Latest / Faster, Please! — The Podcast / ✨ My interview with economist Tyler Cowen on the Age of AI (and more)
Transcript
- 0:00SPEAKER_01: Welcome to Faster Please,
- 0:02SPEAKER_01: the podcast, the audio-visual companion to my Faster Please newsletter.
- 0:06SPEAKER_01: I'm your host, Jim Pethokoukis.
- 0:08SPEAKER_01: My guest for this episode is Tyler Cowen.
- 0:10SPEAKER_01: Tyler is an economics professor at George Mason University and co-founder of
- 0:14SPEAKER_01: the economics blog, Marginal Revolution.
- 0:17SPEAKER_01: He's a writer for the Free Press and the author of several books,
- 0:20SPEAKER_01: including The Great Stagnation, Average is Over, and Stubborn Attachments.
- 0:25SPEAKER_01: Today, we discuss why AI's productivity gains have been slow to appear,
- 0:29SPEAKER_01: whether public pushback could stall progress,
- 0:32SPEAKER_01: and what the technology means for jobs and long-run American economic growth.
- 0:37SPEAKER_01: Hope you enjoy it.
- 0:38SPEAKER_01: Tyler, welcome back to the podcast.
- 0:40SPEAKER_01: Happy to be here.
- 0:40SPEAKER_01: Good morning.
- 0:41SPEAKER_01: Good morning.
- 0:43SPEAKER_01: Do you think the current public backlash
- 0:46SPEAKER_01: against data centers will materially slow AI progress in the United States?
- 0:53SPEAKER_01: My baseline here is that as long as investors and businesses think
- 0:59SPEAKER_01: that all this build-out will lead to economic value,
- 1:02SPEAKER_01: that these data centers will get built somewhere,
- 1:06SPEAKER_01: though maybe they'll cost more, it'll take longer.
- 1:09SPEAKER_01: What do you think?
- 1:11SPEAKER_00: I agree with you.
- 1:12SPEAKER_00: We have 50 states,
- 1:13SPEAKER_00: as you well know,
- 1:14SPEAKER_00: and all you need is a few governors who decide they want money and jobs,
- 1:19SPEAKER_00: and America is great at money and jobs.
- 1:22SPEAKER_00: And a lot of them are saying things they don't really mean to keep their voters at bay.
- 1:27SPEAKER_00: I suspect Texas and Pennsylvania are both examples of that.
- 1:31SPEAKER_00: So don't look at what they say, look at what they do.
- 1:34SPEAKER_00: I expect something less than full steam ahead that basically will get the job done.
- 1:41SPEAKER_01: I've been sort of comparing the current backlash against data to nuclear power.
- 1:47SPEAKER_01: And when I think about nuclear power, it was a long sort of brewing backlash against nuclear.
- 1:54SPEAKER_01: There was always a concern about radiation.
- 1:57SPEAKER_01: And then over a decade,
- 1:58SPEAKER_01: there were some minor scares,
- 2:00SPEAKER_01: there were whistleblowers,
- 2:01SPEAKER_01: there were anti-nuclear books,
- 2:03SPEAKER_01: there was the rise of a broader environmental movement.
- 2:06SPEAKER_01: A lot happened before we got to like Three Mile Island,
- 2:09SPEAKER_01: when that seemed to be sort of like the final nail,
- 2:12SPEAKER_01: like in the coffin as far as public support for nuclear.
- 2:15SPEAKER_01: This, whether it's the data,
- 2:18SPEAKER_01: particularly I guess the data center backlash,
- 2:20SPEAKER_01: and maybe AI more broadly,
- 2:22SPEAKER_01: seemed to happen Quite quickly, at least based on that historical example.
- 2:28SPEAKER_01: So how did these rather anonymous light industrial buildings lose their social
- 2:35SPEAKER_01: permission so quickly and so completely?
- 2:39SPEAKER_00: Everything happens more quickly when discourse occurs over the internet.
- 2:43SPEAKER_00: Some people suggest China has manipulated the discourse.
- 2:47SPEAKER_00: I don't know.
- 2:48SPEAKER_00: Hard to judge that.
- 2:50SPEAKER_00: I can't imagine that's the dominant effect going on.
- 2:54SPEAKER_00: But the news cycle is so sped up.
- 2:55SPEAKER_00: You know,
- 2:56SPEAKER_00: Trump does one outrageous thing after the other and a day goes by
- 3:00SPEAKER_00: and you stop talking about it and you're on to the next one and then you're on to
- 3:03SPEAKER_00: the next one.
- 3:04SPEAKER_00: So this may just be part of the super sped up news cycle and it might prove fairly evanescent.
- 3:10SPEAKER_00: That would not surprise me.
- 3:12SPEAKER_01: Yeah, I mean,
- 3:12SPEAKER_01: I think on that point,
- 3:15SPEAKER_01: to the extent that some of the complaints about data centers just seem
- 3:19SPEAKER_01: to be just factually erroneous,
- 3:22SPEAKER_01: And I'd like to think that eventually sort of the facts will like break through like
- 3:27SPEAKER_01: that filter.
- 3:29SPEAKER_01: But also to the extent that these concerns are built on like predictions
- 3:33SPEAKER_01: that people have heard that all the white collar jobs will be gone in 18 months.
- 3:39SPEAKER_01: We won't know what the world will look like in 20 months.
- 3:43SPEAKER_01: If we're sitting here 18 months from now or a year from now and People who sit
- 3:49SPEAKER_01: behind computer screens still have jobs,
- 3:51SPEAKER_01: and they look out the window and the world doesn't look radically changed.
- 3:56SPEAKER_01: I wonder if that doesn't not just undermine the data center backlash,
- 3:59SPEAKER_01: but really these terrible polling numbers for AI more broadly.
- 4:05SPEAKER_00: I'm not sure how much the facts will matter.
- 4:07SPEAKER_00: I just think people might exhaust themselves.
- 4:10SPEAKER_00: There's not really a villain with a face.
- 4:13SPEAKER_00: So whether or not people love the CEOs of OpenAI and Anthropic,
- 4:18SPEAKER_00: The faces of those people are not connected to data centers,
- 4:22SPEAKER_00: the way, say, Sam Bankman Freed was connected to crypto.
- 4:25SPEAKER_00: So unless there's a villain, I don't think it's going to stick.
- 4:30SPEAKER_00: And there'll be plenty of other things to worry and complain about.
- 4:33SPEAKER_00: The world throws those up pretty regularly.
- 4:35SPEAKER_00: And I would just predict we'll move on is my best guess.
- 4:39SPEAKER_01: For me, it's easy to listen to people complaining.
- 4:44SPEAKER_01: About, I guess,
- 4:45SPEAKER_01: data centers,
- 4:46SPEAKER_01: but about AI more broadly and think,
- 4:49SPEAKER_01: you know,
- 4:51SPEAKER_01: they've listened,
- 4:52SPEAKER_01: they've watched too many TikTok videos,
- 4:54SPEAKER_01: they've looked at just too many headlines,
- 4:56SPEAKER_01: we might have a natural proclivity to believe in these kind of scenarios,
- 5:00SPEAKER_01: especially I haven't seen 50 years of movies about AI going wrong.
- 5:04SPEAKER_01: But then again, I mean,
- 5:05SPEAKER_01: can you really believe people,
- 5:07SPEAKER_01: or rather,
- 5:07SPEAKER_01: can you blame people when the CEOs of these companies,
- 5:12SPEAKER_01: the ones in charge of the technology,
- 5:15SPEAKER_01: Are giving them these very apocalyptic scenarios about jobs.
- 5:21SPEAKER_01: And even though they'll give the upsides too,
- 5:23SPEAKER_01: they're painting a world very different, coming very soon.
- 5:30SPEAKER_01: And I guess I fundamentally can't blame people.
- 5:34SPEAKER_01: Most people I think are fairly risk averse.
- 5:36SPEAKER_01: Most people don't become entrepreneurs.
- 5:40SPEAKER_01: They may want the latest iPhone,
- 5:41SPEAKER_01: but that's very different than wanting AI to change the world they know it.
- 5:48SPEAKER_01: Do you blame the CEOs a bit for this backlash?
- 5:52SPEAKER_00: I don't know if the word blame is the right word here.
- 5:55SPEAKER_00: I think a lot of the PR has not shown enough equanimity and that the people
- 6:01SPEAKER_00: who are motivated to work on this are precisely the same people
- 6:05SPEAKER_00: who think it will have very extreme effects.
- 6:08SPEAKER_00: And those companies have been remarkably productive,
- 6:11SPEAKER_00: gotten a lot done at a truly incredible pace,
- 6:13SPEAKER_00: even by American or tech world standards.
- 6:16SPEAKER_00: So the notion that they have semi-religious views of what this all means is
- 6:21SPEAKER_00: a very San Francisco kind of thing.
- 6:24SPEAKER_00: Like, do I blame people in the 1960s who thought that LSD was creating a new world,
- 6:30SPEAKER_00: new way of living, new way of life?
- 6:31SPEAKER_00: Well, sort of, but that's what we are.
- 6:36SPEAKER_00: You know,
- 6:36SPEAKER_00: we're a little nutty in the head as a country,
- 6:39SPEAKER_00: and this is California, and then it's San Francisco.
- 6:42SPEAKER_00: So we also need to embrace that nuttiness at the same time.
- 6:46SPEAKER_00: So I think that's the bigger picture I try to keep in mind when maintaining my own equanimity.
- 6:53SPEAKER_01: So take seriously that they are building an important technology,
- 7:01SPEAKER_01: somewhat less seriously their 12-month job forecasts.
- 7:05SPEAKER_00: Correct.
- 7:06SPEAKER_00: Or, you know, some people have said in two or three years we'll have cured all these diseases.
- 7:10SPEAKER_00: I'm quite sure that's not true.
- 7:12SPEAKER_00: And I've spoken with bio experts who work with AI.
- 7:15SPEAKER_00: You know, they're not AI skeptics at all.
- 7:17SPEAKER_00: And they just say no way.
- 7:20SPEAKER_00: Just think of the regulatory obstacles, clinical trials, experiment, verification.
- 7:25SPEAKER_00: There's so many parts of the loop.
- 7:28SPEAKER_00: With humans still in them,
- 7:29SPEAKER_00: even assuming the AI met all of those fantastic expectations,
- 7:34SPEAKER_00: it's just simply quite impossible.
- 7:38SPEAKER_01: I mean, I remember talking to some Silicon Valley people,
- 7:42SPEAKER_01: I don't know, 2017, 2018,
- 7:45SPEAKER_01: when already there was talk about AI taking jobs and we'd all be driving around
- 7:51SPEAKER_01: in robo-taxis again within one year.
- 7:55SPEAKER_01: And it seemed at the time they were considering a massive job loss and talking
- 7:59SPEAKER_01: about universal basic income.
- 8:03SPEAKER_01: But I felt they were overly confident in their ability to understand a field,
- 8:09SPEAKER_01: economics and To some extent,
- 8:12SPEAKER_01: economics is also the study of sort of constraints,
- 8:14SPEAKER_01: like the constraints that some of which you just mentioned on that kind of change,
- 8:20SPEAKER_01: sort of the political economy of the real world versus what you might be able to do
- 8:25SPEAKER_01: in a lab.
- 8:28SPEAKER_00: I agree,
- 8:28SPEAKER_00: but keep in mind,
- 8:29SPEAKER_00: they've been more right than wrong,
- 8:31SPEAKER_00: and a lot of pundits,
- 8:33SPEAKER_00: economists, public intellectuals, most missed the AI revolution.
- 8:38SPEAKER_00: Those people were more or less completely right.
- 8:41SPEAKER_00: So it's very hard to tell them all of a sudden,
- 8:44SPEAKER_00: oh, well, you're wrong on this, that, and the other.
- 8:47SPEAKER_00: So, you know, their track record on the tech, at least, is amazingly good.
- 8:54SPEAKER_00: Now, I'm one of those people, one of the few economists who predicted AI would be a big thing.
- 8:58SPEAKER_00: This was back in 2013.
- 9:00SPEAKER_00: But I also said it wouldn't totally overturn the world as we know it.
- 9:04SPEAKER_00: I think I've been right on both counts.
- 9:06SPEAKER_00: I don't think that many people have been right on both counts.
- 9:09SPEAKER_00: So that's my boast for the morning.
- 9:14SPEAKER_01: So your confidence level, I did an interview with Eric Brynjolfsson not so long ago.
- 9:20SPEAKER_01: And I don't think I'm reading into it.
- 9:22SPEAKER_01: I think he was fairly plain that he's been a little disappointed with sort of
- 9:26SPEAKER_01: the return on investment.
- 9:28SPEAKER_01: We've seen,
- 9:28SPEAKER_01: even though he,
- 9:29SPEAKER_01: you know,
- 9:30SPEAKER_01: it's still early and he's written some of the,
- 9:32SPEAKER_01: you know,
- 9:32SPEAKER_01: the most important research on how long it takes to incorporate these technologies.
- 9:36SPEAKER_01: But he seems disappointed about the return on these technologies so far
- 9:41SPEAKER_01: while being optimistic that,
- 9:43SPEAKER_01: you know,
- 9:43SPEAKER_01: that there was going to be a big return and big productivity gains and all that.
- 9:47SPEAKER_01: Are you just a little disappointed?
- 9:52SPEAKER_00: Not at all.
- 9:53SPEAKER_00: You know,
- 9:53SPEAKER_00: Eric and I had a formal debate on exactly this at an event called Techonomy way back
- 9:58SPEAKER_00: when.
- 9:59SPEAKER_00: Maybe it's like 2015.
- 10:00SPEAKER_00: I'm not sure.
- 10:01SPEAKER_00: And I just said, Eric, these things take a long time.
- 10:04SPEAKER_00: It's going to take decades.
- 10:05SPEAKER_00: It's underway.
- 10:06SPEAKER_00: That's great.
- 10:07SPEAKER_00: But don't expect too much too quickly.
- 10:11SPEAKER_00: I've actually been pleasantly surprised it's gone as well as it has.
- 10:15SPEAKER_00: I agree, so far the productivity dividend, it's close to zero.
- 10:19SPEAKER_00: Not in programming,
- 10:20SPEAKER_00: but in the economy as a whole,
- 10:22SPEAKER_00: it's not clear you can even see it in the aggregate numbers.
- 10:26SPEAKER_01: No, I think you're right.
- 10:29SPEAKER_01: And so we're at a place where we're not seeing,
- 10:32SPEAKER_01: set aside cancer cures, we're not seeing it in the aggregate productivity numbers.
- 10:37SPEAKER_01: Companies,
- 10:38SPEAKER_01: when they talk about the technology,
- 10:40SPEAKER_01: still are talking about return in sort of future tense,
- 10:44SPEAKER_01: not really present tense for the most part,
- 10:46SPEAKER_01: particularly if you get beyond very narrow tech sectors.
- 10:50SPEAKER_00: There is a lot of revenue coming into the companies, right?
- 10:53SPEAKER_00: I mean, that's great.
- 10:55SPEAKER_00: It will help them keep on going, but that is itself different from productivity per se.
- 11:00SPEAKER_00: But finally, the revenue numbers are really extremely positive.
- 11:05SPEAKER_01: I think that's right.
- 11:06SPEAKER_01: So if we're not seeing the upside,
- 11:09SPEAKER_01: like how vulnerable sort of,
- 11:12SPEAKER_01: not just public opinion,
- 11:13SPEAKER_01: but sort of Washington to stories and incidents like
- 11:18SPEAKER_01: the OpenAI hugging face incident.
- 11:21SPEAKER_01: I was just reading some tweets from a colleague who,
- 11:26SPEAKER_01: when he's been talking to people in Washington,
- 11:28SPEAKER_01: they seem right now With the exception of Bernie Sanders,
- 11:31SPEAKER_01: who seems all over it, unaware of this open AI hugging face hack.
- 11:37SPEAKER_01: And my concern is that this will be kind of a catalyst for the people,
- 11:43SPEAKER_01: for people who already want to pause this technology or really slow it down.
- 11:47SPEAKER_01: They're already there.
- 11:49SPEAKER_01: But I'm wondering if this incident,
- 11:51SPEAKER_01: once Washington gets back,
- 11:53SPEAKER_01: Doesn't become the kind of thing which then leads to more calls
- 11:57SPEAKER_01: for stricter regulation or some kind of pause.
- 12:02SPEAKER_00: There will be all those calls.
- 12:04SPEAKER_00: The discourse will be ugly and stupid,
- 12:06SPEAKER_00: no matter what your point of view, even if you favor a lot more regulation.
- 12:10SPEAKER_00: But it's very hard to stop fundamental new technologies.
- 12:14SPEAKER_00: There's also Chinese open source.
- 12:16SPEAKER_00: There's a national security imperative here.
- 12:19SPEAKER_00: I just think it's going to happen.
- 12:21SPEAKER_00: It will pain me every day to see the debates and some of the laws
- 12:24SPEAKER_00: that probably end up getting passed, but I just don't see how it gets stopped.
- 12:29SPEAKER_00: You know,
- 12:29SPEAKER_00: for better or worse,
- 12:30SPEAKER_00: I would say for better,
- 12:32SPEAKER_00: but there's not a coherent path for telling a story of how this ends.
- 12:38SPEAKER_01: You've been criticized in social media and I think there's actually a new piece in
- 12:43SPEAKER_01: the free press countering one of your recent columns saying
- 12:47SPEAKER_01: that you're just underestimating the potential damage that can be caused by sort
- 12:51SPEAKER_01: of out of control AI agents doing their own thing or AI agents employed
- 12:58SPEAKER_01: by bad people who want to do bad things, some cyber attacks.
- 13:05SPEAKER_01: Have you rethought that at all?
- 13:06SPEAKER_01: Do you think you're maybe underestimating the potential damage and we should take
- 13:09SPEAKER_01: that more seriously when thinking about regulation?
- 13:13SPEAKER_00: I always say to these people, show me the numbers.
- 13:15SPEAKER_00: I'm very open-minded.
- 13:17SPEAKER_00: They love to talk in extreme parables of Skynet goes live or a bioterror attack
- 13:23SPEAKER_00: where all humans die.
- 13:25SPEAKER_00: But you know,
- 13:25SPEAKER_00: in the short run,
- 13:26SPEAKER_00: the intermediate term, like what share prices should we expect to go down?
- 13:32SPEAKER_00: What variables should we expect to change?
- 13:35SPEAKER_00: Give me a number for what cybersecurity costs will be next year and
- 13:38SPEAKER_00: by what percentage that will go up.
- 13:41SPEAKER_00: I was just reading a piece by some people in the cybersecurity insurance sector,
- 13:46SPEAKER_00: and their forecast was AI in the next year will boost cybersecurity costs by 20%.
- 13:52SPEAKER_00: Now, I don't know that's the right number, but they're experts and they're saying 20%.
- 13:57SPEAKER_00: I'm saying to the doomers, Give me a better number.
- 14:01SPEAKER_00: They don't have any numbers at all.
- 14:03SPEAKER_00: Now, if the equilibrium is cybersecurity costs go up 20%, that's actually a lot of money.
- 14:08SPEAKER_00: It's worth investing a lot to limit that happening.
- 14:11SPEAKER_00: But it's not remotely close to the scenarios these people are talking about.
- 14:16SPEAKER_00: So I'm just insisting,
- 14:18SPEAKER_00: give me the numbers,
- 14:19SPEAKER_00: make falsifiable predictions,
- 14:21SPEAKER_00: be concrete,
- 14:22SPEAKER_00: don't talk in metaphor,
- 14:24SPEAKER_00: and you know, oh,
- 14:25SPEAKER_00: these things are civilizations and beings are sentient,
- 14:28SPEAKER_00: like whatever you want to think.
- 14:30SPEAKER_00: But please give me concrete numbers and predictions.
- 14:32SPEAKER_00: And they just don't do it.
- 14:34SPEAKER_00: So that alone doesn't mean they're wrong.
- 14:37SPEAKER_00: But if they're really going to take this seriously,
- 14:39SPEAKER_00: they need to get their act together and act like actual scientists,
- 14:43SPEAKER_00: the way the climate change people mostly have done.
- 14:46SPEAKER_00: So far, we're not seeing that.
- 14:48SPEAKER_00: There was another estimate I saw.
- 14:50SPEAKER_00: It said...
- 14:51SPEAKER_00: Cybersecurity insurance costs will double over a five-year time period.
- 14:56SPEAKER_00: This was from experts in the sector.
- 14:58SPEAKER_00: Again, I'm not endorsing that number, but it's a number, and it puts it in perspective.
- 15:03SPEAKER_00: It's terrible if those costs double,
- 15:06SPEAKER_00: but again, it's not an insane scenario that we can't live with.
- 15:12SPEAKER_01: I understand your point sort of about like the momentum of an important technology.
- 15:19SPEAKER_01: And assuming this is an important general purpose technology,
- 15:25SPEAKER_01: thinking that we just won't go forward.
- 15:28SPEAKER_01: Also given that we have, there's a national, one, there seems to be a lot of economic value.
- 15:33SPEAKER_01: There's sort of a national security imperative.
- 15:37SPEAKER_01: Also, these models are kind of out there.
- 15:39SPEAKER_01: Sort of going back seems...
- 15:42SPEAKER_01: Ridiculous.
- 15:43SPEAKER_01: That said,
- 15:44SPEAKER_01: I do fear that given the current sort of public pessimism,
- 15:52SPEAKER_01: you know,
- 15:52SPEAKER_01: polls saying they're far more excited, they're far more worried than excited by it.
- 15:57SPEAKER_01: I'm worried that even an incident,
- 15:59SPEAKER_01: let's say bigger than what we saw with OpenAI and HuggingFace where
- 16:04SPEAKER_01: A model ends up somehow breaking into the operations network of some power company.
- 16:11SPEAKER_01: No actual damage, you know, somewhat like Three Mile Island.
- 16:14SPEAKER_01: Nobody gets killed,
- 16:16SPEAKER_01: but that kind of incident,
- 16:19SPEAKER_01: bigger than what we've already seen,
- 16:20SPEAKER_01: that that's enough to sort of really push people into becoming so negative
- 16:25SPEAKER_01: that Washington does act with very harmful legislation.
- 16:30SPEAKER_01: And, you know, we didn't shut all the nuclear reactors.
- 16:35SPEAKER_01: They did so in Japan and Germany after Fukushima.
- 16:37SPEAKER_01: We didn't shut down our nuclear reactors after Three Mile Island,
- 16:40SPEAKER_01: but we kind of didn't build any more.
- 16:42SPEAKER_01: So why isn't that a good analogy?
- 16:48SPEAKER_00: We believe Iran did this already in the Midwest.
- 16:51SPEAKER_00: Maybe it's not confirmed it was Iran, but we know what happened.
- 16:55SPEAKER_00: And we haven't stopped.
- 16:56SPEAKER_00: Look, we need these models for cyber defense is an important, important point.
- 17:01SPEAKER_00: So if we stop our models, we're a sitting duck for the rest of the world.
- 17:05SPEAKER_00: First up is China.
- 17:06SPEAKER_00: But over time,
- 17:08SPEAKER_00: we've seen China, Russia,
- 17:09SPEAKER_00: others, they're very happy to spread technology to Iran, North Korea, other parties.
- 17:14SPEAKER_00: And the notion that America is just going to sit here
- 17:17SPEAKER_00: and cannot use its own top models in essence to defend itself,
- 17:22SPEAKER_00: I just don't think Even an irresponsible American government is going to let
- 17:26SPEAKER_00: that happen.
- 17:27SPEAKER_00: So I could be proven wrong on that, but that's my best guess.
- 17:32SPEAKER_01: The Economist magazine, they did a piece recently about AGI-pilled economists.
- 17:39SPEAKER_01: They listed about 12 folks.
- 17:44SPEAKER_01: I think,
- 17:45SPEAKER_01: to their shame,
- 17:46SPEAKER_01: did not include you on the list of key economists who think and talk about AI.
- 17:52SPEAKER_01: I think you are on the more optimistic side.
- 17:55SPEAKER_01: I don't know if I would say you're AGI-pilled or superintelligence-pilled,
- 17:58SPEAKER_01: but you're certainly on the more optimistic side.
- 18:01SPEAKER_01: What do you tend to see that so many other economists don't,
- 18:05SPEAKER_01: whose predictions are like,
- 18:07SPEAKER_01: I see this technology adding like a quarter percent to GDP growth upon deployment
- 18:14SPEAKER_01: over a decade.
- 18:16SPEAKER_01: You seem to be more optimistic than that.
- 18:18SPEAKER_01: I don't want to put words in your mouth.
- 18:19SPEAKER_01: But if you are more optimistic than that, then what are they missing?
- 18:24SPEAKER_00: Well,
- 18:24SPEAKER_00: I've predicted half a percentage point added to GDP or productivity growth over,
- 18:29SPEAKER_00: say, the next 10 years.
- 18:31SPEAKER_00: So half a point is more optimistic than a quarter of a point.
- 18:35SPEAKER_00: And it compounds over time, but it's not that different.
- 18:39SPEAKER_00: I think a major difference is just I've been studying these models for a long time.
- 18:45SPEAKER_00: I studied the topic in 2013.
- 18:48SPEAKER_00: I have a long-standing history of following chess and AI in chess,
- 18:52SPEAKER_00: where I've seen how well that gone and how many problems it conquered,
- 18:56SPEAKER_00: and I've spent a lot of time visiting the major labs and talking to people,
- 19:01SPEAKER_00: and I feel I'm pretty well informed,
- 19:03SPEAKER_00: and most economists are busy doing other things,
- 19:06SPEAKER_00: which is fine,
- 19:07SPEAKER_00: but I think in terms of knowledge of AI and taking seriously what
- 19:11SPEAKER_00: the lab people say,
- 19:13SPEAKER_00: I'm just ahead of most economists,
- 19:15SPEAKER_00: but someone like Eric Brynolfsson, I'm not ahead of him in this regard.
- 19:19SPEAKER_00: He lives in the Bay Area.
- 19:21SPEAKER_00: He goes to these places.
- 19:22SPEAKER_00: He talks to people.
- 19:23SPEAKER_00: There's about 20 other people like that I could name.
- 19:26SPEAKER_00: Their numbers are growing rapidly.
- 19:29SPEAKER_00: But I'm one of them.
- 19:31SPEAKER_00: A good question is simply, have you ever been to the labs?
- 19:35SPEAKER_00: And that predicts pretty well how much an economist is clued in on this topic.
- 19:42SPEAKER_01: I've sort of talked to economists who would say,
- 19:46SPEAKER_01: have you been to the labs,
- 19:49SPEAKER_01: means that you have been,
- 19:51SPEAKER_01: not you,
- 19:51SPEAKER_01: but more optimistic economists have sort of been bedazzled by the technologists and
- 19:59SPEAKER_01: the demos and fancy graphics,
- 20:03SPEAKER_01: PowerPoint presentations,
- 20:05SPEAKER_01: and that That ends up weighing too much into your analysis,
- 20:12SPEAKER_01: spending time in sort of,
- 20:14SPEAKER_01: you know, walking through the reality filter to the Bay Area.
- 20:18SPEAKER_00: You should be bedazzled.
- 20:19SPEAKER_00: That's the whole point.
- 20:20SPEAKER_00: And look, you can go back to when GPT-4 was released.
- 20:24SPEAKER_00: I remember how many people complained,
- 20:25SPEAKER_00: there's hallucination, it can't do this, it can't do that.
- 20:28SPEAKER_00: How many R's are in Strawberry?
- 20:30SPEAKER_00: I, along with many other people, said, look, in a year or two, all this will be patched up.
- 20:36SPEAKER_00: And we were completely correct.
- 20:37SPEAKER_00: And the doubters were completely wrong.
- 20:39SPEAKER_00: So if you want to look at recent predictions, we've been acing it.
- 20:43SPEAKER_00: And the critics and the Gary Marcus types and whoever else, they've failed.
- 20:48SPEAKER_00: They've fallen flat on their faces.
- 20:50SPEAKER_00: So it doesn't mean we're right going forward, but I think it's relevant information.
- 20:56SPEAKER_01: So right now,
- 20:57SPEAKER_01: I'm looking for signs that businesses are using this technology in a productive,
- 21:03SPEAKER_01: high return on investment way.
- 21:05SPEAKER_01: And, you know, I pay attention to earnings calls, what they're saying there, that kind of thing.
- 21:13SPEAKER_01: Is there something that would happen in which you would raise sort
- 21:18SPEAKER_01: of your baseline economic forecast to something that does look more sort
- 21:23SPEAKER_01: of AGI pill,
- 21:24SPEAKER_01: not a half a point,
- 21:25SPEAKER_01: but Something that would really take us out of like what we've traditionally seen
- 21:31SPEAKER_01: for the economic impact of these technologies.
- 21:33SPEAKER_01: So growth of,
- 21:34SPEAKER_01: you know,
- 21:35SPEAKER_01: economic growth for a big economy like ours of 5%,
- 21:37SPEAKER_01: 7%, 10%, and certainly there are people moving that decimal point over.
- 21:42SPEAKER_01: What would sort of really get you gung-ho?
- 21:45SPEAKER_01: Can you imagine some sort of something happening like that?
- 21:49SPEAKER_00: The thing being smarter is not the problem.
- 21:51SPEAKER_00: It's already very smart, the things.
- 21:54SPEAKER_00: The real issue is,
- 21:55SPEAKER_00: if you're in a mid-size to large organization,
- 21:58SPEAKER_00: reorganizing your systems of HR,
- 22:01SPEAKER_00: finance,
- 22:01SPEAKER_00: whatever,
- 22:02SPEAKER_00: all the things they do at AEI,
- 22:04SPEAKER_00: not just where people on an individual basis use, say,
- 22:08SPEAKER_00: ChatGPT to augment their labor,
- 22:11SPEAKER_00: but actually building the whole system around AI in a way that's reliable
- 22:14SPEAKER_00: and coherent and everyone there knows how to work with.
- 22:17SPEAKER_00: That's very, very,
- 22:18SPEAKER_00: very hard and slow,
- 22:20SPEAKER_00: and if I saw more progress on that,
- 22:23SPEAKER_00: Then I would really think,
- 22:24SPEAKER_00: oh, the rates of productivity growth will be much higher soon.
- 22:27SPEAKER_00: But I'm not seeing more progress on that.
- 22:30SPEAKER_00: Human beings are sticky.
- 22:31SPEAKER_00: There's plenty of bottlenecks.
- 22:33SPEAKER_00: No one quite knows how to do it.
- 22:34SPEAKER_00: You can't just call up McKinsey and ask them, even if you're willing to pay their fees.
- 22:39SPEAKER_00: And that's where we're stuck.
- 22:42SPEAKER_00: And just getting a smarter AI,
- 22:43SPEAKER_00: most people don't even notice that, say, the new model is better than the old model.
- 22:48SPEAKER_00: They don't know enough themselves to tell the difference.
- 22:51SPEAKER_00: So it's the human element.
- 22:53SPEAKER_00: And restructuring, you know, divisions within larger organizations.
- 22:57SPEAKER_00: There's plenty of AI native startups with small numbers of people
- 23:00SPEAKER_00: that right now are doing amazing things.
- 23:02SPEAKER_00: They're just not that big a portion of the economy.
- 23:07SPEAKER_01: There's often called the Forecasting Research Institute,
- 23:10SPEAKER_01: and they've been running surveys of academic economists,
- 23:13SPEAKER_01: AI experts, super forecasters.
- 23:16SPEAKER_01: And when they do these surveys, they give them various progress scenarios.
- 23:21SPEAKER_01: And like their slow progress scenario is,
- 23:26SPEAKER_01: think of the year 2030,
- 23:28SPEAKER_01: and in that year,
- 23:30SPEAKER_01: AI can write literature reviews at the level of a capable PhD student,
- 23:35SPEAKER_01: handle one day freelance software engineering jobs.
- 23:39SPEAKER_01: Maybe we'll have robots that can unload dishwashers.
- 23:44SPEAKER_01: Is that world a world, if it should happen, that you would view as slow progress?
- 23:49SPEAKER_01: Would you expect more than that over the next four to five years?
- 23:53SPEAKER_00: Well, that study already sounds off base to me.
- 23:56SPEAKER_00: Writing literature reviews at a PhD level, they can do that now.
- 24:03SPEAKER_00: The robot unloading the dishwasher, that's a harder projection.
- 24:08SPEAKER_00: But I don't think in an economically useful way that will be possible by 2030.
- 24:12SPEAKER_00: So I would say two of the main predictions are already quite off.
- 24:17SPEAKER_00: So what do I think of the survey?
- 24:20SPEAKER_00: They had three scenarios.
- 24:21SPEAKER_01: That was the slow scenario.
- 24:22SPEAKER_01: They also had more rapid ones.
- 24:25SPEAKER_00: Yeah, but the slow scenario thinks something takes four years that's already true.
- 24:29SPEAKER_00: So maybe they just don't know enough to be giving good forecasts.
- 24:33SPEAKER_00: For me, the slow scenario...
- 24:35SPEAKER_00: Is we have another five years where productivity hardly budges upward and that's possible.
- 24:40SPEAKER_00: It wouldn't be my prediction, but it's not at all to be ruled out.
- 24:47SPEAKER_01: Can we have a world of extremely advanced AI?
- 24:54SPEAKER_01: You can give it whatever term you want.
- 24:58SPEAKER_01: So maybe 10 years from now, I don't know.
- 25:01SPEAKER_01: Can we have a world of extremely advanced,
- 25:03SPEAKER_01: capable AI, and also a world with lots of jobs for people?
- 25:10SPEAKER_01: And those things, well, those things fundamentally just come into context.
- 25:13SPEAKER_00: The word eventually always makes me nervous.
- 25:16SPEAKER_00: In any foreseeable future, we can achieve full employment if we don't screw up other policies.
- 25:22SPEAKER_00: Just think of the number of jobs that will be available,
- 25:25SPEAKER_00: testing the new ideas that AIs come up with, most of all in the biomedical field.
- 25:30SPEAKER_00: Think of all the new jobs that will be available, gathering and processing data for the AIs.
- 25:36SPEAKER_00: So just in AEI, there's so much data about how to run a nonprofit, right?
- 25:41SPEAKER_00: And what you've all done.
- 25:42SPEAKER_00: And referee reports.
- 25:44SPEAKER_00: And I strongly suspect none of that is fed into either Claude or, you know, ChatGPT.
- 25:50SPEAKER_00: So that's an incredible job.
- 25:51SPEAKER_00: Thank you.
- 26:11SPEAKER_00: Gathering it and getting that into a form for the AIs to digest it,
- 26:15SPEAKER_00: it wouldn't shock me if in some distant future that was like a third of
- 26:18SPEAKER_00: the whole labor force,
- 26:20SPEAKER_00: the way today,
- 26:20SPEAKER_00: you know,
- 26:21SPEAKER_00: services are some very high percent of the labor force,
- 26:24SPEAKER_00: that people in the Industrial Revolution never would have mattered.
- 26:28SPEAKER_00: So there's entertainment,
- 26:29SPEAKER_00: there's leisure,
- 26:30SPEAKER_00: there's travel,
- 26:30SPEAKER_00: there's personal services,
- 26:32SPEAKER_00: there's, you know, the paradigmatic gardener, and so on and so on.
- 26:37SPEAKER_00: We'll have the chance to transform the whole earth if we want to.
- 26:39SPEAKER_00: Fertility crisis may intersect with that,
- 26:42SPEAKER_00: but there'll be so many projects and so many jobs connected to those projects.
- 26:47SPEAKER_00: Scarcity of the proper talent will be a much bigger issue.
- 26:52SPEAKER_01: You mentioned China a little earlier about the national security impetus
- 26:55SPEAKER_01: of us continuing to push this technology forward.
- 27:00SPEAKER_01: I sort of remember from my days majoring in Soviet foreign relations back in college
- 27:08SPEAKER_01: that as a country we spent a lot of time trying to think about how
- 27:13SPEAKER_01: the Soviet Union thought about nuclear weapons, nuclear war strategy.
- 27:19SPEAKER_01: And what I sort of remember is that we didn't think the same way.
- 27:22SPEAKER_01: We may have broadly thought about mutual assured destruction as an outcome,
- 27:28SPEAKER_01: But how that actually affected what we did was different.
- 27:31SPEAKER_01: I started remembering the Soviets being more willing to sort of fight and win like
- 27:36SPEAKER_01: a limited nuclear war.
- 27:38SPEAKER_01: They had perhaps a more permissive attitude toward that.
- 27:41SPEAKER_01: Do you feel like we really understand how China thinks about AI and AGI and
- 27:46SPEAKER_01: the national security implications?
- 27:48SPEAKER_00: I don't think we understand it.
- 27:50SPEAKER_00: I'm not sure they understand themselves,
- 27:52SPEAKER_00: but it seems to me as a non-Christian society I just read
- 27:56SPEAKER_00: a Wall Street Journal story, venture capitalists.
- 28:21SPEAKER_01: Are pivoting to physical experiences as they seek to insulate themselves from AI.
- 28:27SPEAKER_01: One example is buying sports teams.
- 28:30SPEAKER_01: Does that seem like a particularly good investment in an age of ubiquitous, powerful AI?
- 28:36SPEAKER_00: Absolutely.
- 28:37SPEAKER_00: Now, $12 billion for the Lakers nonetheless struck me as a bit high.
- 28:42SPEAKER_00: But events in general,
- 28:43SPEAKER_00: live events,
- 28:44SPEAKER_00: have grown phenomenally over the last few years, and I'm very long on that sector.
- 28:50SPEAKER_00: And for other reasons, nothing to do with AI.
- 28:52SPEAKER_00: There's this loneliness epidemic in America.
- 28:55SPEAKER_00: People want to get out and see other people and entrepreneurs are figuring out ways
- 28:59SPEAKER_00: to make that happen.
- 29:00SPEAKER_00: That's great.
- 29:02SPEAKER_00: I assume AI is doing, you know, some of its own.
- 29:04SPEAKER_00: Good for you.
- 29:06SPEAKER_01: How damaging do you think it would be if protectionism and industrial policy became
- 29:13SPEAKER_01: the baseline economic approach
- 29:22SPEAKER_00: Welcome to 2026.
- 29:24SPEAKER_00: I'm totally against it, but I think a lot of free market types overrate how bad it is.
- 29:30SPEAKER_00: I think if you have,
- 29:32SPEAKER_00: you know,
- 29:32SPEAKER_00: the wind behind your sails on tech and the real sector of the economy,
- 29:37SPEAKER_00: that you can survive a lot more stupidity than policy economists like to think.
- 29:42SPEAKER_00: So whether your tastes in technology are matched up is underrated as a factor,
- 29:47SPEAKER_00: and how good policy is is overrated.
- 29:50SPEAKER_00: But what you said is today, and I hate it.
- 29:53SPEAKER_01: In Stubborn Attachments,
- 29:54SPEAKER_01: your book Stubborn Attachments,
- 29:56SPEAKER_01: you wrote that,
- 29:57SPEAKER_01: above all,
- 29:57SPEAKER_01: we should focus on maximizing sustainable economic growth and trying to raise
- 30:03SPEAKER_01: that rate of growth.
- 30:06SPEAKER_00: People seem
- 30:08SPEAKER_01: averse to that.
- 30:11SPEAKER_00: Or indifferent, or they don't know what it
- 30:12SPEAKER_01: means.
- 30:13SPEAKER_01: Or they think that if the economy grows faster, only the rich people will benefit.
- 30:19SPEAKER_01: Maybe it'll be neutral on them.
- 30:22SPEAKER_01: Maybe it'll be harmful for them.
- 30:25SPEAKER_01: It feels like that opinion has grown.
- 30:27SPEAKER_01: Does that matter, that people think that?
- 30:29SPEAKER_00: Of course, and it's terrible.
- 30:31SPEAKER_00: And you see all these bad policies like rent control coming back.
- 30:34SPEAKER_00: That superficially address affordability,
- 30:37SPEAKER_00: but don't augment supply, and in fact will decrease supply.
- 30:40SPEAKER_00: And they're just flat out mistakes, which economists would tell you on a bipartisan basis.
- 30:46SPEAKER_00: And I think we're going to see a lot more of that, and you see it from both parties.
- 30:51SPEAKER_01: The nuclear war theorist,
- 30:52SPEAKER_01: something I've written about him,
- 30:54SPEAKER_01: Herman Kahn,
- 30:55SPEAKER_01: who in the 1970s became a very sunny, sort of techno-capitalist futurist.
- 31:00SPEAKER_01: Back in the 70s,
- 31:01SPEAKER_01: he was asked if he disagreed With sort of the limits to growth idea
- 31:09SPEAKER_01: that we had limits to growth,
- 31:10SPEAKER_01: that we're using up all the resources, too many people and all that.
- 31:13SPEAKER_01: And they asked him if you disagree with that notion and he said,
- 31:16SPEAKER_01: I absolutely disagree given only two caveats.
- 31:20SPEAKER_01: One, we don't have disastrously bad luck and we don't do things that are incredibly dumb.
- 31:28SPEAKER_01: It takes a moderate but not extraordinarily good level of decision making
- 31:33SPEAKER_01: To overcome the problems we can imagine in the future.
- 31:38SPEAKER_01: Do you think American decision making meets or clears that hurdle?
- 31:44SPEAKER_01: Is it good enough to deal with the problems we face?
- 31:48SPEAKER_00: I think Herman Kahn was a little too pessimistic on that issue.
- 31:52SPEAKER_00: So our decision making pretty often has been awful, like say stopping nuclear.
- 31:57SPEAKER_00: But just something like progress in solar and batteries or possibly fusion will pay off.
- 32:04SPEAKER_00: A lot of breaks have gone our way, actually, in the last 10 years.
- 32:08SPEAKER_00: You could say it's very good decentralized decision making,
- 32:11SPEAKER_00: but it's going to go very well for us in terms of resource scarcity,
- 32:15SPEAKER_00: making the economy more intangible, coming up with new sources of supply.
- 32:20SPEAKER_00: There are very few supplies that are not elastic, sufficiently elastic to keep us going.
- 32:29SPEAKER_00: It's really been the Julian Simon scenario.
- 32:32SPEAKER_01: Tyler, thanks for coming on the podcast.
- 32:34SPEAKER_00: My pleasure, James.
- 32:35SPEAKER_00: Thank you.