Latest / Faster, Please! — The Podcast / ✨💪 AI, jobs, and productivity: My chat with economist Erik Brynjolfsson
Transcript
- 0:00SPEAKER_01: Will artificial intelligence displace workers or make them more valuable?
- 0:05SPEAKER_01: Probably plenty of both.
- 0:07SPEAKER_01: But how much and how fast will all this change arrive?
- 0:11SPEAKER_01: Today on Faster, Please!
- 0:12SPEAKER_01: The Podcast,
- 0:13SPEAKER_01: I'm joined by Eric Brynjolfsson,
- 0:15SPEAKER_01: one of the world's top economists studying how AI is reshaping productivity,
- 0:20SPEAKER_01: jobs, and the economy.
- 0:23SPEAKER_01: He's the co-author,
- 0:24SPEAKER_01: along with Andrew McAfee,
- 0:25SPEAKER_01: of The Second Machine Age,
- 0:27SPEAKER_01: and author of The Turing Trap,
- 0:29SPEAKER_01: co-founder of Work Helix, and director of the Stanford Digital Economy Lab.
- 0:35SPEAKER_01: Why should workers be excited?
- 0:37SPEAKER_01: Workers in particular, why should they be excited by the AI revolution?
- 0:41SPEAKER_01: They hear CEOs,
- 0:44SPEAKER_01: you know,
- 0:44SPEAKER_01: predict that if you sit behind a computer, your job will be gone in 18 months.
- 0:50SPEAKER_01: They've tempered their predictions a bit lately,
- 0:52SPEAKER_01: but still,
- 0:53SPEAKER_01: I certainly could forgive someone for thinking that unless I'm in the business
- 0:59SPEAKER_01: of constructing a data center and I'm an actual worker building a data center,
- 1:04SPEAKER_01: You know,
- 1:06SPEAKER_00: it's so frustrating that the narrative has become dominated by the doom
- 1:10SPEAKER_00: and gloom side.
- 1:11SPEAKER_00: I don't want to sugarcoat it.
- 1:12SPEAKER_00: There are a lot of disruption.
- 1:13SPEAKER_00: There are jobs disappearing.
- 1:16SPEAKER_00: But I think the bigger story is all the stuff that's being created.
- 1:19SPEAKER_00: Let me tell you a little anecdote.
- 1:20SPEAKER_00: It's a little bit sad,
- 1:22SPEAKER_00: but I think it has a happy ending, which is that, you know, there was a...
- 1:27SPEAKER_00: There was a graduating senior who came to my office hours about a month ago,
- 1:32SPEAKER_00: and she was very distraught.
- 1:33SPEAKER_00: She said she did not have a job,
- 1:36SPEAKER_00: and her friends didn't have jobs,
- 1:37SPEAKER_00: and she said, quote, you know, is my generation doomed?
- 1:42SPEAKER_00: You know,
- 1:42SPEAKER_00: she was she was worried and I didn't want to say,
- 1:44SPEAKER_00: hey, you know, there's nothing wrong.
- 1:46SPEAKER_00: This is it is a challenge for people.
- 1:48SPEAKER_00: The data show that college graduates are having a harder time getting jobs than before.
- 1:53SPEAKER_00: But I'm teaching a class called the AI Awakening.
- 1:56SPEAKER_00: And the point of the class is that everybody in it is building real products with
- 2:03SPEAKER_00: Vibe Coding.
- 2:04SPEAKER_00: They're using Cloud, Co-Work, and other tools to create new kinds of enterprises.
- 2:11SPEAKER_00: In past years, they would have ended the class with a PowerPoint presentation.
- 2:14SPEAKER_00: This year, I said, no, no, no, none of that.
- 2:16SPEAKER_00: You have to have a working product.
- 2:19SPEAKER_00: And that's something that just wouldn't have been possible five years ago or even one year ago.
- 2:25SPEAKER_00: And so I told this student,
- 2:26SPEAKER_00: you know, you got to lean into that side of it, all the stuff that can be created.
- 2:31SPEAKER_00: This is, I think, the best time to be alive.
- 2:33SPEAKER_00: For somebody who's got agency and ambition and intention and wants to do something,
- 2:38SPEAKER_00: create new things for themselves and for the world.
- 2:40SPEAKER_00: But that's not the story that's out there.
- 2:42SPEAKER_00: It's all focused on the part that's disappearing, which is real.
- 2:45SPEAKER_00: But equally important, or I think more important, is the opportunities to create new things.
- 2:51SPEAKER_01: If the story is that this is a powerful new technology,
- 2:58SPEAKER_01: and we've had other powerful new technologies,
- 3:02SPEAKER_01: and there is labor market disruption and people's jobs,
- 3:05SPEAKER_01: some jobs may disappear over time,
- 3:08SPEAKER_01: some jobs will change, but will create new businesses and new jobs.
- 3:13SPEAKER_01: If that is a story you're telling,
- 3:17SPEAKER_01: I think it's highly believable and people should pay attention to it
- 3:21SPEAKER_01: and they should have confidence in it.
- 3:23SPEAKER_01: But the story also being told by a lot of people in Silicon Valley
- 3:27SPEAKER_01: that that's not what we're talking about.
- 3:28SPEAKER_01: We're talking about a very different kind of technology that
- 3:31SPEAKER_01: at first may look like other things.
- 3:34SPEAKER_01: General purpose technologies that are powerful, but this is going to be something different.
- 3:38SPEAKER_01: It'll be the last invention we ever need.
- 3:39SPEAKER_01: It'll be general intelligence, super intelligence.
- 3:42SPEAKER_01: And I think people hear that and they're like,
- 3:44SPEAKER_01: I can believe the first story, but they're saying it's going to be way beyond.
- 3:48SPEAKER_01: And how will it not do pretty much all the work unless you're a ballet dancer or
- 3:55SPEAKER_01: an artisanal potter?
- 3:59SPEAKER_00: There's some truth to that.
- 4:01SPEAKER_00: I do think this wave is bigger than anything since the Industrial Revolution.
- 4:05SPEAKER_00: You know,
- 4:05SPEAKER_00: Andy McAfee and I wrote The Second Machine Age comparing AI
- 4:09SPEAKER_00: and cognitive technologies to that scale.
- 4:12SPEAKER_00: And it's going to happen 10 times faster.
- 4:15SPEAKER_00: So that part is true.
- 4:18SPEAKER_00: That said,
- 4:19SPEAKER_00: we all have a cognitive bias that we can see what already exists
- 4:22SPEAKER_00: and imagine how technology can do it.
- 4:25SPEAKER_00: It's much harder for us to see things that have never existed before, all the new inventions.
- 4:30SPEAKER_00: And so it's not hard to imagine how AI will be able to do some of
- 4:34SPEAKER_00: the things we're already doing.
- 4:36SPEAKER_00: But these new discoveries,
- 4:38SPEAKER_00: that's something that we're going to have to figure out as we go along.
- 4:41SPEAKER_00: That's the job of entrepreneurs, that's the job of inventors.
- 4:44SPEAKER_00: And frankly, most of the time,
- 4:45SPEAKER_00: I'm sitting here next to Sand Hill Road,
- 4:50SPEAKER_00: and I talk to the VCs, and 90% of their investments are not home runs.
- 4:54SPEAKER_00: They make all their returns on a small sliver.
- 4:57SPEAKER_00: Those are the ones that really pay off.
- 4:59SPEAKER_00: But they go ahead and invest in the other ones.
- 5:00SPEAKER_00: I'm grateful to all the people who are willing to put blood,
- 5:02SPEAKER_00: sweat, and tears into these ventures that are sometimes long gone.
- 5:16SPEAKER_00: But some percentage of them pay off, and that's where all the value comes.
- 5:20SPEAKER_00: In my paper, The Turing Trap, I have this story of Daedalus, the mythical engineer.
- 5:26SPEAKER_00: And according to legend, he made robots that were lifelike.
- 5:31SPEAKER_00: You couldn't distinguish them from humans.
- 5:33SPEAKER_00: They passed the Turing test.
- 5:34SPEAKER_00: I'm pretty sure it was just a legend.
- 5:36SPEAKER_00: But suppose he really had done that.
- 5:39SPEAKER_00: And suppose he had then set them to work at automating all the things
- 5:43SPEAKER_00: the ancient Greeks did, you know, so you would have as many clay pots as you wanted.
- 5:47SPEAKER_00: My room could be filled with them, you know, tunics, chariots.
- 5:50SPEAKER_00: If you were sick, they would burn incense for you.
- 5:53SPEAKER_00: No one would have to work anymore.
- 5:54SPEAKER_00: Everything would be taken care of.
- 5:56SPEAKER_00: So the good news is,
- 5:57SPEAKER_00: you know,
- 5:57SPEAKER_00: there would be great abundance, you know, technically infinite productivity.
- 6:01SPEAKER_00: But at the same time, we could see that if all they did was just
- 6:05SPEAKER_00: Automate what already existed,
- 6:06SPEAKER_00: that would be a small fraction of the wealth that you and I and the rest
- 6:10SPEAKER_00: of Americans enjoy today.
- 6:13SPEAKER_00: Most of our increase in living standards comes from new products and services that
- 6:17SPEAKER_00: the Greeks literally couldn't imagine.
- 6:20SPEAKER_00: And that was true between 2,000 years ago and today.
- 6:23SPEAKER_00: I don't think we're at the end of the history.
- 6:25SPEAKER_00: Going forward,
- 6:26SPEAKER_00: I think there's going to be just as many new opportunities and things
- 6:29SPEAKER_00: that are being invented.
- 6:31SPEAKER_00: And humans plus machines put together is what's going to create those new opportunities.
- 6:37SPEAKER_00: And that's the part we should be trying to focus on,
- 6:41SPEAKER_00: the part we should be enabling, and the part that gives me a lot of optimism.
- 6:45SPEAKER_01: Is it?
- 6:46SPEAKER_01: And then I kind of want to move on.
- 6:47SPEAKER_01: But can I hold both these beliefs in my head?
- 6:51SPEAKER_01: And it kind of should make sense.
- 6:54SPEAKER_01: Can I believe that there will be lots of jobs in the future?
- 7:02SPEAKER_01: People will find new things to do.
- 7:05SPEAKER_01: And if you look at a labor market in 10 years,
- 7:07SPEAKER_01: you would find all kinds of work and it would not look dystopian.
- 7:12SPEAKER_01: Can I believe that?
- 7:14SPEAKER_01: And can I also believe
- 7:16SPEAKER_01: That AI,
- 7:18SPEAKER_01: sooner rather than later maybe,
- 7:19SPEAKER_01: will advance to a point where you could easily call it AGI or superintelligence.
- 7:26SPEAKER_01: Can we have really advanced AI plus lots and lots of jobs?
- 7:30SPEAKER_01: Can I hold both those beliefs in my head and not be crazy?
- 7:34SPEAKER_00: I think you can.
- 7:35SPEAKER_00: I think one of the things that people need to think hard about is the timelines.
- 7:42SPEAKER_00: The world I'm talking about where all this stuff has been created
- 7:44SPEAKER_00: and humans continue to have a role,
- 7:45SPEAKER_00: I think it's going to extend for many years, decades.
- 7:49SPEAKER_00: That said,
- 7:49SPEAKER_00: I can't imagine a far future time when there's little or nothing for humans to add
- 7:55SPEAKER_00: to the equation.
- 7:58SPEAKER_00: When Garry Kasparov created Freestyle Chess,
- 8:01SPEAKER_00: Centaur Chess,
- 8:02SPEAKER_00: for a while the humans and machines defeated the best machines and the best humans,
- 8:08SPEAKER_00: of course.
- 8:09SPEAKER_00: Eventually, there was little for the humans to add.
- 8:13SPEAKER_00: It lasted for a little 10 to 12 years or so.
- 8:16SPEAKER_00: And I actually take some optimism from that because that was just a little board
- 8:20SPEAKER_00: with,
- 8:20SPEAKER_00: you know,
- 8:21SPEAKER_00: eight black and white squares by eight black and white squares,
- 8:23SPEAKER_00: a very constrained world.
- 8:25SPEAKER_00: And it took that long for machines to be able to beat humans and to beat the centaurs at that.
- 8:32SPEAKER_00: The real world is much more complicated.
- 8:34SPEAKER_00: And I think we're going to have several decades worth of humans and machines working together.
- 8:38SPEAKER_00: And I'd like to extend that window where we can
- 8:43SPEAKER_00: We should probably be preparing for some further time in the future
- 8:55SPEAKER_00: when there's less of a role for people.
- 8:57SPEAKER_00: But most of my friends here in Silicon Valley,
- 8:59SPEAKER_00: I think their timelines are way too short for when humans no longer have a role.
- 9:05SPEAKER_00: I think they're probably right about a lot of the
- 9:07SPEAKER_00: Is that a question?
- 9:30SPEAKER_01: Is that a good world?
- 9:31SPEAKER_01: That world where humans don't have to add,
- 9:34SPEAKER_01: can't generate a lot of economic,
- 9:36SPEAKER_01: whether it's, you know, five, whether it's maybe it's 30 years away.
- 9:39SPEAKER_01: I don't know.
- 9:40SPEAKER_01: But that world you described, is that a good world?
- 9:44SPEAKER_01: Is that a world you would want to live in?
- 9:45SPEAKER_01: Is that a world I should want my kids to live in?
- 9:50SPEAKER_00: The honest answer is, I don't know.
- 9:52SPEAKER_00: I think the good news is to be a lot of abundance.
- 9:56SPEAKER_00: I also worry that when humans don't have bargaining power
- 10:01SPEAKER_00: and don't have economic leverage,
- 10:03SPEAKER_00: that there could be a lot of centralization,
- 10:05SPEAKER_00: not just of wealth, but of political power.
- 10:08SPEAKER_00: And that could be problematic.
- 10:11SPEAKER_00: I want to emphasize where we are for the next few decades where humans still have a role.
- 10:18SPEAKER_00: I think we should start preparing for a period where AI can do almost everything
- 10:23SPEAKER_00: and we need to come up with mechanisms that we
- 10:26SPEAKER_00: We still have freedom and power in that kind of world.
- 10:30SPEAKER_00: I don't think that's automatic.
- 10:32SPEAKER_00: And one of my biggest concerns, to be frank, is not that we don't have abundance.
- 10:36SPEAKER_00: I think we will.
- 10:37SPEAKER_00: But it's that we don't have freedom and autonomy.
- 10:41SPEAKER_00: And that's something that's not to be taken for granted.
- 10:43SPEAKER_00: And we need to put in place ways that we not only have the wealth,
- 10:48SPEAKER_00: but we also have widely shared prosperity and widely shared decision making rights.
- 10:55SPEAKER_01: That was the sort of Silicon Valley portion of our conversation.
- 11:00SPEAKER_01: Because when I speak to a lot of economists not out there,
- 11:04SPEAKER_01: but I'm on the East Coast and Washington and,
- 11:07SPEAKER_01: you know,
- 11:08SPEAKER_01: people at the Fed or at investment banks,
- 11:12SPEAKER_01: their concern is not super intelligence and what we'll do is they're wondering
- 11:19SPEAKER_01: Is this a technology that businesses have yet to find economically valuable?
- 11:25SPEAKER_01: Whatever AI does on benchmarks or in the lab,
- 11:33SPEAKER_01: they question whether it's actually doing anything yet in actual businesses
- 11:38SPEAKER_01: to produce value.
- 11:39SPEAKER_00: Is it?
- 11:40SPEAKER_00: That's a great question, and I think that's what we should be focused on right now.
- 11:44SPEAKER_00: I've never seen such a gap between the perception, say,
- 11:48SPEAKER_00: in the Bay Area and on the East Coast or the rest of the world,
- 11:51SPEAKER_00: or between the technologists and the business users of it.
- 11:56SPEAKER_00: I think we all can see that the technology is just mind-blowing.
- 11:59SPEAKER_00: What it does is almost magical.
- 12:02SPEAKER_00: We produce this thing called the Stanford AI Index,
- 12:04SPEAKER_00: and it just has 500 pages of charts going up and to the right faster and faster.
- 12:10SPEAKER_00: It's breathtaking on the technical capabilities.
- 12:13SPEAKER_00: At the same time,
- 12:13SPEAKER_00: as you know,
- 12:14SPEAKER_00: the productivity numbers, you know, they're not really moving that much.
- 12:17SPEAKER_00: You know, last year we had 2.8%, which is a little better.
- 12:20SPEAKER_00: There may be some evidence that it's beginning to pick up,
- 12:22SPEAKER_00: but I think it's too early to say much about that.
- 12:25SPEAKER_00: And if you talk to businesses, there's a lot of frustration out there.
- 12:30SPEAKER_00: These projects are not instantly delivering.
- 12:35SPEAKER_00: Some companies are having some success,
- 12:37SPEAKER_00: some aren't,
- 12:37SPEAKER_00: but if you look at them more broadly,
- 12:39SPEAKER_00: I think that the returns have been somewhat disappointing.
- 12:44SPEAKER_00: To me, that's totally natural.
- 12:47SPEAKER_00: That's totally understandable.
- 12:50SPEAKER_00: We've done a lot of work.
- 12:51SPEAKER_00: We call it the Productivity J Curve on the need for complementary investments,
- 12:55SPEAKER_00: for intangible investments in new business process design,
- 12:59SPEAKER_00: in new skills for the workforce, even new products and services.
- 13:04SPEAKER_00: Those take time.
- 13:05SPEAKER_00: In with past general purpose technologies like the steam engine and electricity,
- 13:10SPEAKER_00: it took literally decades before you got those returns.
- 13:13SPEAKER_00: I don't expect it to take decades this time, you know, years perhaps.
- 13:17SPEAKER_00: And we're just beginning to turn the corner on that.
- 13:20SPEAKER_00: This is normal.
- 13:21SPEAKER_00: This is to be expected that these incredibly powerful technical capabilities
- 13:26SPEAKER_00: Don't translate to business productivity and improve business performance
- 13:31SPEAKER_00: until you make the complementary changes.
- 13:33SPEAKER_00: That's what companies are doing.
- 13:35SPEAKER_00: I have a startup called Work Helix that helps them do that more rapidly.
- 13:39SPEAKER_00: And as we help them use the
- 13:42SPEAKER_00: I think within the next couple of years, it's going to be visible in the aggregate numbers.
- 13:49SPEAKER_00: Right now,
- 13:49SPEAKER_00: it's only visible in a few use cases,
- 13:53SPEAKER_00: in call centers,
- 13:54SPEAKER_00: in coding, parts of sales for the companies that are doing it effectively.
- 14:00SPEAKER_00: But I'm not at all surprised by that gap.
- 14:04SPEAKER_01: So, I think you wrote this,
- 14:07SPEAKER_01: maybe it was an interview with you,
- 14:08SPEAKER_01: you wrote in the Financial Times earlier this year,
- 14:11SPEAKER_01: that to the extent companies are using technology,
- 14:14SPEAKER_01: it's in really narrow ways,
- 14:16SPEAKER_01: sort of the summarizing or translation and drafting,
- 14:21SPEAKER_01: but they're not redesigning workflows around this technology.
- 14:25SPEAKER_01: Have you seen, do you have any examples, whether it's
- 14:27SPEAKER_01: Yeah, there are a few companies,
- 14:29SPEAKER_01: you know,
- 14:30SPEAKER_01: at WorkHelix we work with Nasdaq and Ford and other companies and we're beginning to
- 14:42SPEAKER_00: To see them happen there,
- 14:43SPEAKER_00: you know,
- 14:45SPEAKER_00: at every company,
- 14:46SPEAKER_00: I see a very similar pattern,
- 14:47SPEAKER_00: which is there's basically a power law even within the company as well as
- 14:51SPEAKER_00: across companies.
- 14:52SPEAKER_00: But within the companies,
- 14:53SPEAKER_00: there are a few super users,
- 14:55SPEAKER_00: you know,
- 14:55SPEAKER_00: single digit percentage of people in sales and coding and marketing
- 15:00SPEAKER_00: who are literally 10x.
- 15:03SPEAKER_00: They're using the technology incredibly effectively.
- 15:07SPEAKER_00: They're using it many hours a day.
- 15:08SPEAKER_00: They're burning through millions of tokens.
- 15:11SPEAKER_00: And they're just like super users.
- 15:13SPEAKER_00: But the majority of users is this long tail and the median user is basically,
- 15:17SPEAKER_00: you know, writing more eloquent emails or summarizing some documents.
- 15:21SPEAKER_00: You know, it's nice to have, but it's not fundamentally changing what they're doing.
- 15:25SPEAKER_00: To me, that's good news,
- 15:26SPEAKER_00: because what it means is there's a huge opportunity to level up the typical user up
- 15:31SPEAKER_00: to the level of that super user or even halfway to that super user.
- 15:35SPEAKER_00: And so what we're doing at WorkHelix is we're showing them what the best users are doing.
- 15:40SPEAKER_00: Using that as templates for the rest of the organization.
- 15:42SPEAKER_00: There's sometimes,
- 15:43SPEAKER_00: you know,
- 15:43SPEAKER_00: there's somebody on the same floor or next door who's doing kind of the same work,
- 15:48SPEAKER_00: but doing it much more effectively.
- 15:50SPEAKER_00: And that means there's a lot of low hanging fruit.
- 15:53SPEAKER_00: Over the next few months and years,
- 15:54SPEAKER_00: we're going to see more and more individual users and more
- 15:57SPEAKER_00: and more companies leveling up in that way.
- 15:59SPEAKER_00: And that's why I'm so optimistic about the productivity gains that we're likely to have.
- 16:06SPEAKER_01: There was an economist piece out the other day looking
- 16:09SPEAKER_01: at what I think they called AI-pilled or AGI-pilled economists.
- 16:14SPEAKER_01: And leading the pack, the most optimistic, you may disagree with that, you led.
- 16:21SPEAKER_01: You were the most AI-pilled economist.
- 16:26SPEAKER_01: But it's listed from most to least.
- 16:29SPEAKER_01: And at the bottom, there's, I think, Daryl Onesimolu,
- 16:33SPEAKER_00: Yeah, I talked to Daron about that.
- 16:37SPEAKER_00: I don't agree with either end of that.
- 16:39SPEAKER_00: I don't either.
- 16:41SPEAKER_00: It's fun to be on the list.
- 16:42SPEAKER_00: Daron also disagreed.
- 16:44SPEAKER_00: But he and I have a lot in common.
- 16:46SPEAKER_00: I was at MIT with him for many years and we spent a lot of time talking about AI.
- 16:50SPEAKER_00: Back then, I thought we were a little closer.
- 16:51SPEAKER_00: We're definitely further apart now, but there's still a lot of common ground.
- 16:55SPEAKER_00: Well, what is the uncommon ground?
- 16:57SPEAKER_01: I mean, there must be some fundamental thing that you are seeing that
- 17:02SPEAKER_01: Jim, we've known each other a long time.
- 17:04SPEAKER_01: You have always been leaning into the technology and I've always been excited about
- 17:07SPEAKER_01: the potential.
- 17:17SPEAKER_00: Sometimes maybe a little too excited, but I saw the exponential coming.
- 17:22SPEAKER_00: I used GPT-3 back in 2020,
- 17:26SPEAKER_00: and I was showing it to my NBER colleagues that,
- 17:28SPEAKER_00: guys, there's this big wave coming, and for four or five
- 17:32SPEAKER_00: I don't think they can dismiss it anymore.
- 17:37SPEAKER_00: And since I came to the West Coast,
- 17:38SPEAKER_00: sitting here in the Bay Area,
- 17:40SPEAKER_00: every time I go to a barbecue or a dinner or a coffee,
- 17:43SPEAKER_00: somebody is showing me something amazing.
- 17:46SPEAKER_00: It's partly what they're showing
- 17:47SPEAKER_00: And it's honestly a cultural thing.
- 17:50SPEAKER_00: Economists aren't good at measuring this,
- 17:51SPEAKER_00: but there's a different vibe out here of people having these big dreams
- 17:56SPEAKER_00: and expecting big things to happen.
- 17:58SPEAKER_00: And I think some of that has rubbed off on me.
- 18:01SPEAKER_00: And I think they're right.
- 18:02SPEAKER_00: At the same time,
- 18:04SPEAKER_00: those of my friends who are still on the East Coast,
- 18:07SPEAKER_00: they're not exposed to that on a daily basis the way I am.
- 18:10SPEAKER_00: And I think it's going to diffuse.
- 18:13SPEAKER_00: And they see a lot of the worry.
- 18:14SPEAKER_00: I mean, there's this big narrative out there about,
- 18:17SPEAKER_00: we talked at the start about,
- 18:18SPEAKER_00: you know,
- 18:19SPEAKER_00: AI taking jobs,
- 18:20SPEAKER_00: not to mention,
- 18:21SPEAKER_00: you know,
- 18:21SPEAKER_00: things that are happening,
- 18:23SPEAKER_00: you know, catastrophic risks or abuses of the political system.
- 18:27SPEAKER_00: I don't know about you.
- 18:49SPEAKER_01: I think there was another one too where they had created a report about all
- 18:54SPEAKER_01: the great use cases for AI,
- 18:56SPEAKER_01: but then it turned out that a lot of those were hallucinations.
- 18:59SPEAKER_01: That is just shameful.
- 19:00SPEAKER_01: I've seen multiple emails, sarcastic emails.
- 19:06SPEAKER_00: Absolutely.
- 19:06SPEAKER_00: No, and I warn my students not to fall into that trap.
- 19:10SPEAKER_00: I guess it's tempting to be lazy and just do it.
- 19:13SPEAKER_00: I've always been kind of a glass half full kind of guy, and I know that LLMs
- 19:19SPEAKER_00: You can also do wondrous things that are incredibly valuable.
- 19:23SPEAKER_00: My advice to the folks at KPMG and to my students and everyone else is to keep
- 19:29SPEAKER_00: a human in the loop.
- 19:30SPEAKER_00: Ultimately, you, the person, is responsible for the
- 19:34SPEAKER_00: You can identify where the good things are and not the bad things.
- 19:38SPEAKER_00: I mean, one small example.
- 19:39SPEAKER_00: Nowadays, when I write a new academic paper, I have an LLM critique it.
- 19:44SPEAKER_00: I say, you are a tough reviewer at the Quarterly Journal of Economics.
- 19:47SPEAKER_00: Tell me what you like and don't like about this paper.
- 19:50SPEAKER_00: And they will come to me with,
- 19:51SPEAKER_00: you know,
- 19:52SPEAKER_00: half a dozen different suggestions and usually three or four of them are kind
- 19:57SPEAKER_00: of dumb or things I already knew.
- 19:59SPEAKER_00: But one or two of them will be something new.
- 20:00SPEAKER_00: And to me, that's a win.
- 20:02SPEAKER_00: You know, who cares if there's some that are irrelevant?
- 20:05SPEAKER_00: It's the fact that it gives me a couple of new ones.
- 20:07SPEAKER_00: I can ignore the irrelevant ones and then focus on, hey, here's something valuable.
- 20:12SPEAKER_00: And if you have that kind of an attitude of using the human yourself to figure out
- 20:17SPEAKER_00: where the real value is and leveraging that, then it's a net plus.
- 20:23SPEAKER_01: To flip it around...
- 20:25SPEAKER_01: What are the super optimists?
- 20:28SPEAKER_01: I think Elon Musk said the other day that we're going to see 10x growth in
- 20:32SPEAKER_01: the world economy over the next 10 years.
- 20:36SPEAKER_01: That may not be the most ebullient forecast you can find.
- 20:42SPEAKER_01: You can find that are more optimistic.
- 20:45SPEAKER_01: What are they missing?
- 20:47SPEAKER_00: Does that seem likely to you?
- 20:48SPEAKER_00: No, that doesn't seem likely.
- 20:50SPEAKER_00: Like I said, I've never seen such a gap between folks in these different categories.
- 20:55SPEAKER_00: These people with order of magnitude, you know, increases in productivity.
- 21:00SPEAKER_00: And then people who,
- 21:01SPEAKER_00: you know,
- 21:01SPEAKER_00: I talked to a treasury secretary who was like, hey, do you think we can get
- 21:05SPEAKER_00: A former treasury secretary who said,
- 21:07SPEAKER_00: hey, do you think we could get another few tenths of a percent?
- 21:11SPEAKER_00: So there's this amazing gap, like a couple of orders of magnitude.
- 21:15SPEAKER_00: What do the optimists not see?
- 21:17SPEAKER_00: I think they don't see how difficult it is to actually implement
- 21:21SPEAKER_00: You know,
- 21:22SPEAKER_00: most computer scientists,
- 21:23SPEAKER_00: they have this great mindset,
- 21:25SPEAKER_00: which is they simplify the problem down to its essence.
- 21:28SPEAKER_00: And then they see how computers or algorithms can address that core question.
- 21:33SPEAKER_00: And that's great.
- 21:34SPEAKER_00: And AI is really good at addressing a core question.
- 21:37SPEAKER_00: But by assumption, they ignore all the messiness around the edges.
- 21:43SPEAKER_00: But that's what really slows things down.
- 21:46SPEAKER_00: Every real world process is a series of steps and some of them end up being bottlenecks.
- 21:50SPEAKER_00: And if you only focus on the core step and you optimize that,
- 21:54SPEAKER_00: that doesn't necessarily get you high productivity if there are bottlenecks before
- 21:59SPEAKER_00: and after it.
- 22:00SPEAKER_00: Steps and you speed up eight of them,
- 22:03SPEAKER_00: that's not going to get you any additional output if the other two are just as slow
- 22:06SPEAKER_00: as they were before and they're all essential.
- 22:08SPEAKER_00: So what we're seeing right now is that when they build these amazing systems
- 22:13SPEAKER_00: for medical imaging or for coding or for whatever,
- 22:16SPEAKER_00: and they put them into the real world.
- 22:18SPEAKER_00: Getting them all to work with real humans and real customers
- 22:22SPEAKER_00: and real suppliers is just a lot harder than they imagined.
- 22:26SPEAKER_00: And that's why we have the J-curve, is figuring out all that messiness around the edges.
- 22:30SPEAKER_00: It'll happen, but it just takes longer than the typical optimist imagines.
- 22:34SPEAKER_00: And we're still in the messy part of the J-curve.
- 22:36SPEAKER_00: I guess that would be... Yeah, we're still in the messy part and we will be.
- 22:40SPEAKER_00: And I'm kind of at the geometric mean between these folks.
- 22:42SPEAKER_00: So I am more optimistic than some of my East Coast friends.
- 22:46SPEAKER_00: I don't go full AGI pill,
- 22:48SPEAKER_00: despite what the economist said,
- 22:50SPEAKER_00: about being able to get instant massive gains in productivity.
- 22:55SPEAKER_00: We have to make those transitions.
- 23:00SPEAKER_01: So I guess first...
- 23:03SPEAKER_01: To get an update,
- 23:04SPEAKER_01: what is sort of your general broad forecast about what we could expect as far
- 23:10SPEAKER_01: as productivity growth additionally from AI,
- 23:14SPEAKER_01: you know, once we start, once companies really start, you know, extracting value?
- 23:21SPEAKER_01: So first, like, what are we talking about?
- 23:23SPEAKER_00: Well, I think we'll get at least as much as we had in the internet in the 1990s,
- 23:29SPEAKER_00: you know, 3% or more.
- 23:31SPEAKER_00: And just every year,
- 23:33SPEAKER_00: the further you go out,
- 23:34SPEAKER_00: the more it's going to increase,
- 23:35SPEAKER_00: you know, by the mid 2030s, you know, it'll probably be more than 5%.
- 23:42SPEAKER_00: So I'm very optimistic.
- 23:43SPEAKER_01: 5% productivity growth.
- 23:46SPEAKER_01: Yeah.
- 23:47SPEAKER_01: How fast is too fast?
- 23:50SPEAKER_01: Is that a question you even think about?
- 23:53SPEAKER_01: Could it get so disruptive,
- 23:54SPEAKER_01: like five's great,
- 23:56SPEAKER_01: seven's better, ten's... Talk about first world problems, you know, but yeah, no.
- 24:01SPEAKER_00: Is there a number where you would get like, that's like too much disruption?
- 24:06SPEAKER_00: Jason Furman asked me that question a while back,
- 24:08SPEAKER_00: and when he first asked me, I thought that was crazy.
- 24:10SPEAKER_00: You know, of course you always want faster.
- 24:12SPEAKER_00: I mean, what's the name of your podcast?
- 24:13SPEAKER_00: Faster, Please!
- 24:16SPEAKER_00: I just I just assumed,
- 24:17SPEAKER_00: you know, but then, you know, then we talked about a little bit and he's right.
- 24:20SPEAKER_00: At some point, we just cannot absorb it.
- 24:22SPEAKER_00: And there's so much disruption and chaos that,
- 24:25SPEAKER_00: you know, I hope we had that problem that we have to have to worry about that.
- 24:30SPEAKER_00: Right now, I'm still all for pushing on the accelerator in terms of productivity growth.
- 24:36SPEAKER_00: But I do think that some of the risks are things we should genuinely worry about.
- 24:41SPEAKER_00: And so if we could have the productivity growth,
- 24:44SPEAKER_00: the gains in healthcare,
- 24:45SPEAKER_00: curing cancer, without having the risk of, you know, new kinds of viruses and stuff.
- 24:51SPEAKER_00: Cyber attacks and unknown unknowns.
- 24:54SPEAKER_00: I would love to be able to nuance it that way.
- 24:57SPEAKER_00: At some point,
- 24:58SPEAKER_00: I do think you do want to be cautious about just having a free for all
- 25:04SPEAKER_00: of acceleration without knowing where you're going.
- 25:06SPEAKER_00: A big part of the problem is that we're kind of flying blind into this.
- 25:10SPEAKER_00: We don't exactly understand what's ahead of us.
- 25:13SPEAKER_00: There's a ton of investment into the capabilities.
- 25:16SPEAKER_00: There's very little investment into measuring the economic impacts
- 25:20SPEAKER_00: or analyzing how it's all going to play out.
- 25:23SPEAKER_00: You asked me some questions about this future where AI automated a lot of work,
- 25:28SPEAKER_00: and I don't really know how we're going to manage an economic system in that kind of
- 25:33SPEAKER_00: a world.
- 25:34SPEAKER_00: We are not making those kinds of investments right now.
- 25:36SPEAKER_00: And my friends who are doing alignment research,
- 25:40SPEAKER_00: safety research,
- 25:40SPEAKER_00: security research, you know, they tell me they have not solved that problem either.
- 25:46SPEAKER_00: And while all those problems are laying out there, it is becoming increasingly risky.
- 25:51SPEAKER_00: My colleague, Chad Jones, you know, has a paper where he models how you could have faster
- 25:57SPEAKER_00: Uh, progress, but also greater risk.
- 26:00SPEAKER_00: And at some point,
- 26:01SPEAKER_00: you know,
- 26:01SPEAKER_00: turning up the numbers on both of those simultaneously ends up being
- 26:04SPEAKER_00: a bad trade off.
- 26:06SPEAKER_01: You had some, uh, AI company CEOs talk about pausing frontier AI development.
- 26:15SPEAKER_01: Uh,
- 26:16SPEAKER_01: It all seems to be happening at the same time.
- 26:18SPEAKER_01: Do you have an idea?
- 26:19SPEAKER_01: What are they seeing out there?
- 26:20SPEAKER_01: What are they worried about?
- 26:21SPEAKER_00: Do you have any idea?
- 26:23SPEAKER_00: I think we're seeing what I just described that,
- 26:25SPEAKER_00: look, there are some potential for genuinely catastrophic risks.
- 26:29SPEAKER_00: You know, this could be more powerful.
- 26:31SPEAKER_00: If we give the capability to create a lethal virus that has an incubation period
- 26:40SPEAKER_00: of 28 days and is undetectable to them,
- 26:43SPEAKER_00: that is highly contagious,
- 26:45SPEAKER_00: and if we instruct the next generation of AI to design that and figure out how
- 26:52SPEAKER_00: to get it
- 26:55SPEAKER_00: Get the DNA base pairs to do that.
- 26:57SPEAKER_00: You know, that would be terrible.
- 26:59SPEAKER_00: I would not want to distribute that to 7 billion people on Earth and say,
- 27:03SPEAKER_00: hey, you know, please don't kill us all.
- 27:06SPEAKER_00: Maybe 99.9% of them would do the right thing, but it doesn't seem like great odds.
- 27:12SPEAKER_00: That's just one kind of example.
- 27:14SPEAKER_00: The cyber risks, you know, apparently Mythos broke the NSA's
- 27:23SPEAKER_00: And there's unknown unknowns when you go into this world.
- 27:30SPEAKER_00: So we want to, you know, your earlier question, I think, was sensible.
- 27:34SPEAKER_00: Like, how fast do we want to go to this?
- 27:36SPEAKER_00: It's not infinitely fast.
- 27:37SPEAKER_00: We want to be able to digest and manage it.
- 27:40SPEAKER_00: Now, the way I would handle that is I would put more resources into space.
- 27:46SPEAKER_00: We're speeding up our ability to understand and adapt, and we're not doing enough of that.
- 27:50SPEAKER_00: And that means,
- 27:51SPEAKER_00: for instance,
- 27:52SPEAKER_00: instead of cutting the budget for economic statistics,
- 27:57SPEAKER_00: I would be massively boosting it so we get more visibility.
- 28:01SPEAKER_00: At the Stanford Digital Economy Lab,
- 28:03SPEAKER_00: we launched last week something called the AI Economic Indicators to try to fill in
- 28:07SPEAKER_00: a little bit.
- 28:08SPEAKER_00: We have data on the labor force in partnership with ADP.
- 28:12SPEAKER_00: We have data on adoption in partnership with a number of different companies.
- 28:16SPEAKER_00: We have data on what we call the transformation tracker that keeps track of
- 28:21SPEAKER_00: a dozen different metrics.
- 28:23SPEAKER_01: They are really valuable dashboards,
- 28:25SPEAKER_01: which I'm sure I'll link to,
- 28:27SPEAKER_01: because certainly one of the things I keep thinking is other than
- 28:31SPEAKER_01: How do I know things are happening?
- 28:33SPEAKER_01: Unless I get the sort of big picture lagging productivity numbers or listening
- 28:39SPEAKER_00: to conference calls from companies, how else do I know what's going on?
- 28:43SPEAKER_00: Right.
- 28:44SPEAKER_00: And I'm a big fan of the idea that we can't necessarily plan
- 28:47SPEAKER_00: and predict what's going to happen over the next five years,
- 28:50SPEAKER_00: but we can sense and respond.
- 28:52SPEAKER_00: And so we need to invest more in sensing and being able to rapidly respond.
- 28:58SPEAKER_00: I suppose I can imagine a world where
- 29:25SPEAKER_01: Robots are doing a lot of the work,
- 29:26SPEAKER_01: and people are somehow supported by the government,
- 29:29SPEAKER_01: they're getting checks, and we can imagine all kinds of scenarios.
- 29:32SPEAKER_01: But to me, it seems a little early to be worried about that.
- 29:36SPEAKER_01: And for now,
- 29:36SPEAKER_01: isn't the way most of us are going to benefit from AI going to be
- 29:42SPEAKER_01: through higher productivity growth,
- 29:44SPEAKER_01: leading to higher wages,
- 29:46SPEAKER_01: and AI creating new, solving big problems, whether it's medicine or whatever.
- 29:53SPEAKER_01: Aren't those the likely ways AI is going to help us within
- 29:56SPEAKER_01: the foreseeable future rather than we all need to have equity stakes and we all need
- 30:02SPEAKER_01: to own a share of OpenAI or Anthropic or something?
- 30:06SPEAKER_00: I share your optimism about all the value being created.
- 30:09SPEAKER_00: In fact,
- 30:10SPEAKER_00: as you know,
- 30:10SPEAKER_00: we have this measure called GDPB,
- 30:12SPEAKER_00: which measures all the free goods and the stuff that's not showing up
- 30:16SPEAKER_00: in official GDP, which is kind of growing a little bit.
- 30:20SPEAKER_00: But I think the even bigger
- 30:21SPEAKER_00: The growth is in GDPB,
- 30:23SPEAKER_00: the consumer surplus,
- 30:24SPEAKER_00: hundreds of billions of dollars coming from AI chatbots and other tools
- 30:30SPEAKER_00: that we've measured that isn't showing up elsewhere.
- 30:33SPEAKER_00: And that's going to grow very rapidly.
- 30:35SPEAKER_00: So we're getting,
- 30:36SPEAKER_00: not to mention the healthcare and other benefits,
- 30:39SPEAKER_00: curing diseases, curing cancer, I hope, hopefully.
- 30:43SPEAKER_00: So that's the good news.
- 30:46SPEAKER_00: At the same time,
- 30:47SPEAKER_00: I think there's a legitimate concern about human agency and power and whether
- 30:53SPEAKER_00: or not there's going to be increasing concentration of wealth that leads
- 30:57SPEAKER_00: to more concentration of political power.
- 30:59SPEAKER_00: We're already beginning to see a fair amount.
- 31:01SPEAKER_00: I don't think it's too early to start thinking about some of the economic implications.
- 31:05SPEAKER_00: In fact,
- 31:06SPEAKER_00: I think we need to hurry up and work harder on that side of it
- 31:10SPEAKER_00: and not assume it's all going to play out just fine.
- 31:13SPEAKER_00: It would be great to have a world of abundance,
- 31:15SPEAKER_00: and I think there's a good chance we're going to have that.
- 31:17SPEAKER_00: I'm on record as being optimistic about that.
- 31:20SPEAKER_00: But at the same time, we want to maintain our
- 31:24SPEAKER_00: Decentralized decision making and freedom and agency, and that's not inevitable.
- 31:29SPEAKER_00: These tools can be used to really centralize power.
- 31:32SPEAKER_00: Zoe Hitzig and I wrote a paper called AI's Use of Knowledge in Society in Homage
- 31:37SPEAKER_00: to Friedrich Hayek.
- 31:39SPEAKER_00: Great work.
- 31:40SPEAKER_00: And challenging some of the assumptions that were true in 1945 when he wrote
- 31:44SPEAKER_00: about it that may not be as true going forward.
- 31:47SPEAKER_00: And we need to be prepared.
- 31:48SPEAKER_00: So I think there should be a crash program to focus on the economics of
- 31:54SPEAKER_00: I'm trying to work on a lot of that here at Stanford, the Digital Economy Lab.
- 31:57SPEAKER_00: I'm meeting with philanthropists who are helping with that.
- 32:00SPEAKER_00: It would be great if the government also invested more in better data gathering.
- 32:05SPEAKER_00: As the technology rushes ahead, the gap between that and our economic understanding
- 32:09SPEAKER_00: It's going to be a pretty different economy going forward.
- 32:13SPEAKER_00: Right now the prevailing narrative is using that technology to simply automate work.
- 32:18SPEAKER_00: I think we can do better.
- 32:19SPEAKER_00: I think we can use it to enhance what people are doing and create new opportunities
- 32:24SPEAKER_00: I'm not sure that's going to happen automatically without us being very deliberative about it.
- 32:30SPEAKER_01: Who would deliver it?
- 32:31SPEAKER_01: Is that a public policy or is that just businesses deciding that we need to have
- 32:38SPEAKER_01: a kind of society where people work so we're going to focus on augmentation?
- 32:42SPEAKER_01: Is it both?
- 32:44SPEAKER_00: It's both and more.
- 32:46SPEAKER_00: I think it starts with better understanding of what the opportunities are, what the choices are.
- 32:51SPEAKER_00: When I talk to businesses,
- 32:52SPEAKER_00: I tell them that it's in their enlightened self-interest to use the technology,
- 32:56SPEAKER_00: not just to cut costs and automate, but also to create
- 32:59SPEAKER_00: New kinds of goods and services to create new ways of enhancing quality
- 33:04SPEAKER_00: and customer service to create new ways for their employees to contribute.
- 33:08SPEAKER_00: I do that in my course, the AI awakening where my students do it.
- 33:11SPEAKER_00: I have a master class where I'm hoping,
- 33:13SPEAKER_00: you know,
- 33:14SPEAKER_00: hundreds of thousands or millions of people will be able to use these tools
- 33:17SPEAKER_00: to create new kinds of value.
- 33:20SPEAKER_00: I also think there's a role for government to think about how do we invest in skills.
- 33:23SPEAKER_00: You know,
- 33:24SPEAKER_00: the United States has historically led the world in investment in education
- 33:28SPEAKER_00: and infrastructure and science and technology,
- 33:31SPEAKER_00: not so much recently, but over the previous centuries.
- 33:35SPEAKER_00: My last question, would you expect
- 34:02SPEAKER_01: Let's not take it any longer than five or seven years out,
- 34:06SPEAKER_01: that the U.S.
- 34:07SPEAKER_01: labor market will look radically different,
- 34:12SPEAKER_01: whether it's labor force participation,
- 34:14SPEAKER_01: the unemployment rate,
- 34:15SPEAKER_01: or if you look at that labor market, you'll see that work has changed.
- 34:21SPEAKER_01: You know,
- 34:21SPEAKER_01: less work,
- 34:23SPEAKER_01: fewer people participating,
- 34:25SPEAKER_01: something that begins to look like what a lot of people fear.
- 34:28SPEAKER_01: Will we see that in five or seven years?
- 34:30SPEAKER_01: I'm not asking in 20 years or beyond.
- 34:32SPEAKER_00: Yeah, I think we're going to see millions of jobs destroyed and millions of jobs created.
- 34:38SPEAKER_00: And make sure you quote both halves of that.
- 34:39SPEAKER_00: I've had people...
- 34:41SPEAKER_00: I know you will.
- 34:42SPEAKER_00: That's always happening.
- 34:44SPEAKER_00: Jobs are being destroyed.
- 34:45SPEAKER_00: It's going to happen faster and bigger.
- 34:46SPEAKER_00: This is going to happen probably 10 times faster and more disruptive.
- 34:50SPEAKER_00: We got a small taste of it with globalization and the China shock and free trade.
- 34:57SPEAKER_00: A small warm-up act compared to what we have ahead of us and navigating
- 35:01SPEAKER_00: that is going to be is going to take a lot of skill and a lot of people are going
- 35:06SPEAKER_00: to probably be hurt and a lot of people are going to benefit and making sure
- 35:10SPEAKER_00: that we get through that without having a political upheaval is not going
- 35:13SPEAKER_00: to be easy it's going to take some the right kinds
- 35:17SPEAKER_00: of investments public investments private investments is going to take
- 35:20SPEAKER_00: the right kind of attitude I'm
- 35:24SPEAKER_00: I'm excited about the opportunity of all the wealth that's going to be created
- 35:27SPEAKER_00: and all the new goods and services and especially the improvements in health.
- 35:32SPEAKER_00: That's one of the things I'm most excited about.
- 35:34SPEAKER_00: I'm also concerned that if we don't navigate it well, people are going to be rebutting.
- 35:38SPEAKER_00: To me, a lot of that backlash is based on
- 36:05SPEAKER_01: People thinking that when an AI CEO says,
- 36:08SPEAKER_01: you know, within 18 months, all the behind screen jobs would be gone.
- 36:13SPEAKER_00: That's not what you're, you don't think that's likely.
- 36:16SPEAKER_00: Part of it is we got to change the conversation.
- 36:18SPEAKER_00: We got to change the reality.
- 36:19SPEAKER_00: And we have to lean into the fact that a lot of those jobs will disappear.
- 36:23SPEAKER_00: I don't want to sugarcoat that.
- 36:25SPEAKER_00: There's going to be a lot of new jobs created.
- 36:27SPEAKER_00: You got to tell both parts of that story.
- 36:28SPEAKER_00: And of course, you want to lean into the
- 36:30SPEAKER_00: I'm not sure there's a lot of risk of that.
- 36:57SPEAKER_00: No, no, no, no, but it has to be, you know, people are going to question it.
- 37:00SPEAKER_00: And so you need to have the goods when you say that's going to happen.
- 37:05SPEAKER_00: And I believe it's totally possible.
- 37:07SPEAKER_00: I'm investing,
- 37:08SPEAKER_00: you know, my course, both at Stanford and the masterclass is all about doing that.
- 37:12SPEAKER_00: My company, Work Helix, is all about doing that.
- 37:15SPEAKER_00: So I'm doing what I can for my part.
- 37:17SPEAKER_00: I would love to see more people lean into that part of the story.
- 37:21SPEAKER_00: It's something that I think there's an opportunity there for a smart politician,
- 37:25SPEAKER_00: a smart political leader, smart business leaders to lean into that.
- 37:29SPEAKER_00: Well, my fear is that in the next elections,
- 37:32SPEAKER_01: they're going to be leaning the other way, but I hope.