.

499
00:26:09,240 --> 00:26:11,119
Speaker 4: And these were people who were college graduates who do

500
00:26:11,160 --> 00:26:13,639
Speaker 4: this for a living. And one group just used the

501
00:26:13,640 --> 00:26:16,600
Speaker 4: standard tools, so it's basically the Internet word processors. Another

502
00:26:16,640 --> 00:26:18,720
Speaker 4: one actually used the chatbot and this was early enough

503
00:26:18,760 --> 00:26:21,480
Speaker 4: that most people didn't already have it, and there were

504
00:26:21,520 --> 00:26:23,960
Speaker 4: a couple of really nice results. So, first thing, it

505
00:26:24,000 --> 00:26:26,399
Speaker 4: saved everybody time. It cut the time it took people

506
00:26:26,480 --> 00:26:28,880
Speaker 4: to do this work from about thirty minutes to about eighteen.

507
00:26:29,280 --> 00:26:32,359
Speaker 4: The second is it improved the quality on average. So

508
00:26:32,960 --> 00:26:35,719
Speaker 4: the output of the people using this tool was judged

509
00:26:35,760 --> 00:26:38,320
Speaker 4: and by other college graduates who were not confederates in

510
00:26:38,359 --> 00:26:43,920
Speaker 4: the experiment to be more precise, more concise, and more accurate,

511
00:26:44,520 --> 00:26:46,840
Speaker 4: so improve the quality of work and saved time. But

512
00:26:46,880 --> 00:26:49,320
Speaker 4: then the most exciting result was if you looked at

513
00:26:49,320 --> 00:26:53,520
Speaker 4: the quality range of the work people did, it basically

514
00:26:53,600 --> 00:26:56,840
Speaker 4: made the least capable writers using chat GPT were about

515
00:26:56,880 --> 00:26:59,080
Speaker 4: as good as the median writers not using it, So

516
00:26:59,119 --> 00:27:01,760
Speaker 4: it kind of leveled up the bottom. And we've seen

517
00:27:01,760 --> 00:27:05,040
Speaker 4: this in other places as well, folks doing customer support.

518
00:27:05,600 --> 00:27:07,119
Speaker 4: The example I'm thinking of is a kind of an

519
00:27:07,440 --> 00:27:11,520
Speaker 4: enterprise software product and they customers chat in through chat window,

520
00:27:12,160 --> 00:27:16,000
Speaker 4: and then the company installed a tool that suggests responses

521
00:27:16,040 --> 00:27:18,199
Speaker 4: to the customer's chat. You don't have to use them,

522
00:27:18,359 --> 00:27:21,119
Speaker 4: but it will also not just suggest technical responses, but

523
00:27:21,200 --> 00:27:24,240
Speaker 4: polite responses and so on to keep the customer from

524
00:27:24,440 --> 00:27:29,240
Speaker 4: getting overheated. And the result is that it speeds the

525
00:27:29,320 --> 00:27:32,040
Speaker 4: rate at which people learn. So it used to take

526
00:27:32,080 --> 00:27:36,280
Speaker 4: people ten months to reach peak capacity. Now it takes

527
00:27:36,320 --> 00:27:40,480
Speaker 4: them about three months. They're somewhat faster when that's done,

528
00:27:40,720 --> 00:27:43,280
Speaker 4: so it's not that it eliminates the training or learning.

529
00:27:43,400 --> 00:27:47,080
Speaker 4: Everyone starts off bad at this job, but they get faster.

530
00:27:47,200 --> 00:27:50,040
Speaker 4: They converge towards expert level more quickly. But this tool,

531
00:27:50,119 --> 00:27:54,320
Speaker 4: and also really interestingly, people quit a lot less. And

532
00:27:54,400 --> 00:27:57,120
Speaker 4: the reason is, you know, customer service work is actually

533
00:27:57,520 --> 00:28:00,560
Speaker 4: really difficult. It's very heavy emotional lay and you have

534
00:28:00,640 --> 00:28:03,080
Speaker 4: to take a lot of incoming abuse actually from customers.

535
00:28:03,119 --> 00:28:06,760
Speaker 4: It's hard to keep your cool. And the sentiment analysis

536
00:28:06,760 --> 00:28:09,240
Speaker 4: of this tool, of the chats that occurred through it,

537
00:28:09,280 --> 00:28:12,240
Speaker 4: is that it basically reduced the level of hostility from

538
00:28:12,320 --> 00:28:15,080
Speaker 4: customers to workers and from workers to customers. So it

539
00:28:15,080 --> 00:28:17,760
Speaker 4: actually did a lot of the emotional labor. So it

540
00:28:17,800 --> 00:28:20,960
Speaker 4: didn't eliminate the need for skills in doing this work,

541
00:28:20,960 --> 00:28:25,800
Speaker 4: but it enabled people to become more efficient, more rapidly,

542
00:28:26,240 --> 00:28:29,720
Speaker 4: with less stress. And so that's the good scenario. There's

543
00:28:29,760 --> 00:28:32,520
Speaker 4: a lot of work that needs to be done. And

544
00:28:32,840 --> 00:28:36,480
Speaker 4: right now, what are the most expensive things? The things

545
00:28:36,520 --> 00:28:38,600
Speaker 4: that are growing more and more costly all the time

546
00:28:38,720 --> 00:28:44,120
Speaker 4: are education, healthcare, legal services. Why is that? Why are

547
00:28:44,120 --> 00:28:47,160
Speaker 4: those things getting so expensive? Well, during the industrial era,

548
00:28:47,200 --> 00:28:54,280
Speaker 4: we got really efficient and manufacturing goods. Right, so TVs, automobiles, coffeemakers, right,

549
00:28:54,320 --> 00:28:59,280
Speaker 4: mobile phones, these things are actually remarkably good and relatively cheap. Why, well,

550
00:28:59,440 --> 00:29:03,000
Speaker 4: we've automated them and the labor content is relatively low.

551
00:29:03,320 --> 00:29:08,080
Speaker 4: On the other hand, healthcare, education, law, Right, we've not

552
00:29:08,120 --> 00:29:11,080
Speaker 4: gotten any more efficient to those things, and they require

553
00:29:11,120 --> 00:29:13,880
Speaker 4: people who've gotten more and more expensive over time because

554
00:29:13,920 --> 00:29:15,880
Speaker 4: as we've automated the other work, the people who are

555
00:29:15,880 --> 00:29:19,440
Speaker 4: the degreed professionals or have become the bottleneck. So that

556
00:29:19,680 --> 00:29:22,760
Speaker 4: slows the growth of productivity. It makes the cost of

557
00:29:22,800 --> 00:29:26,040
Speaker 4: living higher for the typical person. Right, typical person is

558
00:29:26,040 --> 00:29:28,320
Speaker 4: not a lawyer, it's not a professor, it's not a doctor.

559
00:29:28,320 --> 00:29:31,000
Speaker 4: But they're paying for all those things. So if we

560
00:29:31,000 --> 00:29:34,640
Speaker 4: could enable more people without as much training, and I

561
00:29:34,640 --> 00:29:37,440
Speaker 4: don't mean no judgment, I mean some training. If we

562
00:29:37,440 --> 00:29:39,320
Speaker 4: could allow paralegals to do more legal work, if we

563
00:29:39,360 --> 00:29:42,280
Speaker 4: could allow nurse practitioners to do a larger range of

564
00:29:42,320 --> 00:29:45,000
Speaker 4: medical tasks. If we could enable people who are doing

565
00:29:45,560 --> 00:29:48,760
Speaker 4: working as contractors also to do more design. Right, if

566
00:29:48,760 --> 00:29:51,800
Speaker 4: we're enabling people who don't have computer science degrees to

567
00:29:51,800 --> 00:29:54,400
Speaker 4: do more software development, not only would that reduce the

568
00:29:54,560 --> 00:29:56,960
Speaker 4: cost of these expensive services, but when I prove the

569
00:29:57,040 --> 00:29:59,680
Speaker 4: quality of work that people could do, it allow them

570
00:29:59,680 --> 00:30:04,160
Speaker 4: to take some expertise and make it go further. So

571
00:30:04,240 --> 00:30:05,320
Speaker 4: that's the good scenario.

572
00:30:05,680 --> 00:30:08,440
Speaker 1: Joe, I like the idea of using AI to reduce

573
00:30:08,600 --> 00:30:11,880
Speaker 1: emotional labor. I wonder if I can start automating some

574
00:30:12,400 --> 00:30:17,840
Speaker 1: responses on Twitter to toxic bitcoin maximlist. That's interesting, Tracy.

575
00:30:17,960 --> 00:30:21,760
Speaker 2: The block button is right there. So there's so many

576
00:30:21,840 --> 00:30:24,160
Speaker 2: different questions now that I have in my mind. But

577
00:30:24,400 --> 00:30:27,040
Speaker 2: you know, look, we're only near the beginning. I mean

578
00:30:27,160 --> 00:30:29,560
Speaker 2: chat GPT, which is sort of what's brought us all

579
00:30:29,600 --> 00:30:33,720
Speaker 2: into consciousness, was unveiled to the public in late twenty

580
00:30:33,800 --> 00:30:37,200
Speaker 2: twenty two, so not even two years into that this

581
00:30:37,360 --> 00:30:39,480
Speaker 2: sort of breakthrough that enabled it. As just a few

582
00:30:39,520 --> 00:30:43,400
Speaker 2: years older than that. The concern would be, well, yes,

583
00:30:43,480 --> 00:30:48,080
Speaker 2: at this point some training plus AI enables many people

584
00:30:48,160 --> 00:30:51,120
Speaker 2: to become much more productive and have this sort of

585
00:30:51,160 --> 00:30:54,640
Speaker 2: output that was previously associated with people with years of experience.

586
00:30:55,280 --> 00:30:58,960
Speaker 2: Like the fear would be that in multiple generations down

587
00:30:58,960 --> 00:31:03,520
Speaker 2: the road, you don't even need that initial training first.

588
00:31:03,720 --> 00:31:06,520
Speaker 4: I fully agreed. Word just at the beginning. The tools

589
00:31:06,560 --> 00:31:08,080
Speaker 4: are only so good, they're going to get much better.

590
00:31:08,360 --> 00:31:10,920
Speaker 4: Are Understanding how to use them is also very primitive.

591
00:31:10,960 --> 00:31:13,480
Speaker 4: We often don't know how to interact well with AI.

592
00:31:13,560 --> 00:31:15,520
Speaker 4: In fact, I could give you examples of cases where

593
00:31:15,520 --> 00:31:17,760
Speaker 4: it goes pretty badly, even though the tool is good.

594
00:31:18,320 --> 00:31:20,360
Speaker 4: So I think there are sort of two concerns built

595
00:31:20,360 --> 00:31:22,400
Speaker 4: into what you said. One is it basically, for now

596
00:31:22,520 --> 00:31:24,760
Speaker 4: it's a helper, and then eventually it's just your replacement.

597
00:31:25,400 --> 00:31:27,960
Speaker 4: And the other is that even if it just makes

598
00:31:28,000 --> 00:31:30,720
Speaker 4: everyone more efficient, eventually, well, we just saturate the world

599
00:31:30,760 --> 00:31:33,960
Speaker 4: with whatever that thing is, and then it's super cheap. Right. So,

600
00:31:34,000 --> 00:31:36,560
Speaker 4: there's only so many PowerPoint presentations the world can tolerate,

601
00:31:37,240 --> 00:31:39,240
Speaker 4: and if you get really fast at making them, eventually

602
00:31:39,240 --> 00:31:40,320
Speaker 4: people will pay you to stop.

603
00:31:41,360 --> 00:31:44,040
Speaker 2: Yes, we're there now, maybe, but anyway, keep going.

604
00:31:45,680 --> 00:31:48,160
Speaker 4: So I think that that will occur in some cases.

605
00:31:48,240 --> 00:31:50,320
Speaker 4: There's no question that in some cases the tool will

606
00:31:50,360 --> 00:31:53,120
Speaker 4: initially be a supplement and eventually be a replacement. Right.

607
00:31:53,240 --> 00:31:55,440
Speaker 4: So maybe air traffic controllers would be an example like that,

608
00:31:55,520 --> 00:31:58,240
Speaker 4: right where eventually almost all the air traffic control will

609
00:31:58,280 --> 00:32:01,200
Speaker 4: be done by machines. But I don't think every job

610
00:32:01,280 --> 00:32:03,640
Speaker 4: is like that. I don't think that's the case in medicine.

611
00:32:03,640 --> 00:32:06,960
Speaker 4: Medicine will be a hands on occupation for a very

612
00:32:07,000 --> 00:32:10,400
Speaker 4: long time. So will law, where there's a lot of

613
00:32:10,520 --> 00:32:13,600
Speaker 4: high stakes decision making, so will design. So I don't

614
00:32:13,760 --> 00:32:17,200
Speaker 4: think that we're going to automate everything away. I know

615
00:32:17,240 --> 00:32:19,800
Speaker 4: people think that, and I think it's a valid concern.

616
00:32:19,880 --> 00:32:22,400
Speaker 4: I don't think that's the most likely scenario. But I

617
00:32:22,400 --> 00:32:25,240
Speaker 4: also want to stress something that's said too little in

618
00:32:25,280 --> 00:32:28,000
Speaker 4: these discussions, which is, when you think about what you

619
00:32:28,080 --> 00:32:30,320
Speaker 4: can do with a new tool, most people think, well,

620
00:32:30,320 --> 00:32:32,240
Speaker 4: what can I automate? What is the thing that I'm

621
00:32:32,280 --> 00:32:34,440
Speaker 4: doing now that I could now have the machine do

622
00:32:34,600 --> 00:32:36,680
Speaker 4: for me? And that's important, and we do a lot

623
00:32:36,680 --> 00:32:40,560
Speaker 4: of automation, but automation is not the primary source of

624
00:32:40,640 --> 00:32:44,600
Speaker 4: how innovation improves our lives. Right. Many of the things

625
00:32:44,640 --> 00:32:48,560
Speaker 4: that we do with new tools is create new capabilities

626
00:32:48,960 --> 00:32:52,080
Speaker 4: that we didn't previously have. Right. So, airplanes did not

627
00:32:52,240 --> 00:32:54,680
Speaker 4: automate the way we used to fly. We just didn't

628
00:32:54,760 --> 00:32:58,120
Speaker 4: fly before we had airplanes. Right. The scanning electron microscope

629
00:32:58,280 --> 00:33:02,120
Speaker 4: didn't automate the way we used to look at subatomic particles.

630
00:33:02,440 --> 00:33:05,680
Speaker 4: We simply couldn't see them without that microscope. Right. So

631
00:33:06,120 --> 00:33:09,200
Speaker 4: think of the thought experiment of automating everything in Asian Greece,

632
00:33:09,560 --> 00:33:12,440
Speaker 4: you know, two thousand years ago. Even if you automated

633
00:33:12,520 --> 00:33:15,760
Speaker 4: everything in ancient Greece, it wouldn't be modern America, right,

634
00:33:16,080 --> 00:33:19,520
Speaker 4: It wouldn't have electricity, it wouldn't have computers, it wouldn't

635
00:33:19,560 --> 00:33:23,640
Speaker 4: have airplanes, it wouldn't have penicillin, it wouldn't have a

636
00:33:23,760 --> 00:33:27,560
Speaker 4: million tools technologies that we take for granted. So the

637
00:33:27,640 --> 00:33:31,400
Speaker 4: most important applications of technology are to enable capabilities that

638
00:33:31,480 --> 00:33:34,080
Speaker 4: didn't previously exist, and I think AI will do that

639
00:33:34,160 --> 00:33:36,440
Speaker 4: as well. So you know, we couldn't be having this

640
00:33:36,520 --> 00:33:39,840
Speaker 4: conversation were it not for our computers. Right. If someone

641
00:33:39,840 --> 00:33:41,720
Speaker 4: took my computer away from me, I couldn't even do

642
00:33:41,800 --> 00:33:44,320
Speaker 4: my job, right, It's just my job wouldn't exist in

643
00:33:44,360 --> 00:33:46,800
Speaker 4: its current form. And so what we do with new

644
00:33:46,800 --> 00:33:51,160
Speaker 4: technology is create new capabilities, and then human expertises often

645
00:33:51,240 --> 00:33:54,640
Speaker 4: needed to support those capabilities. Right, we didn't have pilots

646
00:33:54,680 --> 00:33:58,240
Speaker 4: before we had airplanes, and we didn't have pediatric oncologists

647
00:33:58,240 --> 00:34:01,080
Speaker 4: before we had all kinds of tools and knowledge to

648
00:34:01,280 --> 00:34:05,080
Speaker 4: treat cancer or cancer in children. And so as we

649
00:34:05,120 --> 00:34:10,040
Speaker 4: instantiate these new capabilities, we often require new human skills

650
00:34:10,040 --> 00:34:12,799
Speaker 4: and expertise that are valuable. And so much of what

651
00:34:12,840 --> 00:34:15,800
Speaker 4: we do with these tools is to change our lives

652
00:34:15,840 --> 00:34:18,839
Speaker 4: by pushing out the possibility set, rather than simply just

653
00:34:19,280 --> 00:34:21,640
Speaker 4: automating the things that we already do, and I think

654
00:34:21,640 --> 00:34:24,239
Speaker 4: AI will also be really important for that.

655
00:34:39,560 --> 00:34:42,480
Speaker 1: One thing I wanted to ask you is you are

656
00:34:42,800 --> 00:34:45,959
Speaker 1: very very clear in your piece that this is more

657
00:34:46,040 --> 00:34:51,000
Speaker 1: of an informed thesis than an actual forecast. And here

658
00:34:51,480 --> 00:34:53,880
Speaker 1: I am actually leaning on chat GPT when I asked

659
00:34:53,880 --> 00:34:56,960
Speaker 1: it to poke holes in your argument. One of the

660
00:34:56,960 --> 00:34:59,720
Speaker 1: ones it spat out had to do with this exact question.

661
00:35:00,600 --> 00:35:05,440
Speaker 1: Are there specific measures or policies that we could be

662
00:35:05,480 --> 00:35:09,600
Speaker 1: doing right now to make the probability of this outcome

663
00:35:10,280 --> 00:35:15,160
Speaker 1: better rather than the sort of like destructive AI dooomerism

664
00:35:15,200 --> 00:35:16,840
Speaker 1: outcome that everyone is worried about.

665
00:35:17,200 --> 00:35:20,280
Speaker 4: Yeah, so I appreciate your saying that the future should

666
00:35:20,320 --> 00:35:23,040
Speaker 4: not be treated as a forecasting or prediction exercise. It

667
00:35:23,040 --> 00:35:26,000
Speaker 4: should be treated as a design problem. Because the future

668
00:35:26,080 --> 00:35:27,640
Speaker 4: is not like the weather that we just wait and

669
00:35:27,640 --> 00:35:29,960
Speaker 4: see what happens. Right, We're making our own weather. We

670
00:35:30,000 --> 00:35:32,439
Speaker 4: have enormous control over the future in which we live

671
00:35:32,760 --> 00:35:36,320
Speaker 4: and depends on the investments and structures that we create today,

672
00:35:36,360 --> 00:35:39,719
Speaker 4: whether that's democracies, whether that's you know, education, whether that's

673
00:35:39,760 --> 00:35:42,239
Speaker 4: how we use tools and science that whether we use

674
00:35:42,239 --> 00:35:45,279
Speaker 4: fissionable material to make bombs or to make energy. Right,

675
00:35:45,280 --> 00:35:47,920
Speaker 4: we have lots and lots of agency here. So in

676
00:35:48,000 --> 00:35:51,200
Speaker 4: terms of using AI well, So first of all, let

677
00:35:51,239 --> 00:35:52,880
Speaker 4: me say what would be a metric how would we

678
00:35:52,960 --> 00:35:55,759
Speaker 4: know we were using AI? Well, because it's not like

679
00:35:55,800 --> 00:35:57,680
Speaker 4: carbon dioxide, where you know, we say, oh, we know

680
00:35:57,760 --> 00:35:59,799
Speaker 4: we're reducing carbon dioxide, you can just measure it, right,

681
00:35:59,800 --> 00:36:02,480
Speaker 4: How would we know we're using AI well, I would

682
00:36:02,520 --> 00:36:05,640
Speaker 4: say we know we're using it well when we see

683
00:36:05,880 --> 00:36:09,920
Speaker 4: people who don't have for your college degrees doing work

684
00:36:09,920 --> 00:36:12,560
Speaker 4: that we would think of as expert decision making work,

685
00:36:12,600 --> 00:36:16,600
Speaker 4: whether that's coding, whether that's you know, medical vocational work,

686
00:36:16,840 --> 00:36:20,760
Speaker 4: whether that's design and contracting, or even whether it allows

687
00:36:20,840 --> 00:36:23,680
Speaker 4: skilled repair people to work on a broader range of

688
00:36:23,800 --> 00:36:26,719
Speaker 4: products or tools or engines or whatever. So that's my

689
00:36:26,920 --> 00:36:31,279
Speaker 4: metric of success, that it opens up new job opportunities

690
00:36:31,280 --> 00:36:34,839
Speaker 4: to people who are not at the absolute elite of

691
00:36:34,880 --> 00:36:37,759
Speaker 4: a field. How do we get there? So I think

692
00:36:37,760 --> 00:36:41,240
Speaker 4: that's a super central question. And I think most thoughts

693
00:36:41,239 --> 00:36:44,320
Speaker 4: about you know, policies about AI are about regulating, controlling,

694
00:36:44,400 --> 00:36:46,800
Speaker 4: and some of that has to happen, and I feel

695
00:36:46,840 --> 00:36:51,000
Speaker 4: reasonably confident that it will. This is much more about investing, right,

696
00:36:51,360 --> 00:36:53,440
Speaker 4: So say, look, you know, in the United States, for example,

697
00:36:53,880 --> 00:36:57,759
Speaker 4: about twenty percent of GDP two in ten dollars goes

698
00:36:57,760 --> 00:37:01,200
Speaker 4: to education and healthcare. More than half that money is

699
00:37:01,239 --> 00:37:04,279
Speaker 4: public money, so in fact, we have a lot of

700
00:37:04,280 --> 00:37:08,200
Speaker 4: control over how education and healthcare delivered. So healthcare would

701
00:37:08,200 --> 00:37:09,920
Speaker 4: be the best place to start to say, all right,

702
00:37:10,040 --> 00:37:13,400
Speaker 4: let's redesign the tools or invest in the tools in

703
00:37:13,440 --> 00:37:16,520
Speaker 4: a way that enables more people to deliver this work.

704
00:37:16,560 --> 00:37:19,080
Speaker 4: And not only would that make better jobs, it would

705
00:37:19,080 --> 00:37:23,040
Speaker 4: also improve access to healthcare potentially lower those costs. We

706
00:37:23,040 --> 00:37:25,640
Speaker 4: could do the same in education. How can we make education,

707
00:37:26,280 --> 00:37:29,360
Speaker 4: you know, make better use of teachers, provide better services

708
00:37:29,400 --> 00:37:33,320
Speaker 4: to students, and also make education more accessible, immersive, engaging

709
00:37:33,360 --> 00:37:35,600
Speaker 4: for adults. Right, we have lots of adults who need

710
00:37:35,640 --> 00:37:38,120
Speaker 4: to learn, and traditional classrooms are really not the best

711
00:37:38,120 --> 00:37:39,960
Speaker 4: place to do that. So I do think you have

712
00:37:40,040 --> 00:37:43,399
Speaker 4: to think about these moonshots and governments can invest in them.

713
00:37:43,680 --> 00:37:45,759
Speaker 4: That doesn't mean the government has to run them, but

714
00:37:45,880 --> 00:37:50,160
Speaker 4: you know, governments often fund basic science, governments fund education.

715
00:37:50,640 --> 00:37:52,520
Speaker 4: Most health innovation in the United States is paid for

716
00:37:52,600 --> 00:37:54,919
Speaker 4: by the National Institute of Health, which is much much

717
00:37:54,960 --> 00:37:58,560
Speaker 4: larger than the National Science Foundation, for example, So I

718
00:37:58,600 --> 00:38:01,200
Speaker 4: think that is the biggest chance is to look for

719
00:38:01,200 --> 00:38:06,719
Speaker 4: those opportunities and then design with the intention of creating

720
00:38:07,120 --> 00:38:09,560
Speaker 4: a more effective way to structure work that uses the

721
00:38:09,600 --> 00:38:12,680
Speaker 4: tools and uses human skills better. And let me say,

722
00:38:12,719 --> 00:38:14,480
Speaker 4: you might say, well, you know, why doesn't this apply

723
00:38:14,520 --> 00:38:16,960
Speaker 4: equally well to the last era. So first of all,

724
00:38:16,960 --> 00:38:19,279
Speaker 4: we didn't design, and probably we should have done more.

725
00:38:19,320 --> 00:38:23,319
Speaker 4: But essentially, computers are good at following rules and so

726
00:38:23,840 --> 00:38:27,360
Speaker 4: they could replicate a lot of work that was just that,

727
00:38:27,520 --> 00:38:31,200
Speaker 4: but they weren't good at supplementing skills at enabling people

728
00:38:31,280 --> 00:38:35,480
Speaker 4: to do these high stakes decision making tasks. So it's important. Unsually,

729
00:38:35,520 --> 00:38:38,520
Speaker 4: AI is almost the inverse of traditional computing. Right if

730
00:38:38,560 --> 00:38:40,480
Speaker 4: I told you I have the most advanced technology in

731
00:38:40,480 --> 00:38:42,399
Speaker 4: the world, but you know, it really can't do math

732
00:38:42,400 --> 00:38:45,000
Speaker 4: and it's not reliable with facts and figures, you would say, well,

733
00:38:45,040 --> 00:38:46,799
Speaker 4: what kind of technology is that? And I would say, well,

734
00:38:46,800 --> 00:38:50,279
Speaker 4: that's artificial intelligence. It is really quite the opposite. So

735
00:38:50,360 --> 00:38:53,200
Speaker 4: I think it has quite different capabilities. And in some

736
00:38:53,239 --> 00:38:57,400
Speaker 4: sense you could say traditional computing was really complementary to

737
00:38:57,840 --> 00:39:00,600
Speaker 4: you know, the most elite professionals. And it's quite possible

738
00:39:00,640 --> 00:39:03,080
Speaker 4: that AI will enable more people to compete with them,

739
00:39:03,520 --> 00:39:07,719
Speaker 4: and that's a really good thing because that improves the

740
00:39:07,800 --> 00:39:10,040
Speaker 4: quality of services and improves the quality of jaws for

741
00:39:10,120 --> 00:39:12,280
Speaker 4: people who were not at that leading edge.

742
00:39:12,760 --> 00:39:14,839
Speaker 2: This is I think the key thing because in your

743
00:39:14,960 --> 00:39:17,640
Speaker 2: piece and and other testimony you've given, you've talked about

744
00:39:17,640 --> 00:39:20,640
Speaker 2: this idea of collective decision making, And when I think

745
00:39:20,640 --> 00:39:24,480
Speaker 2: about modern American society or modern society in general, I

746
00:39:24,520 --> 00:39:28,640
Speaker 2: don't necessarily think that collective decision making is something we're

747
00:39:28,680 --> 00:39:30,239
Speaker 2: particularly strong on.

748
00:39:30,400 --> 00:39:30,879
Speaker 4: So if the.

749
00:39:30,840 --> 00:39:34,879
Speaker 2: Future depends on making good collective decisions, then that makes

750
00:39:34,920 --> 00:39:38,160
Speaker 2: me anxious. But you know, you talk about investment, but

751
00:39:38,280 --> 00:39:41,960
Speaker 2: it sounds like the other element here. And you mentioned

752
00:39:42,000 --> 00:39:44,920
Speaker 2: that the rise of the nurse practitioner had to happen

753
00:39:45,040 --> 00:39:47,960
Speaker 2: over the kicking and screaming of the American Medical Association,

754
00:39:48,360 --> 00:39:52,040
Speaker 2: which represents that a top strata of healthcare professionals, the

755
00:39:52,280 --> 00:39:55,360
Speaker 2: elite doctors. How much of this is going to be

756
00:39:56,000 --> 00:40:00,880
Speaker 2: a political fight ultimately in which the doctor and the

757
00:40:01,000 --> 00:40:07,040
Speaker 2: lawyers and the podcasters collectively resist other people who are

758
00:40:07,120 --> 00:40:09,440
Speaker 2: using these tools to do our jobs. And how much

759
00:40:09,520 --> 00:40:12,319
Speaker 2: is that really like where the collective fight is going

760
00:40:12,360 --> 00:40:12,760
Speaker 2: to happen.

761
00:40:13,520 --> 00:40:15,360
Speaker 4: Yeah, if we have to take on the podcasters.

762
00:40:15,400 --> 00:40:18,360
Speaker 2: I think we're doomed, but yeah, yeah, we're going to

763
00:40:18,400 --> 00:40:20,000
Speaker 2: fight this kicking and screaming for sha.

764
00:40:20,040 --> 00:40:23,120
Speaker 4: The AMA is one thing, yeah, but the podcasters, that's

765
00:40:23,160 --> 00:40:26,040
Speaker 4: a whole different army. Some of that will absolutely be

766
00:40:26,120 --> 00:40:29,440
Speaker 4: terf warfare. Right the professions, we think, oh, you know that,

767
00:40:29,560 --> 00:40:31,840
Speaker 4: you know, oil companies and so on don't like competition

768
00:40:31,920 --> 00:40:33,839
Speaker 4: and they're always trying to rig the market, But in fact,

769
00:40:33,880 --> 00:40:36,680
Speaker 4: the professions rig the markets as well, right they what

770
00:40:36,880 --> 00:40:39,440
Speaker 4: a profession is, actually what it means is an occupation

771
00:40:39,480 --> 00:40:41,719
Speaker 4: that gets to certify its own members and decide who's

772
00:40:41,760 --> 00:40:45,280
Speaker 4: in and who's out right. And so it's the medical

773
00:40:45,400 --> 00:40:49,040
Speaker 4: profession that creates training standards and certification standards. It's universities

774
00:40:49,040 --> 00:40:52,360
Speaker 4: that decide what skills enable you to have a PhD

775
00:40:52,880 --> 00:40:57,160
Speaker 4: and therefore become a professor. So it absolutely is going

776
00:40:57,239 --> 00:40:59,960
Speaker 4: to be a challenge. Like lawyers will try very hard

777
00:41:00,080 --> 00:41:02,440
Speaker 4: say well, that can't be a legal document unless a

778
00:41:02,480 --> 00:41:05,799
Speaker 4: lawyer has signed it someone with a JD and has

779
00:41:05,880 --> 00:41:09,239
Speaker 4: passed the bar. So that will be a source of

780
00:41:09,280 --> 00:41:12,480
Speaker 4: resistance for sure. On the other hand, if there's a

781
00:41:12,560 --> 00:41:14,839
Speaker 4: really good competing alternative, if you can say, look, these

782
00:41:14,920 --> 00:41:18,240
Speaker 4: nurse practitioners can do a lot of this diagnostic work.

783
00:41:18,440 --> 00:41:20,120
Speaker 4: You know, they work well with doctors, but they can

784
00:41:20,160 --> 00:41:22,359
Speaker 4: do some things that doctors would be more expensive doing,

785
00:41:22,440 --> 00:41:24,920
Speaker 4: and you can make that case. Or a paralegal using

786
00:41:24,960 --> 00:41:28,239
Speaker 4: the software can create a lot of routine documents, or

787
00:41:28,360 --> 00:41:32,919
Speaker 4: a software developer using GitHub copilot can go pretty far.

788
00:41:33,520 --> 00:41:36,640
Speaker 4: Then that creates a lot of economic pressure that tends,

789
00:41:36,680 --> 00:41:40,560
Speaker 4: over long periods of time to erode these gills. So

790
00:41:40,719 --> 00:41:43,400
Speaker 4: I think that they will not go quietly into this

791
00:41:43,480 --> 00:41:47,759
Speaker 4: dark knight. But if the models are successful, it does

792
00:41:47,880 --> 00:41:52,200
Speaker 4: create a strong incentive for eventually that to become adopted.

793
00:41:52,840 --> 00:41:56,880
Speaker 1: I think part of the concern around AI has to

794
00:41:56,920 --> 00:42:01,839
Speaker 1: do also with how any productivity gains are actually distributed

795
00:42:01,880 --> 00:42:05,280
Speaker 1: and whether or not people are compensated for doing more.

796
00:42:06,080 --> 00:42:09,280
Speaker 1: And I asked chat GPT obviously to provide a summary

797
00:42:09,360 --> 00:42:12,080
Speaker 1: of dust capital before I came on here. No, I

798
00:42:12,120 --> 00:42:14,440
Speaker 1: do think there is this concern about Okay, in an

799
00:42:14,480 --> 00:42:17,840
Speaker 1: ideal scenario, we're all more efficient in terms of our labor,

800
00:42:18,200 --> 00:42:22,440
Speaker 1: and maybe some types of work are even better to perform.

801
00:42:22,520 --> 00:42:26,359
Speaker 1: Maybe we reduce that emotional labor. But aside from that

802
00:42:26,400 --> 00:42:31,200
Speaker 1: particular benefit, how do we distribute the additional productivity gains?

803
00:42:31,360 --> 00:42:34,960
Speaker 1: Is there any evidence or any reason to believe that

804
00:42:35,120 --> 00:42:39,440
Speaker 1: these benefits are going to go to labor, to actual

805
00:42:39,480 --> 00:42:42,360
Speaker 1: workers and individuals versus to companies in capital.

806
00:42:42,719 --> 00:42:44,960
Speaker 4: Yeah. Good. So let me give you two answers that question.

807
00:42:45,280 --> 00:42:49,440
Speaker 4: One is it really does depend on institutions, not just

808
00:42:49,680 --> 00:42:51,520
Speaker 4: on decentralized labor markets.

809
00:42:51,560 --> 00:42:51,680
Speaker 1: Right.

810
00:42:51,719 --> 00:42:55,799
Speaker 4: So if you compare the US versus Germany versus Scandinavia, right,

811
00:42:55,880 --> 00:42:58,520
Speaker 4: we have so much in common. We have the same technologies,

812
00:42:58,719 --> 00:43:00,680
Speaker 4: we have the same agent population, we have the same

813
00:43:00,760 --> 00:43:03,280
Speaker 4: rising education levels, we have the same China as a competitor,

814
00:43:03,400 --> 00:43:05,960
Speaker 4: we have lots of immigration. And yet these countries have big,

815
00:43:06,080 --> 00:43:08,200
Speaker 4: very different cakes with the same ingredients. Right. The US

816
00:43:08,280 --> 00:43:11,239
Speaker 4: is kind of cowboy capitalism, very high levels inequality and

817
00:43:11,320 --> 00:43:14,040
Speaker 4: disparity and not so much sharing with workers. And if

818
00:43:14,080 --> 00:43:17,400
Speaker 4: you look at Scandinavia or Germany, it's much more cuddly capitalism. Right,

819
00:43:17,400 --> 00:43:20,800
Speaker 4: it's not nearly as unequal. And that's really a question

820
00:43:20,920 --> 00:43:25,560
Speaker 4: of tax regulation, it's a question of the role of

821
00:43:25,680 --> 00:43:28,759
Speaker 4: labor unions and labor voice, and it's a question of

822
00:43:28,760 --> 00:43:30,960
Speaker 4: social norms. And so I guess we should not take

823
00:43:30,960 --> 00:43:33,400
Speaker 4: it as inevitable that the outcomes we have are the

824
00:43:33,400 --> 00:43:35,960
Speaker 4: only ones the market could tolerate. But at the same time,

825
00:43:35,960 --> 00:43:40,000
Speaker 4: we should recognize that without those sort of counterveling forces,

826
00:43:40,360 --> 00:43:43,279
Speaker 4: the outcomes can look pretty bad. So I do think

827
00:43:43,360 --> 00:43:47,720
Speaker 4: you know, I'm happy about the rise of collective bargaining

828
00:43:47,800 --> 00:43:49,799
Speaker 4: again in the United States, although it's from a very

829
00:43:49,800 --> 00:43:53,160
Speaker 4: low level. I'm happy that more states are passing minimum

830
00:43:53,200 --> 00:43:57,600
Speaker 4: wage regulations. I'm happy that the Biden administration is trying

831
00:43:57,640 --> 00:44:00,319
Speaker 4: to sort of beef up the Occtational Safety in Health

832
00:44:00,320 --> 00:44:04,400
Speaker 4: Administration and the Equal Employment Opportunity Commission and so on.

833
00:44:04,480 --> 00:44:06,759
Speaker 4: So I think those things matter a great deal. So

834
00:44:07,160 --> 00:44:09,200
Speaker 4: one should not take it for granted that just because

835
00:44:09,239 --> 00:44:12,560
Speaker 4: productivity rises, workers benefit in many countries. That's true, but

836
00:44:12,600 --> 00:44:13,840
Speaker 4: not so much in the United States.

837
00:44:13,840 --> 00:44:16,320
Speaker 2: But I want to press you right here on this point,

838
00:44:16,360 --> 00:44:20,239
Speaker 2: because why doesn't this undermine much of the argument. If

839
00:44:20,280 --> 00:44:23,280
Speaker 2: these different countries, whether it's Sweden and Germany the US,

840
00:44:23,680 --> 00:44:27,160
Speaker 2: can have very different sort of distributional outcomes with the

841
00:44:27,239 --> 00:44:31,280
Speaker 2: same cake ingredients, with roughly similar technology and labor markets,

842
00:44:31,800 --> 00:44:36,239
Speaker 2: why then take the assumption that it's the technology that

843
00:44:36,360 --> 00:44:40,800
Speaker 2: has the distributional impact rather than just those policies themselves.

844
00:44:41,440 --> 00:44:44,239
Speaker 4: Okay, this is an excellent question. So I think the

845
00:44:44,239 --> 00:44:48,000
Speaker 4: technology provides headwinds and tailwinds with which policy can work.

846
00:44:48,160 --> 00:44:50,800
Speaker 4: So all of these countries I mentioned have become more unequal.

847
00:44:51,040 --> 00:44:51,879
Speaker 3: Okay, all of.

848
00:44:51,800 --> 00:44:54,200
Speaker 4: These countries have seen a decline in middle scale work.

849
00:44:54,360 --> 00:44:57,439
Speaker 4: All of these countries have seen the mean wage rise

850
00:44:57,480 --> 00:45:00,480
Speaker 4: relative to the median, meaning the upper wages have risen

851
00:45:00,680 --> 00:45:03,080
Speaker 4: more than the center. But the degree to which countries

852
00:45:03,080 --> 00:45:06,080
Speaker 4: have pushed back against that is a function of their institutions.

853
00:45:06,440 --> 00:45:09,120
Speaker 4: In the prior era, prior to computerization, all of these

854
00:45:09,120 --> 00:45:11,840
Speaker 4: countries saw their middle classes grow together along with the

855
00:45:11,920 --> 00:45:15,279
Speaker 4: upper class and lower class, and so the industrial era

856
00:45:15,480 --> 00:45:19,759
Speaker 4: prior computerization was very friendly, sort of intrinsically towards the

857
00:45:19,760 --> 00:45:22,719
Speaker 4: middle class. The computer era was much much less so,

858
00:45:23,400 --> 00:45:26,640
Speaker 4: and then policy helped ameliorate those impacts, and much less

859
00:45:26,640 --> 00:45:28,440
Speaker 4: so in the United States. So I do think that

860
00:45:28,560 --> 00:45:32,120
Speaker 4: technology plays a role. I just we should simultaneously believe

861
00:45:32,320 --> 00:45:35,600
Speaker 4: that these underlying forces of technology and globalization creates strong

862
00:45:35,640 --> 00:45:38,080
Speaker 4: pressures in one way or another, and then policy can

863
00:45:38,120 --> 00:45:41,920
Speaker 4: shape how those pressures play out. It won't undo them,

864
00:45:41,960 --> 00:45:44,799
Speaker 4: but it can channel them more or less effectively. So

865
00:45:44,840 --> 00:45:47,200
Speaker 4: you're asking both the right questions, and I think the

866
00:45:47,280 --> 00:45:50,280
Speaker 4: answer is both are true, but we should think it's

867
00:45:50,360 --> 00:45:53,960
Speaker 4: not one or the other. And in some periods those

868
00:45:54,000 --> 00:45:56,680
Speaker 4: forces are very favorable and policy has to do less

869
00:45:56,680 --> 00:46:00,520
Speaker 4: hard work, and other periods they're relatively unfavorable. Policy, if

870
00:46:00,520 --> 00:46:03,040
Speaker 4: it's working well, has to do more work. The other

871
00:46:03,120 --> 00:46:04,319
Speaker 4: point I want to make, and this is why I'm

872
00:46:04,360 --> 00:46:08,160
Speaker 4: so focused on expertise, is expert work is intrinsically well paid.

873
00:46:08,600 --> 00:46:12,560
Speaker 4: It's scarce, and it's necessary. And that's why if we

874
00:46:12,640 --> 00:46:14,239
Speaker 4: live in a world where all the work can be

875
00:46:14,280 --> 00:46:19,279
Speaker 4: done by machines, we're completely dependent upon redistribution, right the

876
00:46:19,320 --> 00:46:22,200
Speaker 4: people who own the machines to share with everyone else.

877
00:46:22,239 --> 00:46:26,080
Speaker 4: And I'm not so optimistic about people's excitement about sharing

878
00:46:26,080 --> 00:46:28,000
Speaker 4: with everyone else. And even when people say, oh, we'll

879
00:46:28,000 --> 00:46:31,720
Speaker 4: have universal basic income, they really mean universal basic income

880
00:46:32,200 --> 00:46:34,279
Speaker 4: within the borders of the United States. They don't mean

881
00:46:34,360 --> 00:46:36,600
Speaker 4: universal basic income for the rest of the world. Right,

882
00:46:36,680 --> 00:46:40,200
Speaker 4: So people's notion of sharing is very limited. So I

883
00:46:40,480 --> 00:46:44,440
Speaker 4: do think it's extremely important that labor remains valuable, and

884
00:46:44,480 --> 00:46:48,640
Speaker 4: that's actually an achievement of the industrialized world that so

885
00:46:48,719 --> 00:46:52,440
Speaker 4: many people can make a good reasonable standard living based

886
00:46:52,520 --> 00:46:55,759
Speaker 4: on their skills, and so technologies and tools that make

887
00:46:55,840 --> 00:46:59,680
Speaker 4: human expertise more valuable by allowing to go further are

888
00:46:59,719 --> 00:47:04,880
Speaker 4: really favorable towards income distribution. Technologies that just automate away work,

889
00:47:05,400 --> 00:47:08,600
Speaker 4: even though they raise productivity, are not favorable to its

890
00:47:08,640 --> 00:47:11,560
Speaker 4: income distribution because it means it goes to ownership of capital,

891
00:47:11,560 --> 00:47:15,120
Speaker 4: and ownership of capital is intrinsically more centralized in ownership

892
00:47:15,160 --> 00:47:18,719
Speaker 4: of labor, because in a country that doesn't have slavery

893
00:47:18,920 --> 00:47:23,480
Speaker 4: and doesn't have labor coersion, everyone owns one worker themselves,

894
00:47:23,920 --> 00:47:28,799
Speaker 4: and so that inherently creates some tendency towards equality when

895
00:47:29,080 --> 00:47:30,120
Speaker 4: labor is valuable.

896
00:47:30,880 --> 00:47:35,160
Speaker 2: The efforts of the Biden administration to reindustrialize the US

897
00:47:35,239 --> 00:47:36,880
Speaker 2: and sort of counter some of the effects of the

898
00:47:36,960 --> 00:47:38,920
Speaker 2: last twenty years that you wrote about, do you have

899
00:47:38,960 --> 00:47:41,640
Speaker 2: any optimism that those trends can be reversed. I know

900
00:47:41,680 --> 00:47:44,000
Speaker 2: this is a very simple, straightforward question that you're going

901
00:47:44,040 --> 00:47:46,480
Speaker 2: to answer in about thirty seconds, so good luck.

902
00:47:46,560 --> 00:47:49,680
Speaker 4: I don't think they can be completely reversed, but you

903
00:47:49,760 --> 00:47:52,600
Speaker 4: can stem the tide, right, So it's not that the

904
00:47:52,680 --> 00:47:56,440
Speaker 4: U this has stabilized. The US continues to lose industrial capacity, right,

905
00:47:56,480 --> 00:47:59,560
Speaker 4: whether it's in semiconductors, whether it's automobiles, whether it's an aircraft,

906
00:47:59,600 --> 00:48:04,080
Speaker 4: Thank you, and so on. So I think reinvesting can

907
00:48:04,440 --> 00:48:07,480
Speaker 4: help solidify those sectors, and I think it's very important

908
00:48:07,520 --> 00:48:09,280
Speaker 4: to do so, because now they're not just a question

909
00:48:09,320 --> 00:48:12,680
Speaker 4: of jobs. It really is about leadership of the key

910
00:48:13,120 --> 00:48:18,000
Speaker 4: profit and idea generating activities in the modern world, and

911
00:48:18,040 --> 00:48:22,520
Speaker 4: we don't want to lose a leadership place in those activities.

912
00:48:23,440 --> 00:48:27,680
Speaker 2: Good concise answer to what probably could be multiple future episodes,

913
00:48:27,960 --> 00:48:30,959
Speaker 2: David Otter, Thank you so much for coming on out Laws.

914
00:48:31,040 --> 00:48:33,640
Speaker 2: That really was a fascinating conversation. We probably could get

915
00:48:33,719 --> 00:48:37,359
Speaker 2: multiple episodes out of this conversation with you, but really

916
00:48:37,360 --> 00:48:38,200
Speaker 2: appreciate your time.

917
00:48:38,360 --> 00:48:40,120
Speaker 4: Thank you very much. Nice to speak with both of you.

918
00:48:40,120 --> 00:48:54,880
Speaker 3: Have a good day, Tracy. I'm convinced. I think everything

919
00:48:54,920 --> 00:48:55,399
Speaker 3: will be fun.

920
00:48:55,640 --> 00:48:57,240
Speaker 2: I'm no longer worried.

921
00:48:57,360 --> 00:48:59,479
Speaker 1: Well, first of all, I would say it was nice

922
00:48:59,520 --> 00:49:03,640
Speaker 1: to hear a slightly more optimistic argument from David. There

923
00:49:03,680 --> 00:49:06,400
Speaker 1: were a lot of quotable sentences in there. So I

924
00:49:06,480 --> 00:49:09,680
Speaker 1: like the idea that everyone's their own individual capitalist in

925
00:49:09,719 --> 00:49:12,400
Speaker 1: the sense that we each have one worker to direct

926
00:49:12,600 --> 00:49:15,560
Speaker 1: and get the most money out of So that's how

927
00:49:15,560 --> 00:49:19,759
Speaker 1: I'm going to start thinking cuddly capitalism, which, as our

928
00:49:19,840 --> 00:49:23,080
Speaker 1: producer Klee observes, is a much more appealing name than

929
00:49:23,120 --> 00:49:26,280
Speaker 1: the Swedish model. I like that. What I would say

930
00:49:26,800 --> 00:49:31,640
Speaker 1: is again putting on my cynical journalist hat, and I

931
00:49:31,640 --> 00:49:34,000
Speaker 1: guess I don't have an opinion because I am a

932
00:49:34,080 --> 00:49:37,840
Speaker 1: journalist whose expertise is about to be automated away. But

933
00:49:38,400 --> 00:49:42,200
Speaker 1: my non consensus take here, or my sort of hot

934
00:49:42,280 --> 00:49:45,319
Speaker 1: take here, is that I agree with David that we

935
00:49:45,440 --> 00:49:48,359
Speaker 1: are going to get more jobs out of AI, and

936
00:49:48,400 --> 00:49:52,239
Speaker 1: probably more than a lot of people currently anticipate. I

937
00:49:52,280 --> 00:49:57,000
Speaker 1: guess I'm less convinced about how useful those jobs are

938
00:49:57,040 --> 00:49:59,160
Speaker 1: going to be, So going back to his point about

939
00:49:59,160 --> 00:50:02,480
Speaker 1: how do we measure how well we're using AI, I

940
00:50:02,520 --> 00:50:05,160
Speaker 1: have a feeling that a lot of it is going

941
00:50:05,200 --> 00:50:07,799
Speaker 1: to end up basically creating a whole new layer of

942
00:50:08,280 --> 00:50:10,840
Speaker 1: BS jobs that don't actually do much. So there's going

943
00:50:10,920 --> 00:50:14,520
Speaker 1: to be all these decision making bodies attached to AI.

944
00:50:14,960 --> 00:50:19,520
Speaker 1: There's going to be big discussions about how you implement AI, fairness,

945
00:50:19,800 --> 00:50:22,839
Speaker 1: litigating its results, and things like that. I guess I'm

946
00:50:22,840 --> 00:50:25,560
Speaker 1: a little bit pessimistic about the ability of AI to

947
00:50:25,719 --> 00:50:29,640
Speaker 1: generate additional bureaucracy in addition to additional productivity.

948
00:50:30,120 --> 00:50:32,440
Speaker 2: The other term that was great was when you said

949
00:50:32,480 --> 00:50:34,960
Speaker 2: the future is not like the weather. Yeah, but also

950
00:50:35,320 --> 00:50:38,160
Speaker 2: like I am worried about any notion that to achieve

951
00:50:38,239 --> 00:50:41,719
Speaker 2: the good outcome, the good equilibrium, we have to make good,

952
00:50:41,840 --> 00:50:45,680
Speaker 2: correct collective decisions because I have almost zero confidence in

953
00:50:45,960 --> 00:50:48,640
Speaker 2: whether it's just the US specifically or globally to make

954
00:50:49,040 --> 00:50:52,600
Speaker 2: collective decisions. I do think like going after like these

955
00:50:52,840 --> 00:50:56,520
Speaker 2: guilds like the American Medical Association, which for all of

956
00:50:56,560 --> 00:50:58,960
Speaker 2: the rise of nurse practitioners, it doesn't seem like we're

957
00:50:59,000 --> 00:51:01,560
Speaker 2: doing great on like ben the cost of healthcare jobs

958
00:51:01,680 --> 00:51:05,040
Speaker 2: or really having a healthcare capacity. That's going to be

959
00:51:05,040 --> 00:51:07,000
Speaker 2: like really tough. And those fights are going to be

960
00:51:07,080 --> 00:51:10,200
Speaker 2: really intense, whether it's with lawyers, whether it's with doctors,

961
00:51:10,239 --> 00:51:13,319
Speaker 2: whether it's with teachers, whether it's with podcasters, whether it's

962
00:51:13,320 --> 00:51:16,680
Speaker 2: professional architects, et cetera. Like, those fights are going to

963
00:51:16,719 --> 00:51:22,520
Speaker 2: be extremely intense. But like the basic intuition sounds very

964
00:51:22,560 --> 00:51:24,759
Speaker 2: compelling to me. The other thing is like you know

965
00:51:24,800 --> 00:51:27,440
Speaker 2: this idea of like oh, yeah, some training plus AI,

966
00:51:27,480 --> 00:51:29,440
Speaker 2: like I am worried, like maybe you just won't need

967
00:51:29,440 --> 00:51:31,920
Speaker 2: the training, and maybe it's just AI from the start.

968
00:51:32,040 --> 00:51:33,160
Speaker 3: So I don't know.

969
00:51:33,640 --> 00:51:35,759
Speaker 1: Well, in some respects, I think that would almost be

970
00:51:36,320 --> 00:51:40,719
Speaker 1: a better outcome in terms of democratizing AI. But yeah,

971
00:51:40,719 --> 00:51:44,120
Speaker 1: there are so many questions, uncertainty, as you mentioned in

972
00:51:44,160 --> 00:51:47,279
Speaker 1: the intro, lots of different takes at the moment. I

973
00:51:47,320 --> 00:51:49,880
Speaker 1: guess we'll see how it plays out and whether or

974
00:51:49,920 --> 00:51:52,200
Speaker 1: not you and I have jobs in ten years time.

975
00:51:52,239 --> 00:51:54,879
Speaker 2: We'll see, Well, we'll have David back when we're just like.

976
00:51:55,000 --> 00:51:58,440
Speaker 1: When we're automated voices. Yeah, exactly, all right, shall we

977
00:51:58,520 --> 00:51:59,239
Speaker 1: leave it there for now?

978
00:51:59,320 --> 00:52:00,000
Speaker 3: Let's leave it there.

979
00:52:00,120 --> 00:52:03,160
Speaker 1: Okay. This has been another episode of the Authoughts podcast.

980
00:52:03,239 --> 00:52:06,000
Speaker 1: I'm Tracy Alloway. You can follow me at Tracy Alloway.

981
00:52:06,120 --> 00:52:08,800
Speaker 2: And I'm Joe Wisenthal. You can follow me at the Stalwart.

982
00:52:09,040 --> 00:52:12,400
Speaker 2: Follow our guest David Otter, He's at David Otter. Follow

983
00:52:12,440 --> 00:52:16,200
Speaker 2: our producers Carmen Rodriguez at Carman Arman dashl Bennett at

984
00:52:16,320 --> 00:52:19,000
Speaker 2: Dashbot and kel Brooks at kel Brooks. And thank you

985
00:52:19,040 --> 00:52:22,400
Speaker 2: to our producer Moses Ondam. Form our oddlogs content go

986
00:52:22,440 --> 00:52:24,680
Speaker 2: to Bloomberg dot com slash odd Lots, where we have

987
00:52:24,719 --> 00:52:27,160
Speaker 2: a blog, we post transcripts, and we have a weekly

988
00:52:27,239 --> 00:52:30,279
Speaker 2: newsletter and you can chat with fellow listeners twenty four

989
00:52:30,280 --> 00:52:33,400
Speaker 2: to seven in the discord discord dot gg slash odd Logs.

990
00:52:33,400 --> 00:52:36,160
Speaker 2: There's even an AI room in there where people are

991
00:52:36,160 --> 00:52:38,960
Speaker 2: talking about all these things. So imagine there will be

992
00:52:39,000 --> 00:52:40,479
Speaker 2: some conversation about this there.

993
00:52:40,719 --> 00:52:43,200
Speaker 1: And if you enjoy odd Lots, if you like it

994
00:52:43,239 --> 00:52:46,240
Speaker 1: when we do deep dives into AI, how it works,

995
00:52:46,280 --> 00:52:48,800
Speaker 1: what it means for the economy and society, then please

996
00:52:48,920 --> 00:52:52,239
Speaker 1: leave us a positive review on your favorite podcast platform.

997
00:52:52,440 --> 00:52:55,279
Speaker 1: And remember, if you are a Bloomberg subscriber, you can

998
00:52:55,320 --> 00:52:58,879
Speaker 1: listen to all of our episodes absolutely ad free. All

999
00:52:58,920 --> 00:53:01,520
Speaker 1: you need to do is connect to your Bloomberg subscription

1000
00:53:01,680 --> 00:53:04,080
Speaker 1: to Apple Podcasts. Thanks for listening.