Well, Casey, a lot is happening in AI this week. Specifically, I feel like it's been a very big OpenAI news week.
Yeah, and I feel like it's actually been a minute since we checked in on them, and so I was glad that there was like such a series of events that I felt like, okay, yeah, we can actually return to this company now.
So, before we do this, of course, we should make our AI disclosures. I work for The New York Times, which is suing OpenAI along with Microsoft and Perplexity, and my fiancée works at Anthropic. So, let's just get a few things ticked off here. The first thing was a personnel shift. Fidji Simo, the former CEO of applications, I believe was her title at OpenAI.
That was her initial title. It was later changed to CEO of AGI deployment.
Okay. So, she is leaving the company. She was on an extended medical leave due to some health problems she had had. She is now officially sort of stepping back from that role, and her roles will be divided up among other executives. Casey, what did you make of this?
I mean, very sad story. Obviously, anytime somebody has to leave a job for like reasons other than they wanted to, it is sad. Simo was there for less than a year and I think was able to accomplish some of what she wanted and did play a role in pushing the company to streamline some of its offerings and cut back on some of the projects that, in my view, were distractions, such as Sora, you know, AI video generator. But I think she had much bigger ambitions there that she didn't get to realize. And you know, in terms of OpenAI going forward, it is yet more executive turmoil at a company that has had a lot of that over the past year.
And speaking of turmoil, we should also talk about this lawsuit that Apple filed last week against OpenAI, alleging that they stole some trade secrets. Casey, tell us the story of this lawsuit.
This came out on Friday. Apple filed a lawsuit against OpenAI and IO, which was the sort of Jony Ive co-founded company that they bought for more than $6 billion. The suit also names two current OpenAI employees, Changlu and Tang Tan, who incidentally are former Apple employees who now work at OpenAI. They write in this filing, quote, 'At every level, from members of its technical staff to its chief hardware officer' — that's Tang Tan — 'and in coordination with business partners, OpenAI has been stealing Apple's trade secrets and confidential information. As a natural result, OpenAI's nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misappropriated trade secrets.' So that was the kind of, you know, quote that you saw in a lot of stories about this particular case, but the details are really pretty wild.
Yeah, let's talk about some of those details. It was a long filing. There was a lot of juicy stuff in there. The thing that I think got the most attention was this allegation by Apple that Tang Tan, this OpenAI chief hardware officer, had directed job candidates who are applying for jobs at OpenAI to bring physical hardware components, blueprints, and prototypes into job interviews for what the filing calls 'show-and-tell' sessions. And like, you think it's hard enough to get a job in this economy, and then your recruiter says, 'Hey, when you come in, make sure to bring two forms of ID and then your resume and also an unreleased Apple product that you sort of snuck out of the lab.' Like, what? That's such a big ask. How are you supposed to deal, to get a job in this climate?
I mean, my goodness. So then there's also these allegations about Changlu, the former senior system electrical engineer at Apple. Apple says in its filing that he kept a work laptop after leaving the company. He then apparently exploited an unknown security vulnerability to access confidential information and bragged about this to a colleague. I'll just read his apparent quote from the filing: 'LOL, I found out I can access the network storage. So funny.'
You know, and I just, I'm glad you read that in full because here's what I want people to know. If you say LOL in a message, you should not add 'so funny,' right? Because those two phrases are doing the same thing.
Yes. Omit needless words.
So, there's a lot more allegations in the filing. One of the things it says is that OpenAI allegedly approached one of Apple's manufacturing partners and tricked them into showing them some sort of secret metal treatment finishing process that Apple uses.
Yeah. If you've ever picked up an iPhone, you know that it has a very beautiful metal finish. And the allegation here is that OpenAI wanted to know how that happened. And so they just went to the supplier and they were like, 'Just show us what you do for Apple. It's cool. They told us it's cool.'
Yeah. Tim Cook said, 'We have permission to do this.'
Yeah. Tim Cook said, 'I would love for Sam Altman to know how to do this. Just go ahead and show them.' And then they show — these are the allegations. Yeah.
But Casey, what did you make of this filing as you looked through all of the juicy details?
So, I'm of two minds here because it is true that Apple in general treats everything that has ever happened in the city of Cupertino as an Apple trade secret, right? They defend their intellectual property extremely vigorously. And so I would not be surprised if, as this case unfolds, you know, a judge or a jury might push back on some of the claims here and say, you know what, it's actually not illegal for Tang Tan to mention the code name of an unreleased Apple product to a prospective candidate. Right. Okay. That said, with some of the particular allegations against the LOL-so-funny person, the allegation that they were like coaching job candidates to bring unreleased products into job interviews, that does just seem transparently wrong to me. And so my real question as this case unfolds now, Kevin, is basically how high up the chain did this go? You know, some bloggers have noted that the case goes out of its way not to mention Jony Ive, who's like a foundational figure at this company. If there was actually some sort of, you know, scheme to do this, it seems extremely unlikely that Jony Ive would be unaware of it. And yet, he's not in the filing. Neither is Sam Altman. Right? So, these are some of the questions that I have. But I do think it, you know, if Apple is successful here, it could have some, you know, very tough consequences for OpenAI. What would you make of it?
Yeah, I mean, I think these guys — knowing nothing else about this case other than the filing, I'm sure there will be, you know, counter-filings and explanations from OpenAI about why this all wasn't as bad as it looks on paper — but like, to me, these guys seem guilty as hell. That said, it is not clear that OpenAI is actually building anything that makes use of Apple's trade secrets. We got a report this week from Mark Gurman over at Bloomberg, who is sort of the closest and most plugged-in Apple reporter, who reported on OpenAI's first hardware product, the thing that Jony Ive has been working on with them, being some sort of a mobile, screen-free home smart speaker. That doesn't sound to me like anything that Apple has built or maybe is planning on building.
Well, I mean, they have that HomePod, right? You know, like what Gurman described, as I read it, was basically a next-generation like Amazon Echo, which many companies now have their own versions of, including Apple with the HomePod. And so a screen-free speaker that connects to AI, like, Apple does have one of those, and I would not be surprised if they said that whatever OpenAI is building somehow builds off of their efforts here.
Yeah, I mean, I am not particularly bullish on the OpenAI hardware push in general. But I think this made me even more skeptical that they are going to come out with some category-breaking, you know, definitive technology here. I could be wrong, of course, but you know, asking employees, job candidates, to bring in prototypes from their Apple secret work does not seem to me like a company that has a real vision for what it is doing on its own. You know what I thought was funny about this was that a little over a year ago at the first Hard Fork Live, I asked Sam Altman about their hardware project. And I said, 'Is it just Alexa?' And he said, 'It's not just Alexa. Like, you know, give us some time. Like, we're thinking we're going to do something better than Alexa.' And I read the Gurman report this week and I thought, 'Sounds a lot like Alexa.' So, you know, we'll see. I mean, the AI in it is going to be better, you know, but is it really more than that? Anyways, yeah.
I mean, I think this is very much in keeping with Apple's sort of historical posture for secrecy. You know, you remember when the iPhone — I think it was the iPhone 4 — was sort of leaked, there were law enforcement investigations. Like, this company takes its trade secrets very seriously. At the same time, Apple is also struggling in AI. I think it's fair to say that. And so this to me feels very much like the kind of lawsuit that someone who is not in the lead or close to the lead would file.
Yeah, I think that that is reasonable. You know, in addition to the, you know, alleged inspection of unreleased devices over the past couple of years, OpenAI has hired more than 400 people away from Apple. So even at a big company like Apple, that is the kind of recruiting effort that you notice. And I'm sure that has been extremely frustrating to the folks over at Apple as they try to build their own next-generation devices. So I think that there was clearly a wound there, and I think it is fair to read this lawsuit in part as Apple lashing back.
Yeah. I mean, this is a bigger conversation than we have time for today. But I'm always just fascinated by how leaky and sort of porous the boundaries between these AI companies are. I mean, one of the sort of hidden factors in this AI race is, I think, the fact that in California, you can't really enforce a non-compete. So, people are always moving jobs from OpenAI to Anthropic, from Google DeepMind to OpenAI. Like, people are just bouncing around all of the time and they're taking their secrets with them, right? Like, you can't take intellectual property. You can't like literally take the code for your model, but all this stuff like exists in their head. And so part of the reason that you would hire someone from Apple if you're OpenAI is like they have this stuff in their head and it's just a matter of like, oh, they can't literally bring the designs or the prototype or the IP over, that's illegal, but just knowing how to build something is not a protected class of IP.
Right. And I'm sure that OpenAI will make that argument as this case moves forward. Kevin, what has OpenAI said in the meantime about these allegations? So, a spokesperson for OpenAI posted on X, quote, 'We have no interest in other companies' trade secrets. We remain focused on building innovative technology that empowers people everywhere.' How did that strike you?
That struck me as what we call a non-denial denial. It struck me as we're just learning about this and we need to put out a statement, but we can't say anything because we don't know anything. So, what is a series of words that journalists would reprint that would sort of buy us a few more days? That's kind of how that struck me.
There's an art form to this stuff.
Okay. Another thing that is happening at OpenAI this week is that people are starting to play around with Soul, this new model GPT-5.6, that was not available for a period of time under request from the White House and the federal government, but that people are now using and experimenting with. I have not spent much time with Soul so far, but people I know who I trust on this stuff have said it's really good. It's a very capable model.
Yeah, it absolutely is. It is a notable step up from 5.5. I've been doing some vibe coding projects recently. I've had the GPT-5.6 Soul kind of critique them, suggest changes, and I've been really pleased with the feedback. So, I think if you are a ChatGPT user and you're on one of the paid plans, you're going to be really excited that you have GPT Soul.
Is this a thing you do to like, you know, get back at your fiancée when you're having arguments at home? You just like casually stroll in and you're like, 'Hey, been doing some really cool vibe coding with GPT-5.6.'
Yes. I mean, I think he understands that, you know, it's in the nature of my job to always be testing all of the models.
Yeah. You're not a monogamist when it comes to AI models.
No, I'm extremely poly, and I think he appreciates that.
So, a couple things that have come out of this new model release. One is that it has really changed the plans of Anthropic. As we talked about on the show, they had announced that, you know, starting on some arbitrary day in early July, they were going to revoke access to Fable for people using their Claude subscription plans. This is their most powerful model. Basically, you'd still be able to use it through the API, but you're going to have to pay for it on a per-token basis. You're not just going to like get some Fable access included in your bundle of a subscription. And they then extended that deadline by a few days to give people a few more days to play with Fable. Then they have extended it again. And my read on this is like this is in direct response to the fact that OpenAI has a, by all accounts, very good model out right now, and they are afraid that if they cut off access to Fable, people are going to switch to using ChatGPT or Codex.
Absolutely. So, I mean, you think about what their original plan was, which is, you know, we'll put out Fable, and this happened, you know, several weeks before 5.6 eventually came out. And I assume that their thinking was people will have a few weeks to see how good Fable is. And so then whenever some other company launches a new model, it won't matter because people are already going to be addicted to Fable. Instead, they had a big fight with the US government. And by the time they finally got Fable out the door, there were essentially only a few days between that and 5.6 arriving. And so my belief now is that they are trying to essentially get people hooked on Fable and eventually push them toward paying per token. But you know, 5.6 is out now and people are going to be able to maybe try that for a lot cheaper. So,
Right. And the companies, we should say, have been getting a little catty.
A little. I've seen episodes of Drag Race that were more polite than what's been happening on Twitter over the past few days. Kevin.
Yeah. So, let's just talk about that a little bit because I think it's interesting, but also I think potentially very important. So, Anthropic posted a new ad for Claude that is sort of a longer ad. It's kind of a sort of montage. It's talking about hard questions. It starts with this very bizarre sequence of images of one is like a building on fire, and one is with a cemetery, like a row of headstones. It looks to be like a military cemetery.
It's Arlington National Cemetery, where my grandfather is buried. Kevin, what is it doing in this ad?
And so the sort of voiceover on the ad is talking about how, how can it, you know, who's going to, who's going to hit the brakes if we need to slow down this technology? Can AI be trusted? Like, these are the kinds of hard questions. You know, the ad eventually gets to a more hopeful and positive message, but it starts with this kind of like very doomy picture of the world on fire, literally as a result of this technology. And Sam Altman did not like this ad. Many other people did not like this ad.
I mean, I'll say I think this ad is a disaster. Like, I really do. Like, it's an extremely confusing set of messages, and I think it's like, it is unlikely to me that the average person would take away a positive message about Anthropic or AI from this ad.
Oh, I don't know. I want to sort of like meekly defend the ad by saying that I think it's good if AI companies are honest about the fears that they have. I think it would be very bad if they just all turned into like sunny, optimistic marketing companies.
Well, sure, but is the right format for that like brand marketing, you know? Like, I understand you think this might go really, really badly and you're trying to prepare the world for that, but I think like a serious video with your leaders talking about that or like a thoughtful blog post are a better avenue for that than like a video posted on X of a burning building. You know what I mean?
I know. I take that point. I just like, I don't want to have the message that comes out of this be like AI companies should just lie to us, and if they're scared about AI, they should just not tell us that, and they should just put out ads about how great this stuff is.
Well, here to me is — I'm not going to call it a lie, but here's what I find confusing. So, at the end of this ad, which just essentially I believe has like sound clips of average people asking their questions about AI or discussing their wishes for AI. At the very end of it, Anthropic shows its tagline, which is 'Keep Thinking.' And when Anthropic tells me 'Keep Thinking,' I feel like I'm reading one of those blog posts where they tell senior citizens to keep doing crossword puzzles to keep their mind active. You know what I mean? It's like if you're building superintelligence that is going to do all of the thinking, what exactly do you want me to do the thinking for? Do you know what I mean? So, I just think the entire brand message is getting very, very confused, and it is making people feel bad, and I think Anthropic needs to think about that.
Yeah, I think that's right. But let's get back to the OpenAI of it because that was actually where this started to heat up. So, Sam Altman posted on Monday. He reposts the ad and he says, 'I thought this was satire. Kept looking for the handle to be spelled C1-0ud AI,' like basically a sort of scam account or something. And then he sort of mocks the 'hard questions' tagline of the ad. He says, 'Hard questions are great, but only if we deem you worthy enough not to silently downgrade you or even get access at all.' Of course, referring to Anthropic's on-again, off-again Fable rollout and some decisions they had made about sort of like silently downgrading people to a lesser model if it detected they were working on certain types of things.
Yeah. So here is my impression of what is happening: is the folks at OpenAI understand that they are now seen as an underdog, and they're thrilled about it. Like, think about the past year or two that they have had where they were sort of carrying the banner of American AI, and every misstep that they made was under a microscope. And now fast forward to today, Anthropic is seen as being in the lead. And so it's now all of Anthropic's mistakes that are sort of getting magnified, and that lets OpenAI come in and throw some blows and look like a scrappy challenger again. And it's just much more fun to play the scrappy challenger than the sort of, you know, solemn standard bearer of American AI.
Yeah, I buy that. I don't think they're thrilled about being in second place, at least as the public perceives it. I think there's a lot of paranoia and fear over there that they are, you know, losing talented researchers to Anthropic, that they are sort of losing what they call the mandate of heaven to Anthropic by not having the absolute best models on the market.
Wait, what is the mandate of heaven?
Oh my god, you haven't seen this meme?
So, this is a thing that like insane people on social media post about where they're like, 'Oh, you know, the new Gemini is really good. Maybe, maybe Google has the mandate of heaven.' It's just sort of like...
Like sort of like God has chosen you to invent AGI. Yes. Okay. Wonderful.
It's very dumb. Anyone who says the mandate of heaven in sincerity should be checked into a mental hospital. But anyway, I think there's a lot of paranoia. They really like — these companies, like, I don't know if the public understands this, but these companies hate each other. This is more than a Coke and Pepsi thing. Like, what's funny is I was researching this for my book and I learned that technically these companies may still have a non-disparagement agreement with each other. So when Dario and the Anthropic folks were leaving OpenAI in late 2020, they sort of signed this non-disparagement agreement that says like, we won't badmouth you and you won't badmouth us, and like we'll all sort of keep things, you know, placid on the surface. And that has never been rescinded, to my knowledge. And so technically they still have a document on the books that says we're not supposed to trash each other in public.
From the outside, reading sort of how they're talking about each other, I would assume that they had a disparagement agreement.
An only-disparagement agreement.
The minute we leave here, it's on, bro. It's on like Donkey Kong.
Yes. So these companies absolutely hate each other. They are bickering over the sort of terms of these models being accessible to subscribers. I think they are going to start sort of doing some kind of price war with each other. That's my expectation, is like, you know, Anthropic will kind of reduce the price of tokens and then OpenAI will reduce it below that, and it'll sort of be, I think, good for consumers in a very sort of narrow, you know, consumer-interest perspective, because competition among frontier AI labs is good, I would argue.
Yeah, it's definitely good for consumers. I think that there are elements of the, you know, AI race itself that are scary, but perhaps that's a different conversation.
No, I think that's where I want to go with this because I think like this feud between OpenAI and Anthropic, which does not seem to be subsiding and in fact seems to be getting worse, I think it is very dangerous. I do not like this way that these labs are conducting their relations with each other. I do not like how much the people at OpenAI hate Anthropic and are motivated by wanting to crush them. I do not like how much the people at Anthropic hate OpenAI and want to crush them. I think that we are heading into a period where there may be real and dangerous risks associated with these very powerful systems, and these companies should be open to collaboration and solving this problem together. And coordination between labs, I think, is going to become increasingly important, and you can't do that if you're constantly fighting each other.
I completely agree. You know, Demis, who runs DeepMind, put out a piece this week where he talked about an idea for a kind of government regulator that would, you know, create a process around when a frontier lab could release a new model, maybe solve some of the issues that we've seen in recent months. And one of the things that he talks about is it would be great if the labs at some point could successfully collaborate around a slowdown in the development of AI. And now maybe that's something that the government will just, you know, decree. But in the absence of that, you want to see exactly what you're talking about, Kevin, which is like the leaders of these companies being able to pick up the phone and call each other and think about like what is good for the rest of the world and not just what is going to get a bunch of retweets.
Yeah. I just desperately need Dario and Sam and Demis to like all get together, take a little MDMA, work it out, find some compassion, find some common ground, you know, put the beef aside. Yeah, we're all in this together. Sometime, some — what I found is sometimes business leaders will forget that we are all individual facets of the same universal consciousness, and that's when they get themselves into trouble.
All right, Kevin, let's turn our attention to the question of AI and jobs. This is something that I have been fascinated by myself over the past few months. I've been speaking to several economists and tech leaders about the short, medium, and long-term risk of an AI getting so good that all of a sudden employers are going to feel like they don't need people anymore, like they can just spend tokens. And there's frankly still a lot of skepticism on this point. And so that is why I was fascinated to see a new letter come out this week that had been signed by a group of nearly 200 economists and researchers, and they were warning about the impact that AI could have on the economy.
Yeah, I saw this too. This was a statement. It was very short, just three short points. It was titled 'We Must Act Now.' And basically, I mean, we can probably just read the whole thing.
Read the whole thing. It's not that long. And it says, point one: AI may become radically more powerful over the next 10 years. Point two: this could drive an unprecedented transformation of our economy, larger than the Industrial Revolution, but unfolding over a vastly shorter time frame. It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards. Point three: economists, policymakers, and technology leaders must act now to understand the economics of transformative AI and to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.
So the statement itself, I think, is pretty anodyne. It's basically like we should take AI really seriously because it could do all these things. It could do some good things, it could do some bad things, and we should pay attention to it. I think what was more interesting was the list of signatories, including some former Hard Fork guests. Anton Korinek from the University of Virginia and Anthropic was on there, as well as I think some more skeptical economists who have previously been kind of dismissive of the idea that AI could cause large-scale job transformation, at least in the short term.
Yes, at least in the short term. It includes several Nobel laureates, including Daron Acemoglu, sort of a well-known skeptic of the short-term AI job-loss hypothesis, David Autor at MIT, lots of other just sort of like well-known economists and thinkers who are coming together to make this statement, and I thought this was really interesting.
Yeah. And so this week we wanted to talk to a person who helped to put this statement together. That person is Erik Brynjolfsson. He is a Stanford economist who helped to organize the statement, and he is somebody who takes these threats really, really seriously.
Yeah, I've known Erik for years. He's a great economist, and he's been writing about the danger that AI poses to jobs and the economy for a very long time, like longer than most economists, longer than ChatGPT has existed. He's been writing about this stuff. He's a senior fellow at the Stanford Institute for Human-Centered AI and the director of the Stanford Digital Economy Lab. He's been studying AI's impacts on human workers and firms for a very long time. So to meet the demand for the questions that we have for him, we now supply Erik Brynjolfsson.
Erik Brynjolfsson, welcome to Hard Fork. Well, it's a pleasure to be talking to you again, Kevin. So you have been writing and talking and thinking about the transformative potential of AI on the economy for a very long time, longer than almost any economist I know. And I'm curious what inspired you to organize this letter, this 'We Must Act Now' statement from a bunch of your peers and colleagues.
This is something that's been brewing for a long time. You're right. I've been thinking about it a long time. I taught my first course about artificial intelligence back in 1987, before most of your listeners were probably born. And I've just been very convinced about what was happening. But you know, the nature of exponential curves is they grow, you know, slowly at first and then suddenly. And now, especially this year, we're in that suddenly part of the curve. The technology is going so fast, but our economic understanding is nowhere close to keeping up. And so I'm doing what we can to try to change that.
So the letter is titled 'We Must Act Now.' I would say it doesn't really specify an action. So tell us about what your goals are with this letter and what you think the policymakers who read it and all the sort of esteemed signatories, what should they do?
Well, there's two main purposes. First is that there's been a sea change, and we're hoping to catalyze that sea change in attitudes about how seriously we take AI. As recently as a year or two ago, and certainly before that, most of the economists I talked to really were not very AI-pilled. They thought, hey, it's not a big deal. I remember one secretary of treasury saying to me, oh, do you think that maybe, you know, there'll be a couple of tenths of a percent of change in productivity over the coming years. So it was sort of like imagining sort of a very small technological impact. And I personally think it's going to be huge, and I wanted more people to take it seriously and realize that I think this is the most important thing economists should be working on. Whatever other projects they're working on, they should look at it through a lens of AI. Part of what we're calling for is a better understanding, more economists thinking hard about this question, more research going into it. That said, there was the beginning of a direction that, you know, Daron Acemoglu and I and others have been talking about for a while, which is this idea of complementing humans rather than substituting for them. Now, fleshing out exactly how to do that is a bigger question, but at least if we could get, you know, a couple hundred economists and a lot of Nobel Prize winners and other AI researchers around the world to agree on that direction, that's a good start.
Let me hit you with some of the objections that I've heard when I've talked to other tech leaders or economists about the idea that AI might take a lot of jobs. I myself am totally prepared to believe that there could be a real disruption here, but I do hear some skepticism. So, one thing I hear is that job losses almost never have any one cause, and it's basically impossible to pinpoint when you're looking at the data.
Well, it's certainly true. There's lots of things going on in the economy. AI is not the only thing. It's one of the most important, and increasingly the most important thing. You know, we wrote this paper, 'Canaries in the Coal Mine,' that you may be familiar with. We're going to have a new version of it out in a few days, probably in time for your airing of this podcast, where we look at some of the alternatives as well. We look at what's going on with interest rates. We look at remote work. We look at the tech over-hiring. We look at changes in education. All of those are important. And we find that all of them make a difference. But even when you put them in individually or all at the same time into these regressions to try to parse out what's happening, AI continues to be an important factor. And it's not just destroying jobs. AI is also creating jobs, as a lot of people will tell you. And it's the job of us, people like me, like economists, to try to sort that out and see which parts are important as early as possible and as fine-grained as possible. Just because it's not easy doesn't mean we shouldn't try to sort it out.
Yeah. I mean, I confess that I have been wrong about the economic expectations I had for what would happen when AI systems got more powerful. I think if you had time-traveled back to 2020 and showed me like Claude Fable or GPT-5.6, and then you asked me like, this technology will exist in 2026, what is the headline unemployment rate? I probably would have said like 10 or 20%. And what we see instead is that employment rates today are basically the same as they were a few years ago. Headline unemployment is hovering around 4%. So are you as surprised as I am that this technology becoming so powerful has not created more unemployment?
No, not at all. So I've written many papers on this idea of the productivity J-curve and on the way that AI complements changes. You may have heard — I'll tell, I'll retell the story about how when electricity was first introduced, it took about 20 to 30 years before it fundamentally changed the way work was in American factories. At first, there was little or no productivity change. There was little or no employment change because people were just kind of paving the cow paths. They were redoing the same things they used to do, but now with electric motors, and nothing fundamentally changed. It took a lot of reinvention of those factories and ultimately the whole economy to unleash a big change. And that reinvention meant eliminating millions of jobs. It meant creating millions of jobs. All of that, in the case of electricity, like I said, took 20 to 30 years. I don't expect it to take 20 to 30 years this time around, but it's not the kind of thing that happens in 20 to 30 months either. It's something that plays out over a period of years. I think we're in the early stages of that transformation. I have no doubt there's going to be a lot of disruption and a lot of jobs destroyed. There's also going to be a lot of jobs created. We're just in the earliest stages of that right now, but it's going to get bigger.
You mentioned the Stanford Canaries dashboard, which I've been following with great interest. It showed that early-career jobs shrank 2.7% year-over-year, while mid-career jobs grew 1.6% year-over-year. Is there growing consensus among economists that this is AI-driven impact, and how should we respond to those numbers?
I can't speak for the economics...
Profession as a whole. I mean, I believe the numbers. I've spent a lot of time looking at them. I think over time it'll become more and more obvious. Maybe one of the most convincing things, in addition, I mentioned we tested all these other hypotheses and they didn't knock out the AI result. But the other thing, just to, you know, we first published a version of this back in August of 2025 and we saw, you know, basically the same pattern, and people said, 'Oh, it's a blip. It's due to, you know, interest rates or whatever.' It's just grown over time. You can go to the canaries, you can go to the dashboard, AI economic indicators at Stanford, and what you'll see is that these trends have just been continuing to grow over time. It's not something that's just due to COVID or something else.
I want to ask one more question, Eric, about this letter, this statement. Obviously the coalition that signed the letter, like a hugely impressive group of economists and thinkers, tech executives, a real who's who of the field. But there were some people who didn't like this letter and didn't sign it. Either they thought it was too vague or it didn't make enough concrete policy recommendations. But I wanted to single out one reaction from Noah Smith, who's an economics writer and blogger. He wrote that he didn't sign this statement because, among other reasons, he was just very skeptical of this line about trying to steer AI in a direction that complements humans and benefits society. He basically wrote like, this is not possible. We don't know that we can steer this technology toward being complementary. He writes, 'How the heck are business people and inventors supposed to steer AI toward being complementary to human workers? They obviously couldn't predict the labor market effects of the last round of AI, at least in the short term. So why should anyone believe the technologists have the ability to purposefully invent different forms of AI with different labor market effects?' What's your response to that?
Well, I love Noah. He's super smart. He's a friend of mine. We hang out in the Bay Area from time to time. I'm so glad he engaged with the letter and I'm also glad that not everybody just reflexively agreed with it. I think we had just the right amount of FUD there, a few people disagreeing just to make it interesting. It would have been boring if everybody just said, 'Yeah, yeah, this is obvious.' That said, Noah is wrong about this. You know, we have lots of examples in history of technology being directed in different ways. One of the ideas behind carbon taxes is that, you know, you want to do that, and there's research that shows that, you know, countries that impose more taxes, engine technology evolved in different ways. For computers specifically, there's long been two strains. You know, Alan Turing talked about the Turing test, imitating, replacing humans. There's another stream, Doug Engelbart and others, who talked about AI amplifying people. And one of the things I do, I spend a lot of time talking to CEOs. Let me tell you a very concrete story. I met with a CFO and she told me that she wanted to use AI and get higher ROI for it. And I said, 'That's great, that's what you should be doing.' And then she said, 'And therefore, what we're doing is requiring measuring in every department how much headcount reduction they have thanks to AI.' And I said, 'Whoa, hold it just a minute there. That's one way to measure AI is replacing workers. And you know, and sometimes that's a good thing. But there's so many other things you can be doing. You're a CFO. You're being paid millions of dollars.' I didn't actually say that to her, but that was what I was thinking. 'You should be more creative. Like, let's look at the upside. What kind of new products and services? What kind of better customer service? What kind of reduced employee turnover?' There are all these other metrics that a good manager can look at that will increase the upside. And if they think that way, if they change their mentality, believe me, they're going to steer the technology, you're going to have outcomes that involve people more, that complement people more, rather than just replace them.
Yeah, I'm just a little more cynical about that than maybe you are. I want to believe that you're right and that these enlightened corporate leaders will say, 'Well, you know, we really don't want the technology to replace humans because humans are so valuable and there's all these ways that they can be, you know, complemented by the technology.'
Right. But like, well, let me just say this, you know, there's two reasons for doing this, there's three reasons for doing this. Okay. One is that I do think it's the right thing for society, but let's just set that one aside for now. The second reason is that I think it's in their enlightened self-interest. I could teach at business schools corporate strategy, the better strategy, the more defensible strategy is creating new products and services and barriers to entry, not just cutting costs. And then the third thing I'll say is, it's true there are a lot of tax and other incentives right now that skew companies disproportionately towards investing in capital versus labor. For instance, marginal tax rates on capital lower than marginal tax rates on labor. And so if you have a strategy that replaces a lot of workers with machines, you end up paying lower taxes. That's a flaw, I think, in the tax system. That's something we can fix and create better incentives for work for managers to treat them more evenly. And then last but not least, I do think there are a lot of people who will embrace this because it is the right thing for the world, and I think their workforce is going to like working at that company if it's a company that does right by them.
You told Ben Castleman at the New York Times that economists need more reliable data. What are some of the problems with the data that we have right now and what would you like to see more of?
I mean, this is something I'm super focused on right now. Sadly, as we're going into one of the most turbulent periods in all of economic history, we are destroying our instrument panel. The government is cutting back on data collection, eviscerating some of the statistical agencies. I feel for those people as they're trying to do it. This is the time, if anything, we should be doubling down and investing more in visibility. It's a really high ROI activity. It just costs a tiny bit of the budget and it'll help us make better decisions not only as policymakers but also as managers. That said, you know, there's only so much I can do about the federal budget. What I can do is at the Digital Economy Lab create the AI economic indicators, partner with private organizations. We're incredibly lucky that there's a bunch of real-time data that's being just collected by companies anyways. You know, Ramp, Rellio, ADP, and so we can tap into these data sources and get some kinds of visibility. They're not always collected for the purpose of being representative. So you have to do some work to kind of normalize them and measure them. But by creating an infrastructure to get more visibility, we can do more. And we're creating a new taxonomy of the kinds of tasks that are being done in the economy. There's something called O*NET that lists all the tasks that workers do in the American economy. And we're creating an updated version of that. I think, you know, over the coming months and years, we'll be able to get much more real-time visibility in a much more fine-grained way of what's happening in the economy, and that will allow us to sense and respond to what's going on. I mean, you can only do so much about predicting what's going to happen next, but it's sometimes easier to at least measure it quickly and then allow people to respond to that.
