Latest / Trumponomics / AI Is Being Built to Replace You—Not Help You

Transcript

AI Is Being Built to Replace You—Not Help You · Trumponomics

• 0:02
  Speaker 1: Bloomberg Audio Studios, Podcasts, radio News. On our current path,
• 0:09
  Speaker 1: this handful of people will decide what the future of
• 0:12
  Speaker 1: AI is, and the best way to counter balance that
• 0:17
  Speaker 1: is to have a vision that's different and hopefully better
• 0:21
  Speaker 1: for society.
• 0:31
  Speaker 2: Welcome to Trumpnomics, the podcast that looks at the economic
• 0:33
  Speaker 2: world of Donald Trump, how he's already shaped the global
• 0:36
  Speaker 2: economy and modern earth is going to.
• 0:38
  Speaker 1: Happen next.
• 0:41
  Speaker 2: Well, we talk about the big forces affecting our economy
• 0:43
  Speaker 2: in the broader world on this show. And there's no
• 0:45
  Speaker 2: bigger topic in economic policy and general discussion these days
• 0:49
  Speaker 2: than the impact of AI. We've talked at different times
• 0:53
  Speaker 2: about the consequences for jobs, inflation, interest rates, and where
• 0:57
  Speaker 2: the policy makers, let alone ordinary people, ready for any
• 1:00
  Speaker 2: of it. Well, my guest this week, Darren Asimoglu, famously
• 1:04
  Speaker 2: takes the long view on these matters. A recipient of
• 1:06
  Speaker 2: the Nobel Prize for Economics in twenty twenty four, he
• 1:10
  Speaker 2: probably wrote the single most widely read book of economic
• 1:13
  Speaker 2: history of recent times, Why Nations Fail, And more recently
• 1:16
  Speaker 2: he wrote with Simon Johnson, Power and Progress, Our thousand
• 1:20
  Speaker 2: years struggle over technology and prosperity, and I talked to
• 1:24
  Speaker 2: him about that book and the lessons for this AI
• 1:27
  Speaker 2: revolution a while back in the summer of twenty twenty three,
• 1:31
  Speaker 2: but given everything that's been going on, I wanted to
• 1:33
  Speaker 2: have him back to see whether he saw anything in
• 1:37
  Speaker 2: the new waves of AI that we've had since twenty
• 1:41
  Speaker 2: twenty three, particularly in the last few months, whether he'd
• 1:44
  Speaker 2: seen anything to make him change his view on either
• 1:47
  Speaker 2: how fast this technology is going to change our economy
• 1:50
  Speaker 2: or how well placed we are to get the best
• 1:53
  Speaker 2: out of it. Darren, thanks very much for coming back
• 1:56
  Speaker 2: on to this podcast.
• 1:57
  Speaker 1: Thank you. Stephanie's my pleasure to be here.
• 2:04
  Speaker 2: We will get into some of the sort of key
• 2:08
  Speaker 2: dimensions of this in a minute, but I should just
• 2:10
  Speaker 2: get a sense from you. I mean, we're talking now
• 2:13
  Speaker 2: mid March twenty twenty six. There's been so much chatter,
• 2:17
  Speaker 2: and I suspect many people listening have had their own
• 2:19
  Speaker 2: real life experience now of the development of all the
• 2:24
  Speaker 2: different forms of AI, and particularly the sort of agentic
• 2:27
  Speaker 2: AI that we talk about. Are you, in a kind
• 2:30
  Speaker 2: of very broad sense reassessing how fast or how fundamentally
• 2:35
  Speaker 2: this is going to change our world?
• 2:37
  Speaker 1: Yeah? Every day, I think the underlying technology is changing
• 2:42
  Speaker 1: faster than what I would have predicted, what many would
• 2:46
  Speaker 1: have predicted eight and a half ago. So, especially with
• 2:52
  Speaker 1: the recent developments in agentic AI, especially led by Anthropic,
• 2:58
  Speaker 1: there is a real possible ability that these tools can
• 3:02
  Speaker 1: be broadly useful in what people do. There is still
• 3:07
  Speaker 1: a lot of uncertainty, however. First we are not seeing
• 3:11
  Speaker 1: any of the prepackaged, easy to use, reliable applications think
• 3:17
  Speaker 1: of it as the Microsoft Words or Microsoft Offices of
• 3:21
  Speaker 1: AI that can be used across a broad range of
• 3:25
  Speaker 1: occupations or in some specific occupations. Those are not around yet.
• 3:29
  Speaker 1: There is still uncertainty about whether there will be bottlenecks
• 3:34
  Speaker 1: in reaching higher reliability and higher judgment. There is every
• 3:40
  Speaker 1: evidence that there is a lot of rapid progress, but
• 3:45
  Speaker 1: there are some weaknesses in these models that are still persistent,
• 3:50
  Speaker 1: and I don't just mean hallucinations, but lack of a
• 3:54
  Speaker 1: deep understanding. They don't seem to have a conceptual framework,
• 3:57
  Speaker 1: they don't understand the context, and they cannot reason at
• 4:01
  Speaker 1: multiple levels of abstraction about a problem. Yet so those
• 4:06
  Speaker 1: may be overcome, and I think many of those are
• 4:08
  Speaker 1: going to be important in dealing with edge cases in
• 4:12
  Speaker 1: many occupations. So wholesale automation of occupations is still not
• 4:19
  Speaker 1: something we're going to see right away, but some people
• 4:23
  Speaker 1: swear that we're going to see it in one year
• 4:25
  Speaker 1: or two years, three years. So there's a lot of uncertainty.
• 4:28
  Speaker 1: But let me make one thing clear. If we do
• 4:33
  Speaker 1: not up our game about both how we regulate these
• 4:39
  Speaker 1: models and how we actually develop them, there could be
• 4:43
  Speaker 1: a huge amount of damage to society.
• 4:45
  Speaker 2: I think that's very helpful because there's two elements of
• 4:48
  Speaker 2: this where there's obviously, as you say, there's a lot
• 4:51
  Speaker 2: of uncertainty, and there's a wide range of opinion. If possible,
• 4:54
  Speaker 2: I'm going to try and separate them, but obviously they
• 4:56
  Speaker 2: merge into each other. One is this question of the
• 4:58
  Speaker 2: pace of change, faster company is really going to be
• 5:01
  Speaker 2: able to change their practices or capture those productivity improvements.
• 5:04
  Speaker 2: And then the second is which you've just highlighted, is
• 5:07
  Speaker 2: how well are we positioned to make the best of this,
• 5:10
  Speaker 2: not just to get all the productivity growth, but to
• 5:13
  Speaker 2: make sure it's actually positive for most of the population,
• 5:16
  Speaker 2: not just a few. And you highlighted in the twenty
• 5:18
  Speaker 2: twenty three book, you know that none of that was
• 5:20
  Speaker 2: automatic in the case of the industrial revolution, and we
• 5:22
  Speaker 2: may have to do it much faster this time. Just
• 5:25
  Speaker 2: focusing on this speed question, there's a Citrini report which
• 5:29
  Speaker 2: there's been a little mini industry in sort of debunking
• 5:31
  Speaker 2: this research report which sort of went viral because it
• 5:35
  Speaker 2: captured some elements of this sort of faster pace that
• 5:37
  Speaker 2: we were seeing. I think you wrote a paper, you know,
• 5:39
  Speaker 2: the basic Macroeconomics of AI. In the debate, I would
• 5:43
  Speaker 2: say that you're fairly low key about the pace of change,
• 5:47
  Speaker 2: the extent of change in any given year. I think
• 5:49
  Speaker 2: you said at most a few percent over ten years,
• 5:53
  Speaker 2: so maybe even a fraction of a percent of productivity
• 5:55
  Speaker 2: growth overall productivity growth a year. Would you stand by
• 6:00
  Speaker 2: that basic assessment today or do you think maybe the gains,
• 6:03
  Speaker 2: just the purebreductivity gains could be a bit faster.
• 6:06
  Speaker 1: I think they could be a bit faster. There has
• 6:09
  Speaker 1: been faster change of the capabilities of the foundation models. However,
• 6:18
  Speaker 1: it would still require some big breakthroughs, especially at the
• 6:22
  Speaker 1: application layer. So the bottom line of that paper was
• 6:29
  Speaker 1: to point out how we can get a fairly simple
• 6:34
  Speaker 1: understanding of the constituent parts of the contribution of AI
• 6:40
  Speaker 1: to productivity and GDP growth, And that comes from realizing
• 6:46
  Speaker 1: that the GDP contribution of AI is nothing other than
• 6:50
  Speaker 1: what fraction of tasks are going to be taken over
• 6:53
  Speaker 1: or completely transformed by AI in the economy times the
• 6:57
  Speaker 1: average productivity gain or average cost savings in these tasks.
• 7:01
  Speaker 1: So that's the calculation that I did with the available
• 7:04
  Speaker 1: evidence in twenty twenty two. But even then a lot
• 7:08
  Speaker 1: of people took issue at how I interpreted the data, etc.
• 7:14
  Speaker 1: So you could boost some of the numbers that I have,
• 7:17
  Speaker 1: which were that about five percent of the whole economy
• 7:20
  Speaker 1: will be taken by AI within ten years, by twenty
• 7:24
  Speaker 1: thirty or thereabouts, and that that would lead to about
• 7:29
  Speaker 1: twenty five percent cost savings or relative to labor costs
• 7:35
  Speaker 1: that firms used to spend on the same tasks. Now
• 7:39
  Speaker 1: you can boost my numbers by increasing either or both
• 7:45
  Speaker 1: of these quantities. So you can say, no, no, it's
• 7:48
  Speaker 1: not five percent, it's going to be twenty percent of
• 7:51
  Speaker 1: the economy that AI is going to take over, in
• 7:53
  Speaker 1: which case you would quadruple my numbers. Or you could
• 7:56
  Speaker 1: say it's not twenty percent cost savings, but it's going
• 8:00
  Speaker 1: to lead to thirty percent or forty percent cost savings.
• 8:03
  Speaker 1: After all, you know, comp lead to three hundred percent
• 8:06
  Speaker 1: cost savings because you know labor wasn't very expensive anyway.
• 8:09
  Speaker 1: But you see the elbow room to do that kind
• 8:12
  Speaker 1: of thing. But you're not going to come up with
• 8:14
  Speaker 1: revolutionary numbers here. And part of the problem here is
• 8:18
  Speaker 1: that we are right now, and that was the case
• 8:22
  Speaker 1: two years ago and continues to be the case. We
• 8:25
  Speaker 1: are right now focusing on AI as an automation tool,
• 8:29
  Speaker 1: as a tool to replace workers. That's not the best
• 8:33
  Speaker 1: way of using AI. The best way of using AI
• 8:36
  Speaker 1: is to try to complement workers so that they can
• 8:39
  Speaker 1: do new things, they can perform new tasks, they can
• 8:43
  Speaker 1: increase their sophistication level, and also respond to challenges in
• 8:48
  Speaker 1: the world economy from globalization, from aging, from climate change
• 8:52
  Speaker 1: by creating new goods and services, new organizations and so on.
• 8:56
  Speaker 1: If we do that, I think I would be more
• 8:58
  Speaker 1: optimistic about the future away. And it's one of the
• 9:02
  Speaker 1: aspects of the wisdom gap that we have right now.
• 9:05
  Speaker 1: We don't know how to regulate existing models, and we
• 9:07
  Speaker 1: are not really focusing on what we can do best
• 9:10
  Speaker 1: with these models. And that's both for productivity and social consequences.
• 9:13
  Speaker 1: So I've just made the productivity case that we could
• 9:16
  Speaker 1: actually get better productivity consequences. But actually the social consequences
• 9:19
  Speaker 1: are even starker. If we displace people, if we say
• 9:23
  Speaker 1: displaced twenty percent of the population from their jobs and
• 9:25
  Speaker 1: they remain unemployed or they go to lower quality, lower
• 9:28
  Speaker 1: pay jobs, our democracy is not going to survive. We're
• 9:32
  Speaker 1: already struggling to make our democratic system. Yeah, we're not
• 9:37
  Speaker 1: doing that well. And if we put another huge shock
• 9:41
  Speaker 1: on top of that, I'm not very optimistic.
• 9:43
  Speaker 2: So beware and just thinking about the economics of what
• 9:47
  Speaker 2: you're saying, and also thinking about what captured people's imaginations
• 9:51
  Speaker 2: about that Suitrinian report. They say themselves, this is a
• 9:53
  Speaker 2: thought experiment. There was something kind of gripping about the
• 9:56
  Speaker 2: fact that it was claiming to be a memo written
• 9:58
  Speaker 2: in twenty twenty eight. The claim was you would have
• 10:01
  Speaker 2: an extraordinary amount of change in business models in a
• 10:03
  Speaker 2: very short period of time. I assume you would say,
• 10:07
  Speaker 2: given real world frictions and just the way things tend
• 10:12
  Speaker 2: to happen, that a two year timeframe is very unrealistic.
• 10:16
  Speaker 2: But there was another basic assumption built into that that
• 10:19
  Speaker 2: the change that we will most immediately see and will
• 10:21
  Speaker 2: have the biggest impact will be simply to replace labor,
• 10:24
  Speaker 2: not to augment it, and that to the extent that
• 10:27
  Speaker 2: it's creating other stuff that's going to be far outweighed
• 10:32
  Speaker 2: by the job destruction. Even if you don't accept the
• 10:35
  Speaker 2: time frame, do you think there is an emphasis on
• 10:38
  Speaker 2: replacement relative to augmentation or is it a gent ki
• 10:42
  Speaker 2: really both things.
• 10:43
  Speaker 1: I don't know. It's an open question, but my bet
• 10:48
  Speaker 1: would be on the Cititrini side, not on the timeframe,
• 10:52
  Speaker 1: but on the path that we are following. There is
• 10:56
  Speaker 1: so little that these companies are doing in order to
• 11:01
  Speaker 1: understand what work humans do and try to be useful
• 11:05
  Speaker 1: to humans. The whole agenda of all of the leading
• 11:09
  Speaker 1: companies in the United States and now joined by deep
• 11:11
  Speaker 1: Seek in China is agi artificial general intelligence. That is
• 11:17
  Speaker 1: a banner for saying these models are going to do
• 11:21
  Speaker 1: everything better than humans, which of course then leads to
• 11:24
  Speaker 1: the corollary that a lot of companies should just throw
• 11:28
  Speaker 1: away their humans and use these companies. That is an
• 11:30
  Speaker 1: automation agenda that is exactly what Catrini banked on. Now,
• 11:35
  Speaker 1: they then made a number of other assumptions and steps
• 11:39
  Speaker 1: about how that would work out, what its consequences would be,
• 11:42
  Speaker 1: how quickly those would be. Those I don't agree with,
• 11:45
  Speaker 1: but credit to them. They said this was a scenario.
• 11:49
  Speaker 1: They weren't even making a prediction. I don't know why
• 11:51
  Speaker 1: the markets went haywire, given that there was no new
• 11:55
  Speaker 1: information in there. Everything that was in the Sutrini report
• 11:57
  Speaker 1: has been said, and they themselves said, there is no
• 12:00
  Speaker 1: search here that's original. But you know, we are living
• 12:03
  Speaker 1: in such fragile times in everything. You know, the valuation
• 12:08
  Speaker 1: of these companies is all based on very fragile assumptions
• 12:12
  Speaker 1: about what they're going to be able to achieve in
• 12:14
  Speaker 1: the future. If you look at the amount that they're
• 12:17
  Speaker 1: spending and their valuations, this can only be justify if
• 12:21
  Speaker 1: they make something like a trillion dollar revenues in the
• 12:25
  Speaker 1: foreseeable future as an AI industry. I mean, that's just incredible.
• 12:28
  Speaker 1: How are you going to get to a trillion dollar there?
• 12:31
  Speaker 1: They're hardly struggling to make a couple of billion dollars
• 12:33
  Speaker 1: right now as a whole industry. So there is something,
• 12:37
  Speaker 1: you know, glass in this house.
• 12:41
  Speaker 2: Okay, I have to say that I'm slightly depressed by
• 12:42
  Speaker 2: that answer because I thought you were going to push
• 12:44
  Speaker 2: back more heavily against the Citrey assessment. Though I absolutely
• 12:49
  Speaker 2: know that you're I know you have your concerned about
• 12:52
  Speaker 2: our capacity to to cape with this, but I thought
• 12:55
  Speaker 2: I let.
• 12:57
  Speaker 1: Me give me out of the pushback as well. I mean,
• 12:59
  Speaker 1: that's that's the point I want to make throughout. There
• 13:02
  Speaker 1: is the potential to use AI not for automation. That's
• 13:06
  Speaker 1: what I keep emphasizing. But I also want to push
• 13:11
  Speaker 1: very hard against the assumption that either we are going
• 13:15
  Speaker 1: there already, No we're not, or that we can get
• 13:17
  Speaker 1: there automatically. No, we cannot. Because all of these companies
• 13:21
  Speaker 1: have this business model or just let's repoliticplace all the workers.
• 13:25
  Speaker 1: They haven't even put into their calculations much of a
• 13:30
  Speaker 1: revenue stream that they can get from complementing workers, because
• 13:34
  Speaker 1: that's just a very difficult thing to monetize. So I
• 13:36
  Speaker 1: think that's where our wisdom gap is. We are not
• 13:41
  Speaker 1: even wisely thinking about what we should be doing with
• 13:45
  Speaker 1: these very capable models, and the industry is going in
• 13:49
  Speaker 1: its own direction.
• 13:51
  Speaker 2: You have just done a paper with two of your
• 13:54
  Speaker 2: colleagues for Brookings that is trying to give some concrete
• 13:57
  Speaker 2: advice to policymakers in this area to answer specifically that question.
• 14:00
  Speaker 2: I do want to get to that. I just want
• 14:02
  Speaker 2: to quickly one of the things, just to make sure
• 14:05
  Speaker 2: coming out of this conversation. One of the things that
• 14:07
  Speaker 2: we see a lot in this is, and particularly if
• 14:10
  Speaker 2: you look at the research studies in this area, they
• 14:14
  Speaker 2: tend to look at the range of occupations and talk
• 14:17
  Speaker 2: about their degree of exposure quote unquote to AI. And
• 14:21
  Speaker 2: there's a whole range, I think in your original assessment,
• 14:23
  Speaker 2: and a lot of people use this. They sort of
• 14:25
  Speaker 2: thought it was about twenty percent of occupations, and obviously
• 14:28
  Speaker 2: some people have high numbers. If one's thinking about different
• 14:31
  Speaker 2: kinds of AI and the potential of different of AI,
• 14:34
  Speaker 2: how should we read those? Are they just going on
• 14:37
  Speaker 2: your the way that businesses is looking at it now,
• 14:39
  Speaker 2: as you pointed out as very much as a labor
• 14:42
  Speaker 2: replacement technology, should we see that as exposure to replacement
• 14:46
  Speaker 2: or should we see it something more sophisticated?
• 14:48
  Speaker 1: Great set of questions. There are really three questions you're
• 14:51
  Speaker 1: asking here, Stephanie. Let me answer each one of them
• 14:53
  Speaker 1: in turn. One is what does this AI exposure mean?
• 14:57
  Speaker 1: And in general it is an ill defined concept because
• 15:02
  Speaker 1: you could be exposed to AI because you can lose
• 15:05
  Speaker 1: your job with AI, or you could be exposed to
• 15:08
  Speaker 1: AI because you could use AI to increase your contribution
• 15:12
  Speaker 1: to your job.
• 15:13
  Speaker 2: And we see that in the company's exposure as well.
• 15:15
  Speaker 2: And investors can't decide the difference between those two either.
• 15:19
  Speaker 1: That's why they fluctuate between devaluing software companies and giving
• 15:24
  Speaker 1: them a huge boost. So that's the first problem with
• 15:27
  Speaker 1: AI exposure. So when I wrote my paper, I took
• 15:31
  Speaker 1: a position similar to the Citrenia report, and I said,
• 15:33
  Speaker 1: right now, we're going towards automation, so let me focus
• 15:35
  Speaker 1: on that second. Where does that twenty percent number come from? So,
• 15:42
  Speaker 1: roughly speaking, think of it this way right now, and
• 15:46
  Speaker 1: I think in the near future, AI is pretty useless
• 15:51
  Speaker 1: in jobs that involve a huge amount of interaction with
• 15:55
  Speaker 1: the physical world construction, custodial work, manufacturing work, work that
• 16:01
  Speaker 1: involves home care, airdressing, hairdressing. The reason being that we
• 16:07
  Speaker 1: are very far behind in robotics, but also AI models
• 16:12
  Speaker 1: themselves don't have a good conceptual understanding of spatial causal
• 16:16
  Speaker 1: relations that even if we had fantastically flexible robots that
• 16:21
  Speaker 1: could cut your hair or hold your hand. AI models
• 16:25
  Speaker 1: would continuously make mistakes about spatial coosal relations and those
• 16:30
  Speaker 1: unreliabilities would end up breaking your neck. So let's eliminate
• 16:35
  Speaker 1: those jobs. I've also eliminated, again based on other people's coding,
• 16:39
  Speaker 1: any jobs that include a high degree of judgment. So
• 16:44
  Speaker 1: we wouldn't want AI to run air traffic control. So Stephanie,
• 16:51
  Speaker 1: think about it yourself. If the Manchester Airport said from
• 16:54
  Speaker 1: now on, we're not going to have any air traffic controlers.
• 16:57
  Speaker 1: Everything's gonna be done by AI. It might hallucinate, it
• 16:59
  Speaker 1: might make some mistakes, but that's fine. It's cost savings.
• 17:02
  Speaker 1: Would you fly to Manchester Airport? So we don't want that.
• 17:06
  Speaker 1: So those jobs are out, and any job that involves
• 17:09
  Speaker 1: a high degree of social interaction is out as well.
• 17:13
  Speaker 1: So that leaves essentially a range of office cognitive jobs.
• 17:17
  Speaker 1: So that's where the twenty percent comes from. Now what
• 17:20
  Speaker 1: about companies. So companies are indeed going after those jobs.
• 17:24
  Speaker 1: They're going after it jobs, they're going after back office jobs.
• 17:27
  Speaker 1: But there are several new papers that have come out
• 17:30
  Speaker 1: over the last few months and they all find the
• 17:33
  Speaker 1: same things. The companies are talking a big game about AI.
• 17:37
  Speaker 1: They say, oh, we have a lot of AI being used,
• 17:40
  Speaker 1: but it has so far zero impact on the companies,
• 17:43
  Speaker 1: zero impact on employment, zero impact on productivity because it's
• 17:48
  Speaker 1: actually like just other technologies, it's spread slowly. That's us
• 17:53
  Speaker 1: the basis of my numbers. And it's very difficult to
• 17:57
  Speaker 1: integrate AI into what those companies do with the big
• 18:00
  Speaker 1: organizational change. And I think when actually push comes to shove,
• 18:04
  Speaker 1: when they try the organizational change, but they're going to
• 18:07
  Speaker 1: realize that you cannot really replace it security people with AI.
• 18:11
  Speaker 1: You need to use it security people together with AI,
• 18:15
  Speaker 1: and that might actually give us a boost towards more human, complementary,
• 18:17
  Speaker 1: more pro worker AI. But we're not there yet because
• 18:20
  Speaker 1: they're not trying to do that in big scale yet. Now,
• 18:23
  Speaker 1: of course code that's a big advance. Will that change
• 18:27
  Speaker 1: things in twenty twenty six, I don't know. By twenty
• 18:30
  Speaker 1: twenty seven, I'm sure there will be more companies that
• 18:33
  Speaker 1: have attempted to do things, and perhaps we'll have a
• 18:36
  Speaker 1: rude awakening and this is not going to work in
• 18:38
  Speaker 1: the way that we're trying to do it. Perhaps we'll
• 18:40
  Speaker 1: find a new direction. But this is where both policy
• 18:43
  Speaker 1: and public debate are really important.
• 18:45
  Speaker 2: To your point, I think Goldman Sachs added up if
• 18:47
  Speaker 2: you just listen to all the earnings sort of calls
• 18:49
  Speaker 2: that companies are doing and chief executives are giving. Just
• 18:52
  Speaker 2: to your point about all these companies claiming I think
• 18:54
  Speaker 2: the average productivity growth that they're claiming is thirty two percent,
• 18:58
  Speaker 2: but it's not necessary evident in any of the numbers.
• 19:13
  Speaker 2: Let's get onto what policy makers could do about it,
• 19:17
  Speaker 2: because that's something that governments everywhere are obviously very focused on.
• 19:21
  Speaker 2: And I noticed that you had recently done this report
• 19:24
  Speaker 2: for Brookings. I think about a framework for thinking about
• 19:27
  Speaker 2: pro worker AI. What are the main sort of policy
• 19:31
  Speaker 2: areas that you would like people to focus on for that.
• 19:33
  Speaker 1: First of all, just two points I want to make
• 19:37
  Speaker 1: before I talk about policy. The first one is just
• 19:40
  Speaker 1: to clarify that by pro worker AI, I mean exactly
• 19:42
  Speaker 1: the same thing that I was just talking about a
• 19:44
  Speaker 1: second ago. Human complementary AIAI that helps workers do more,
• 19:49
  Speaker 1: help workers become more expert in their jobs, perform new tasks,
• 19:53
  Speaker 1: have better information for problem solving, travel shooting, judgment, and
• 19:57
  Speaker 1: so on. That's what we're talking about with pro worker
• 20:00
  Speaker 1: and not just for office workers. We have a lot
• 20:02
  Speaker 1: of examples in the paper showing how manual workers can
• 20:05
  Speaker 1: benefit from AI. It cannot replace manual workers, but electricians, plumbers,
• 20:11
  Speaker 1: nurses can hugely benefit from having the right kind of
• 20:14
  Speaker 1: AI assistant. But it has to be the right kind
• 20:17
  Speaker 1: of aassistant. It's not going to be chut GBT. So
• 20:19
  Speaker 1: that's the first point. The second point is that my
• 20:22
  Speaker 1: belief is that as important as policy actually is what
• 20:26
  Speaker 1: we're doing right now, Stephanie, is the public debate. Right now,
• 20:31
  Speaker 1: we have delegated the future of this very very important technology.
• 20:38
  Speaker 1: Some would argue, therefore the future of humanity to a
• 20:40
  Speaker 1: handful of people who have no feedback from society, who
• 20:44
  Speaker 1: have no accountability to society, and right now society is confused.
• 20:50
  Speaker 1: So on our current path, this handful of people will
• 20:54
  Speaker 1: decide what the future of AI is. And the best
• 20:57
  Speaker 1: way to counter that is to have a vision that's
• 21:02
  Speaker 1: different and hopefully better for society. And that's what I
• 21:06
  Speaker 1: hope the pro worker AI vision is. So the more
• 21:09
  Speaker 1: people talk about that, the more the public pressure will grow,
• 21:12
  Speaker 1: and the more of an alternative there will be. Look,
• 21:15
  Speaker 1: my understanding from my limited experience is that anthropic Google
• 21:21
  Speaker 1: Open AI are filled with people who are very well meaning.
• 21:26
  Speaker 1: If they thought that there is a socially beneficial and
• 21:30
  Speaker 1: still technically exciting area of AI, they would be much
• 21:34
  Speaker 1: more likely to take the plunge in that direction. It's
• 21:39
  Speaker 1: just that we're not offering them an alternative, and society
• 21:42
  Speaker 1: is not pushing back against some Outlin and his ilk's vision.
• 21:46
  Speaker 1: So that's the point. Policy, in my view, is a
• 21:49
  Speaker 1: supporting set of instruments. It can remove the stortions that
• 21:53
  Speaker 1: exist that solidify the existing system, and it can give
• 21:58
  Speaker 1: a notch to people, as policy has done in the past,
• 22:03
  Speaker 1: to try new things. So on the first bucket, there
• 22:06
  Speaker 1: are many problems in our current system that would make
• 22:13
  Speaker 1: a redirection of AI in a pro worker direction more difficult.
• 22:17
  Speaker 1: I would single out two of them, but there are
• 22:19
  Speaker 1: is more. The first one is that our tax code.
• 22:22
  Speaker 1: That's true in the UK, that's true in the US.
• 22:25
  Speaker 1: Our tax code encourages firms to replace workers because we
• 22:30
  Speaker 1: tax capital essentially at zero percent, labor twenty five to
• 22:34
  Speaker 1: thirty percent, especially in the US once you had the
• 22:37
  Speaker 1: healthcare costs and all the payroll taxes and everything. So
• 22:40
  Speaker 1: that's a massive subsidy to capital that would make firms
• 22:46
  Speaker 1: adopt automation even if automation wasn't better than humans, because
• 22:50
  Speaker 1: they're getting this subsidy. Second, we know from historical evidence
• 22:55
  Speaker 1: and current evidence that new things are done by new firms.
• 23:00
  Speaker 1: Competition is really important. The tech industry has become one
• 23:05
  Speaker 1: of the least competitive industries in history, and moreover, business
• 23:12
  Speaker 1: models that are new and different are likely to get crushed.
• 23:16
  Speaker 1: So encouraging more competition via antitrust by enabling new companies
• 23:23
  Speaker 1: to enter and try new things, I think that's a
• 23:25
  Speaker 1: very important part of way. Now there is a lot
• 23:27
  Speaker 1: of energy in Silicon Valley, but it's all these startups
• 23:30
  Speaker 1: that try to do exactly what open Ai and Anthropic
• 23:33
  Speaker 1: and Google do so that they can be both by them.
• 23:35
  Speaker 1: So that's not the kind of competition I'm talking about.
• 23:37
  Speaker 1: And then in terms of nudging us to do new things,
• 23:42
  Speaker 1: the government is horrible, in my opinion, at being an entrepreneur.
• 23:45
  Speaker 1: It cannot be a venture capitalist, it cannot be an entrepreneur,
• 23:48
  Speaker 1: it cannot be an innovator, but it has great potential
• 23:53
  Speaker 1: to be an aspiring leader. We have had so many
• 23:58
  Speaker 1: examples where a small amount of money from the government
• 24:02
  Speaker 1: has kickstarted industries in nanotechnology, in the internet, in robotics,
• 24:08
  Speaker 1: it was the Robotics Challenge, a million dollar challenge that
• 24:11
  Speaker 1: really focused people's attention to get robots that could actually
• 24:14
  Speaker 1: play a game. So we could do the same with
• 24:17
  Speaker 1: pro worker technologies. So we have given several examples of
• 24:22
  Speaker 1: technologies that are very feasible but are not getting much investment.
• 24:26
  Speaker 1: A few of them are getting some investment from smaller companies.
• 24:29
  Speaker 1: You can come up with another ten, fifteen examples and
• 24:32
  Speaker 1: the government could have an easy competition in these kinds
• 24:35
  Speaker 1: of technologies to focus the mind and show the demonstration
• 24:38
  Speaker 1: effects that would then say to people, wow, you know,
• 24:41
  Speaker 1: we could do this in other industries and other occupations
• 24:43
  Speaker 1: as well.
• 24:44
  Speaker 2: In just thinking about what you've just said, and the
• 24:48
  Speaker 2: paper you wrote for Brookings is trying to encourage us
• 24:51
  Speaker 2: to think about AI policy in a different way. So
• 24:54
  Speaker 2: I can't know what it was called, but it was
• 24:55
  Speaker 2: the AI Action Plan or something that the Trump administration
• 24:58
  Speaker 2: brought out lot and the way that we describe it generally,
• 25:02
  Speaker 2: but particularly when we're talking about China in the US,
• 25:05
  Speaker 2: AI policy is all about how to get there as
• 25:08
  Speaker 2: fast as possible, how to make sure especially in the US,
• 25:11
  Speaker 2: how to make sure we win the race. And there's
• 25:13
  Speaker 2: quite a lot of focus on sort of privacy and
• 25:17
  Speaker 2: concerns around that, and maybe concerns about the pace of adoption,
• 25:21
  Speaker 2: and that's obviously the gap that you're trying to fill,
• 25:22
  Speaker 2: but it doesn't feel like there's much about how to
• 25:25
  Speaker 2: make this work for people. And I'm sort of struck
• 25:27
  Speaker 2: because we had the Chancellor of Rachel Reeves, the UK
• 25:30
  Speaker 2: Finance Minister, on the show a week or two ago,
• 25:33
  Speaker 2: and you know, I think that's one of the things
• 25:35
  Speaker 2: that she's thinking about is we're not going to lead
• 25:38
  Speaker 2: the AI race in the UK, but we have said
• 25:41
  Speaker 2: a lot about leading on the adoption and I guess
• 25:44
  Speaker 2: the piece of that that you would add is you've
• 25:46
  Speaker 2: got to adopt it in a pro worker way. I mean,
• 25:48
  Speaker 2: what does that what would that look like for the UK.
• 25:50
  Speaker 1: Let me first say that the paper that you're referring
• 25:53
  Speaker 1: to is actually co authored with David Arter and Simon Johnson,
• 25:55
  Speaker 1: so let me give a shout out to them as well. Secondly,
• 25:59
  Speaker 1: I think you're absolutely right. While you could give some
• 26:03
  Speaker 1: credit to the Trump administration for emphasizing AI, they are
• 26:08
  Speaker 1: really radardless their only shtick here is this has to
• 26:12
  Speaker 1: be an American technology, and we have to race and
• 26:15
  Speaker 1: we have to get rid of all regulations. That's not
• 26:20
  Speaker 1: a coherent AI policy. But I also fear it's even
• 26:24
  Speaker 1: worse than that, and it's worse in the following way.
• 26:30
  Speaker 1: This AGI winner take all framing is having truly pernicious
• 26:37
  Speaker 1: effects on US China relations because once you are in
• 26:42
  Speaker 1: this mindset that you are locked into this existential race
• 26:47
  Speaker 1: for AI supremacy with China, it means that there's no
• 26:51
  Speaker 1: room for collaboration with China on anything because they are
• 26:53
  Speaker 1: your mortal enemy, because if they get to AI supremacy
• 26:56
  Speaker 1: before you, they are going to destroy you. That's completely false.
• 27:00
  Speaker 1: Models are not going to be at a level that
• 27:02
  Speaker 1: they can just give you global supremacy by themselves, And
• 27:05
  Speaker 1: there are many other things that we can do with AI.
• 27:08
  Speaker 1: In fact, now coming to the UK, China, Germany are
• 27:14
  Speaker 1: doing more interesting things with AI than the US in
• 27:17
  Speaker 1: some domain. Sure, the US has the unrivaled leadership in
• 27:23
  Speaker 1: large language models and foundation models, but I think the
• 27:28
  Speaker 1: real gains from AI, as I hinted at the beginning
• 27:31
  Speaker 1: of our conversation, will come from using AI in applications
• 27:35
  Speaker 1: in manufacturing. Healthcare I think is huge, but manufacturing is
• 27:40
  Speaker 1: going to be easier. And who is leading the efforts
• 27:45
  Speaker 1: to put AI into manufacturing is China. It's Germany even
• 27:49
  Speaker 1: though they have no LLM industry, because they have the
• 27:53
  Speaker 1: manufacturing know how, they have the data, and they are
• 27:55
  Speaker 1: not beholden to this AGI race, so they're trying to
• 27:59
  Speaker 1: do more practical things. I think that's the space in
• 28:02
  Speaker 1: which the UK has to be now. Unfortunately UK doesn't
• 28:05
  Speaker 1: have much manufacturing left, but I think for the remaining
• 28:10
  Speaker 1: manufacturing and other applications, I think that's where UK can
• 28:14
  Speaker 1: have a leadership role because Germany is so far behind.
• 28:17
  Speaker 1: Germany shouldn't have a leadership role. China, of course is
• 28:20
  Speaker 1: going to have a leadership role, but UK can have
• 28:21
  Speaker 1: a leadership role once they have a broader scoping of
• 28:27
  Speaker 1: what it is that we can do with AI, and
• 28:30
  Speaker 1: if we actually manage that, it will have beneficial effects
• 28:34
  Speaker 1: for global balances, because once you get out of this
• 28:38
  Speaker 1: trap of winner take all, we cannot collaborate on anything
• 28:41
  Speaker 1: with China. We have so many global problems, global peace,
• 28:46
  Speaker 1: all the societies that are aging, that require adjustment, climate change, pandemics,
• 28:50
  Speaker 1: There is so much that we actually need to collaborate
• 28:52
  Speaker 1: with China, and if in fact China makes breakthroughs in
• 28:56
  Speaker 1: applying AI to manufacturing, US should learn from them. So
• 29:02
  Speaker 1: there should be information sharing in AI as well.
• 29:05
  Speaker 2: I'm going to run out of time, but I had
• 29:06
  Speaker 2: a couple of more and one is following on from
• 29:08
  Speaker 2: what we were saying about how a country could position
• 29:12
  Speaker 2: itself that's not trying to be in this kind of
• 29:13
  Speaker 2: existential race that the US has positioned itself. In the
• 29:18
  Speaker 2: other conversation, you hear a lot in the UK is
• 29:20
  Speaker 2: and I mentioned it to the Chancellor the other day,
• 29:23
  Speaker 2: is you know that professional services that are successful in
• 29:28
  Speaker 2: the UK, we'd still have some advanced manufacturing, but our
• 29:31
  Speaker 2: strongest categories tend to be along with creative industry, professional services,
• 29:35
  Speaker 2: legal services, accounting, finance, all those things. They seem to
• 29:39
  Speaker 2: be particularly in the frame when it comes to AI,
• 29:42
  Speaker 2: at least in the discussion, and there has been I
• 29:44
  Speaker 2: know a government concern. That means, if we want them
• 29:47
  Speaker 2: still to be leading sectors, they have to be leading
• 29:50
  Speaker 2: in adoption and we have to make sure there are
• 29:51
  Speaker 2: no regulatory or data privacy obstacles in the way of that.
• 29:57
  Speaker 2: I mean, I guess that raises the question in the
• 29:59
  Speaker 2: race to A do you could actually be making the
• 30:01
  Speaker 2: institutional set up worse, not just failing to make it better.
• 30:05
  Speaker 1: Right, you've given me an opening to talk about another
• 30:08
  Speaker 1: one of my topics, which is data. So yes, Indeed,
• 30:13
  Speaker 1: if your objective was pour as much money into AI
• 30:17
  Speaker 1: as possible and get rid of all short term obstacles
• 30:21
  Speaker 1: to AI, you would get rid of privacy and you
• 30:24
  Speaker 1: would allow AI companies to capture as much data as
• 30:29
  Speaker 1: they want freely. That would be and that has been
• 30:33
  Speaker 1: the worst idea you can imagine. It's actually worse for
• 30:37
  Speaker 1: the industry. The future of the industry depends on data.
• 30:42
  Speaker 1: Data is going to be more important for our future
• 30:47
  Speaker 1: as an economy, as a society than land. Can you
• 30:50
  Speaker 1: imagine that if we said any piece of land you want,
• 30:53
  Speaker 1: you can take it. That would be just chaos. But
• 30:56
  Speaker 1: that's how we treat data, and that is actually bad
• 30:59
  Speaker 1: for the industry because it creates a tragedy of the
• 31:02
  Speaker 1: commons where everybody's exploiting data and nobody is investing in data.
• 31:07
  Speaker 1: Especially if you want to do useful things with AI,
• 31:09
  Speaker 1: like the pro worker AI that I was mentioning, you
• 31:11
  Speaker 1: need a lot of high quality use cases. We can
• 31:15
  Speaker 1: do pro worker AI to help teachers, to have nurses,
• 31:18
  Speaker 1: to help electricians. How are you going to do that? Well,
• 31:21
  Speaker 1: you need to train these models on basic knowledge, but
• 31:24
  Speaker 1: you also need to train them on use cases by
• 31:27
  Speaker 1: the most experienced workers in that field, working with edge cases,
• 31:32
  Speaker 1: difficult cases, and they're not going to produce that data
• 31:34
  Speaker 1: unless you pay them. So the current environment where we
• 31:38
  Speaker 1: say privacy doesn't matter data, you should give it as
• 31:41
  Speaker 1: much data to these companies because they're data hungry. That's
• 31:44
  Speaker 1: actually destroying the future of the industry because these models
• 31:47
  Speaker 1: are going to run out of high quality data, they're
• 31:50
  Speaker 1: going to be produced, they're going to be trained on
• 31:53
  Speaker 1: low quality data, and they're going to be more likely
• 31:55
  Speaker 1: to create AI slope rather than the kind of high quality,
• 31:58
  Speaker 1: reliable AI that we need across a range of occupations.
• 32:02
  Speaker 2: I guess just coming back to sort of where we
• 32:05
  Speaker 2: started in the sense of the perspective of your twenty
• 32:08
  Speaker 2: twenty three book, and one of the features of that
• 32:10
  Speaker 2: you and Simon's book was the comparison with the Industrial
• 32:14
  Speaker 2: Revolution and making the point that although we tend to say, oh,
• 32:18
  Speaker 2: it was fine, we ended up with the productivity, and
• 32:20
  Speaker 2: it made everyone better off. You were just pointing to
• 32:23
  Speaker 2: how long the transition lasted, how incredibly costly that was
• 32:27
  Speaker 2: for people, and how much it required active effort to
• 32:30
  Speaker 2: manage it and have a better outcome for people. One
• 32:34
  Speaker 2: of the big differences, it seems, between that Industrial revolution
• 32:38
  Speaker 2: and what we may see now in the next few
• 32:40
  Speaker 2: years with AI is that the workers in the frame
• 32:43
  Speaker 2: fundamentally are white collar workers. And in fact, Dario Medea's
• 32:47
  Speaker 2: talked about half of entry level white collar work. It's
• 32:51
  Speaker 2: quite a sort of safe number because he talks about
• 32:53
  Speaker 2: this and then you could change you could change all
• 32:55
  Speaker 2: the definitions, but half of the entry level white collar
• 32:58
  Speaker 2: jobs will be gone in five years. Does it fundamentally
• 33:02
  Speaker 2: change the challenge for policymakers and even the sort of
• 33:06
  Speaker 2: short term macroeconomic impact If the main workers affected are
• 33:11
  Speaker 2: also white collar workers, they're possibly some of the better paid,
• 33:16
  Speaker 2: greatest consuming members of society.
• 33:19
  Speaker 1: Well, first of all, yes, indeed, there's a lot of
• 33:22
  Speaker 1: uncertainty about what that impact is going to be, but
• 33:25
  Speaker 1: it's true that it's going to be on white color
• 33:26
  Speaker 1: workers more than manufacturing workers. For the reasons that we
• 33:30
  Speaker 1: talked about that these models cannot do physical work or
• 33:33
  Speaker 1: cannot be combined with physical work. Yet now white color
• 33:38
  Speaker 1: workers are college educated, our leaders are college educated, so
• 33:43
  Speaker 1: their plight might have a bigger impact on the political
• 33:47
  Speaker 1: system than the plight of say, high school graduate or
• 33:51
  Speaker 1: high school dropout workers did in the United States or
• 33:54
  Speaker 1: the UK in the nineteen eighties, for example, So that's
• 33:56
  Speaker 1: a possibility. The second important issue is that the Industrial Revolution. Indeed,
• 34:02
  Speaker 1: and this is very important because you hear this sort
• 34:06
  Speaker 1: of grossy view over the Industrial Revolution from Silicon Valley
• 34:10
  Speaker 1: all the time that everything worked that well, it took
• 34:12
  Speaker 1: about one hundred years of pain and suffering before things
• 34:15
  Speaker 1: started getting better. Well, we don't have that kind of time.
• 34:19
  Speaker 1: Our democracy wouldn't survive, and AI is advancing far too rapidly,
• 34:23
  Speaker 1: so our political system needs to be much better and
• 34:26
  Speaker 1: much faster at redirecting things and adjusting to things. So
• 34:30
  Speaker 1: I think those are very important points for us to remember.
• 34:38
  Speaker 1: But finally, I think it's also very important to recognize
• 34:44
  Speaker 1: that the impact will not stop with white color workers,
• 34:49
  Speaker 1: because if college graduates cannot get the jobs that they
• 34:54
  Speaker 1: want to get, they're going to go and compete for
• 34:56
  Speaker 1: other jobs. They're not going to stay at home, they'll
• 35:01
  Speaker 1: create wage downward, wage pressure and job displacement risk for
• 35:05
  Speaker 1: other people, or they will all be pulled into sort
• 35:10
  Speaker 1: of gig work, which then creates all sorts of other
• 35:13
  Speaker 1: problems for the economy and for the labor market. So
• 35:16
  Speaker 1: it's a systemic problem for the labor market as well.
• 35:19
  Speaker 2: Okay, I'm not sure that that was the most uplifting
• 35:22
  Speaker 2: place to end, but it's been embracing but profoundly illuminating
• 35:26
  Speaker 2: to me and clarifying conversation. Darren A Samoglu, thank you
• 35:29
  Speaker 2: so much.
• 35:30
  Speaker 1: Thank you, Stephanie, it was great to talk to you.
• 35:50
  Speaker 2: Thanks for listening to Trumponomics from Bloomberg. It was hosted
• 35:53
  Speaker 2: by me Stephanie Flanders. Trumponomics was produced by Samasadi and Versus,
• 35:57
  Speaker 2: and sound design was by Blake Maple's and Aaron Casper.
• 36:01
  Speaker 2: To help others find the show, please rate and review
• 36:04
  Speaker 2: it highly wherever you listen.