“We Are Being Gaslit By The AI Companies!” – Karen Hao on DOAC Podcast (Transcript)

And back then, as is true today, we have no scientific consensus around what human intelligence is. There’s no definition from psychology, biology, or neurology. And in fact, every attempt in history to quantify and rank human intelligence has been driven by nefarious motives — driven by a desire to prove scientifically that certain groups of people are inferior to other groups of people.

There are no goalposts for this field, and there are no goalposts for the industry. When they say that they are ultimately trying to recreate AI systems that would be as smart as humans, how do we even define what that means? And when are we going to get there if we don’t know how to define the destination?

What that effectively means is that these companies can just use the term Artificial General Intelligence — which is now the term to refer to this ambitious goal to recreate human intelligence — however they want to, and they can define and redefine it based on what is convenient for them.

How OpenAI Defines AGI Differently for Different Audiences

KAREN HAO: In OpenAI’s history, it has defined and redefined AGI many times. When Sam Altman is talking with Congress, AGI is a system that’s going to cure cancer, solve climate change, cure poverty. When he’s talking with consumers that he’s trying to sell his products to, it’s the most amazing digital assistant that you’re ever going to have. When he was talking with Microsoft — in the deal that OpenAI and Microsoft struck, where Microsoft invested in the company — it was defined as a system that will generate $100 billion of revenue. And on OpenAI’s own website, they define it as “highly autonomous systems that outperform humans in most economically valuable work.”

This is not a coherent vision of one technology. These are very different definitions that are spoken out loud to the audience that needs to be mobilized — to ward off regulation, or get more consumer buy-in into the industry’s quest, or to get more capital and more resources for continuing on this journey with ambiguous definitions.

Sam Altman’s Early Warnings About AI

STEVEN BARTLETT: Speaking about different definitions through time — in 2015, in a blog post that Sam Altman wrote before OpenAI was officially announced, he explicitly outlined the existential risk by saying, “Development of superhuman machine intelligence is probably the greatest threat to the continued existence of humanity. There are other threats that I think are more certain to happen — for example, an engineered virus. But AI is probably the most likely way to destroy everything.”

KAREN HAO: In general, when Altman is writing for the public or speaking for the public, he does not just have the public as the audience in mind. There are other people that he is trying to motivate or mobilize when he says these things. And in that particular moment, Altman was trying to convince Elon Musk to join him in co-founding OpenAI.

Musk in particular was spending all of his time sounding the alarm on what he saw as a huge existential threat that AI could pose. And so in that blog post, if you look at the language that Altman uses side by side with the language that Musk was using at the time, it mirrors all the things that Musk—

STEVEN BARTLETT: —was saying 10 years ago. Musk was going on podcasts saying, tweeting, whatever, that the greatest existential risk to humanity was AI.

KAREN HAO: Yeah. And so his parenthetical — “There are other things that might actually be more likely to happen, like engineered viruses” — it’s because up until then, Altman had been talking just about engineered viruses. And so now that he needs to pivot to speak to an audience of one, to Musk, he needs to kind of resolve the contradiction between what he’s now elevating as his new central fear — to be the same as Musk’s new central fear — with what he had previously been saying. So that’s why he’s like, “I think this is now,” even though before he said something different.

Did Sam Altman Manipulate Elon Musk?

STEVEN BARTLETT: And are you saying that Sam Altman manipulated Musk? Because Elon did end up donating a huge amount of money to OpenAI and co-founding it, I believe, with Sam Altman.

KAREN HAO: Elon Musk did end up co-founding it with Altman. And certainly from Musk’s perspective, he does feel manipulated, because he feels like Altman was engineering his language in a way that would make Musk trust him as a partner in this endeavor.

And of course, then Musk leaves. Through some of the documents that came out during the lawsuit that Musk and Altman are engaged in now, it has become clear that there was a degree to which Musk was actually muscled out a little bit. And so that’s why he’s left with this very intense personal vendetta against Altman, saying that somehow Altman tricked him into being part of this.

The Power Struggle at OpenAI’s Founding

STEVEN BARTLETT: So in 2015, Sam Altman is writing these blog posts saying this is one of the greatest existential threats. At the same time, in 2015, Musk is doing some very famous speeches — at MIT, he said that AI was the biggest existential threat and compared developing AI to summoning the demon. And what you’re saying here is that Sam Altman was just mirroring the language that Elon was using to get Elon involved in OpenAI. And later it appears — and again, there’s a legal case taking place now — that Sam might have muscled Elon out in some capacity.

KAREN HAO: Yeah. So we know from the lawsuit and the documents that have come out in the lawsuit that Ilya Sutskever — who was the Chief Scientist of OpenAI at the time — and Greg Brockman — Chief Technology Officer at the time — when they were deciding whether or not to maintain OpenAI as a nonprofit, because it was originally founded as a nonprofit, they decided they needed to create a for-profit entity.

But the question was, who should be the CEO of this for-profit entity? Should it be Musk or should it be Altman? Because they were the two co-chairmen of the nonprofit. And in the emails, it became clear that Ilya and Greg first chose Musk to be the CEO.

But through my reporting, I discovered that Altman then appealed personally to Greg Brockman — who was a friend of his, someone he had known for many years through the Silicon Valley scene — and said, “Don’t you think that it would be a little bit dangerous to have Musk be the CEO of this company, this new for-profit entity? Because he’s a famous guy. He has a lot of pressures in the world. He could be threatened, he could act erratically, he could be unpredictable. And do we really want a technology that could be super powerful in the future to end up in the hands of this man?”

And that convinced Greg, and Greg then convinced Ilya. “I think there’s a point here. Do we really want to give this much power to Musk?” And that is why Musk then leaves — because the two switch their allegiances. They say, “Actually, we want Altman to be the CEO.” And then Musk is like, “If I’m not CEO, I’m out.”

STEVEN BARTLETT: So it sounds like Sam again managed to persuade someone to do something.

KAREN HAO: Mm.

What Do People Really Think of Sam Altman?

STEVEN BARTLETT: I guess this begs the question — what do you think of Sam Altman?

KAREN HAO: I think he’s a very controversial figure.

STEVEN BARTLETT: You did an interesting pause. It’s a pause where someone tries to select their words.

KAREN HAO: Well, this is what’s so interesting about those interviews — people are extremely polarized on Altman. No one has in-between feelings about him. Either they think he’s the greatest tech leader of this generation, akin to the Steve Jobs of the modern era, or they think that he’s really manipulative, an abuser, and a liar.

And what I realized, because I interviewed so many people, is it really comes down to what that person’s vision of the future is and what their goals are. So if you align with Altman’s vision of the future, you’re going to think he’s the greatest asset ever to have on your side. Because this man is really persuasive. He’s incredible at telling stories. He’s incredible at mobilizing capital, at recruiting talent, at getting all the inputs that you need to then make that future happen.

But if you don’t agree with his vision of the future, then you begin to feel like you’re being manipulated by him to support his vision, even if you fundamentally don’t agree with it. And this is the story especially of Dario Amodei, CEO of Anthropic, who was originally an executive at OpenAI.

STEVEN BARTLETT: So for people that don’t know, Dario now runs Anthropic, which is the maker of Claude. A lot of people are probably more familiar with Claude.

The Polarization Around Sam Altman

KAREN HAO: Yeah. And it’s one of the biggest competitors to OpenAI and Anthropic. At the time, when he was an executive at OpenAI, he thought that Altman was on the same page with him. And then over time began to feel that Altman was actually on exactly the opposite page of him and felt that Altman had used Anthropic’s intelligence capabilities, skills to build things and bring about a vision of the future that he actually fundamentally didn’t agree with. And so that’s why people end up with this bad taste in their mouths.

I’ve been covering the tech industry for over eight years and covered many companies. I’ve covered Meta, Google, Microsoft, in addition to OpenAI and Altman. And it’s the only figure that I’ve seen this degree of polarization with, where people cannot decide whether he’s the greatest or the worst.

Dario Amodei and Ilya Sutskever’s Departures from OpenAI

STEVEN BARTLETT: You mentioned Dario there. And what I found really interesting is to look at how people’s quotes evolve over time with their incentives. So I was looking at all of the things they’ve said on the record, on podcasts, in their blog posts, to see how it’s evolved over time.

And Dario, who was the former VP of Research at OpenAI and has now moved on to Anthropic, who are taking a slightly different approach to developing AI, said back in 2017, while he was still at OpenAI, that — and this is a quote — “I think at the extreme end is the Nick Bostrom style of fear that an AGI could destroy humanity. I can’t see any reason in principle why that couldn’t happen. My chance that something goes really quite catastrophically wrong on the scale of human civilization might be somewhere between 10% and 25%.”

And also you mentioned Ilya, who was a co-founder of OpenAI and then left. I guess the first question I’d ask is, why did Ilya leave?

KAREN HAO: That’s a great question. So he was instrumental in trying to get Sam Altman fired. And he’s another one of the people who over time began to feel like he was being manipulated by Altman towards contributing something that he didn’t believe in.

STEVEN BARTLETT: How’d you know?

KAREN HAO: Because I interviewed a lot of people. Ilya in particular had two pillars that he cared about deeply. One is making sure we get to so-called AGI, and the other is making sure that we get to it safely. And he felt that Altman was actively undermining both things. He felt that Altman was creating a very chaotic environment within the company where he was pitting teams against each other, where he was telling different things to different people.

STEVEN BARTLETT: Have you ever spoken to him?

KAREN HAO: I have. So I interviewed him in 2019 for a profile that I did of OpenAI for MIT Technology Review.

STEVEN BARTLETT: And back in 2019, he has a quote where he says, “The future’s going to be good for AIs regardless. It would be nice if it was also good for humans as well. It’s not that it’s going to actively hate humans or want to harm them, but it’s just going to be so powerful. And I think a good analogy would be the way that humans treat animals. It’s not that we hate animals. I think humans love animals and I have a lot of affection for them. But when the time comes to build a highway between two cities, we are not asking the animals for permission, we just do it because it’s important to us. And I think by default, that’s the kind of relationship that’s going to be between us and AI, which are truly autonomous and operating on their own behalf.” And that was in 2019, the year that you interviewed him.

What Is Intelligence? The Hypothesis Behind AI Development

KAREN HAO: One of the things that I feel like we should take a step back to examine is going back to this idea of what even is artificial intelligence, and what do we mean by intelligence. And a huge part of the views of the different people and the quotes that you’re reading derives from a specific belief that they each have in this question of what is intelligence, what constitutes intelligence?

For Ilya, he has, throughout his research career, felt that ultimately our brains are giant statistical models. This is not something that we actually know. But this is his own hypothesis, also the hypothesis of his mentor, Geoffrey Hinton, who also was on this podcast. This is why they have such a strong conviction in the idea of building AI systems that are statistical models, and that this particular approach is going to lead to intelligent systems as we are intelligent. It’s a hypothesis that they have. It’s not one that has been proven by science. And some people vehemently disagree with them on this particular thing.

But if you step into their shoes and take on that hypothesis and assume that it’s true — that our brains are, in fact, statistical engines, and that these systems that they’re building are also statistical engines that they’re making bigger and bigger and bigger until they become the size of the human brain — that’s why they say that making this comparison, where the system will become equal to human intelligence and then maybe exceed human intelligence, is relevant in their framework.

And Ilya gave a talk at one point at this really prominent AI research conference that happens every year called Neural Information Processing Systems. It’s a mouthful, but he gave this keynote where he shows this chart of the size of brains and the intelligence of a species, and it’s roughly linear. The bigger the size of the brain, the more intelligent the species. And so for him, he thinks he’s building a digital brain because he thinks brains are just statistical engines. So from that logic, it’s like, okay, if we then build a bigger statistical engine than the human brain, then based on this chart, it will be more intelligent, and then we will be subjected to the same treatment that we’ve subjected animals.

But it’s really important to understand that these are scientific hypotheses of specific individuals within the AI research community. And there’s a lot of debate about whether this is in fact the case. And some of the biggest critics say it’s very reductive to think of our brains as simply just statistical engines.

STEVEN BARTLETT: Why does it matter to know the mechanism? Is it not just important to know the outcome, which is that it’s going to be able to make a video for me, or agents are going to be able to do the work that I do? Does it really matter for us to know the mechanism behind it?

Why Are We Building AGI? A Critical Question

KAREN HAO: Yes and no. So it matters because these companies are driving their future actions based on this hypothesis. So they have decided that this hypothesis is true, and that they should just continue building larger and larger statistical models in the pursuit of artificial general intelligence. And that’s then having global consequences. In order to continue doing that, they’re hoovering up more and more data, they’re building more and more data centers, and they are exploiting more and more labor in order to continue on this path.

Here’s a question that I think is important to ask: why are we trying to build AI systems that are duplicative of humans? We’re kind of having this conversation right now where we’ve just taken the premise of this industry as a good thing. Like, they said that we should be building AGI, so we say that we should be building AGI. But I would like to ask, why are we doing that? Why is it that we are building a technology that is ultimately designed to replace and automate people away? That is not the enterprise of technology. We should be building technology, and the purpose of technology throughout history has been to improve human flourishing, not to replace people.

And so this is a critical part of my critique of these companies and these scientists that have just adopted this goal and have relentlessly pursued it and have had enormous capital and enormous resources to pursue it. Is this the right goal? Why are we doing this? Why can’t we just build AI systems that do things like accelerate drug discovery and improve people’s healthcare outcomes, which are systems that have nothing to do with the statistical engines that they’re trying to build to duplicate the human brain?

STEVEN BARTLETT: So why are they doing it? I mean, you’ve interviewed all these people. I think it’s what, 300 people in total, 80 or 90 of them from OpenAI, the maker of ChatGPT. Why do you think they’re doing it?

KAREN HAO: I think it’s because they’re driven by an imperial agenda. And that is why I call these companies empires of AI.

STEVEN BARTLETT: What do you mean by an imperial agenda? What does that term mean?

The Empires of AI: An Imperial Agenda

KAREN HAO: Empire is the only metaphor that I’ve ever found to fully encapsulate all of the dimensions of what these companies do and the scale that they operate and what motivates them to do what they do. And there are many parallels that you see between what I call the empires of AI and the empires of old.

They lay claim to resources that are not their own in the pursuit of training these models — that’s the data of individuals, the intellectual property of artists, writers and creators. They’re land grabbing in order to build these supercomputer facilities for training the next generation models.

Second, they exploit an extraordinary amount of labor. They contract hundreds of thousands of workers all around the world, including in the US, to ultimately make these technologies. We can talk about that more. And they also design their tools to be labor automating, so that when the technologies are deployed, it also affects labor rights because it erodes away labor rights. And this is a political choice that they have.

Third, they monopolize knowledge production. So they project this idea that they’re the only ones that really understand how the technology works. And so if the public doesn’t like it, it’s because they don’t actually know enough about this technology. They do this to the public, they do this to policymakers. And they’ve also captured the majority of the scientists that are working on understanding the limitations and capabilities of AI.

STEVEN BARTLETT: You think they’re gaslighting the public?

KAREN HAO: In a way they are, yeah. So if most of the climate scientists in the world were bankrolled by fossil fuel companies, do you think we would get an accurate picture of the climate crisis?

STEVEN BARTLETT: No.

KAREN HAO: And in the same way, the AI industry employs and bankrolls most of the AI researchers in the world. So they set the agenda on AI research in soft ways simply by funneling money to their priorities so that only certain types of AI research are produced. But they also will censor researchers when they do not like what the researcher has found.

And so I talk about the case of Dr. Timnit Gebru in my book, who was the Ethical AI Team Co-lead at Google when she was literally hired to critique the types of AI systems that Google was building. She then co-wrote a critical research paper that was showing how large language models specifically were leading to certain types of harmful outcomes. And in an attempt to try and stop this research from being published, Google ended up firing Gebru and then fired her other co-lead, Margaret Mitchell. And so they control and quash the research that is inconvenient to the Empire’s agenda.

OpenAI’s Campaign of Intimidation Against Critics

STEVEN BARTLETT: Did you have an example where this is happening to journalists as well, that are asking questions of their team members? I think I was watching a video of yours where there was a young man that was saying he had someone show up at his door, knocked on his door and asked for information, emails, text messages. And this person was from one of the big AI companies.

KAREN HAO: This was — OpenAI started subpoenaing some of its critics. Yeah, as part of what appears to be a campaign of intimidation, but also what appeared to be a campaign of fishing for more information to figure out, to map out the network of critics further. But this was a man who runs a small watchdog nonprofit, and they had been doing a lot of work during that time to try and ask questions about OpenAI’s attempt to convert from a nonprofit to a for-profit.

Ultimately, OpenAI was successful in that conversion. But during the period where it was sort of existential for OpenAI to complete this conversion, there were a lot of civil society groups and watchdog groups like Midas who were trying to prevent the process from happening in the dead of night. They were trying to get more transparency. They were trying to have more public debate about this, because it’s unprecedented. And it was then that there was a knock on his door, and he was served papers.

STEVEN BARTLETT: What did the papers say?

KAREN HAO: The papers asked him to reproduce every single piece of communication that he had had that might have involved Musk. So this was like this strange paranoia that OpenAI had that Musk was somehow funding these people to block the conversion. None of them were actually funded by Musk. So in this particular case, the request he simply just answered, “I don’t have any documents because this doesn’t exist.”

The Three Pillars of Empire: Land, Labor, and Knowledge

STEVEN BARTLETT: So going back to this point of empires, you were saying that one of the factors of an empire is land grab.

KAREN HAO: And then the next one was labor exploitation.

STEVEN BARTLETT: Labor exploitation.

The Empire’s Narrative and Control of Information

KAREN HAO: The third one — controlling knowledge production. And one of the other ones that’s really important to understand about the AI empires in particular is empires always have this narrative that they say to the public, like, “We’re the good empire, and we need to be an empire in the first place, because there are also bad empires in the world. And if you allow us to take all the resources and use all of the labor, then we promise we will bring you progress and modernity for everyone. We will bring you to this utopic state akin to an AI Heaven. But if the evil empire does it first, we will descend into a hell.”

STEVEN BARTLETT: And the evil empire being in this —

KAREN HAO: — case, in this case, most often it’s China. But actually, in the early days, OpenAI evoked Google as the evil empire. So all of their decisions were about, “We need to do it first, because otherwise Google — this evil corporation that’s driven by profit — us as a benevolent nonprofit, this is a critical contest of who wins.”

STEVEN BARTLETT: Do you think the people building these AI companies believe that the outcome is going to be all good now? Do you think they think that it’s going to serve everyone, it’s going to be the age of abundance, everything’s going to go well? What do you think they believe? What do you think Sam believes?

KAREN HAO: This is so funny. It’s such a core part of the mythology that they create around the AI industry — it includes the belief that it could go very badly. It goes hand in hand. They need that part of the myth in order to then say, “And that’s why we need to be in control of the technology, because that’s the only way that it’s going to go really, really well.”

And Altman has said publicly, the worst case is lights out for everyone. But best case, we cure cancer, we solve climate change, and there’s abundance. And Dario Amodei — same kind of rhetoric — worst case, catastrophic or existential harm for humanity. Best case, mass human flourishing.

So this is like two sides of the same coin. They have to use both of these narratives in order to continue justifying an extremely anti-democratic approach to AI development, where there should not be broad participation in developing this technology. They must be the ones controlling it at every step of the way.

Sam Altman’s Tweet and Karen’s Book

STEVEN BARTLETT: Sam Altman did a tweet saying, “There are some books coming out about OpenAI and me. We only participated in two of them. One by Keach Hagey, focused on me, and one by Ashlee Vance on OpenAI.” He went on to say, “No book will get everything right, especially when some people are so intent on twisting things. But these two authors are trying to.” You quote-retweeted that tweet from Sam Altman and you said, “The unnamed book, Empire of AI, is mine.” Do you believe that tweet from Sam Altman was in reference to your book?

KAREN HAO: 100%, because there’s only three books coming out about him.

STEVEN BARTLETT: And he’d caught wind that your book was coming out, and he knew —

KAREN HAO: — my book was coming out because I had contacted OpenAI from the very beginning of my process and said, “I’m working on a book now. Will you participate in it?” And actually, initially they said yes.

So my history with OpenAI — I profiled the company for MIT Technology Review. I embedded within the office for three days in 2019. My profile comes out in 2020. The leadership were very unhappy. And in my book, I actually quote an email that I received — that Sam Altman sent to the company about my profile — saying, “Yeah, this is not great.” And from then on, the company’s stance to me was, “We are not going to participate in anything that you do. We are not going to respond to any of the questions that you send.” And this was things that they explicitly articulated. It wasn’t like me inferring.

I had a colleague at MIT Technology Review that also covered AI, and at one point, OpenAI sent him a press release, being like, “We would love for you to cover this story.” And he was like, “I’m really busy. Will you send it to Karen?” And they were like, “Oh, no, we have a history, you understand.” So for three years, they refused to talk to me.

But then I ended up at the Wall Street Journal, where they felt a bit compelled — because it was the Journal — to reopen the lines of communication. And so I started having more dialogue with them. Every time I wrote a piece, I would always send them a request for comment. I would always ask them, “Will you sit for interviews?” And we did get to a more productive relationship.

Then I embarked on the book. I left the Journal to focus on the book full time, and I told them right away, “I’m working on this book. I want to continue this productive conversation where I make sure I reflect OpenAI’s perspective in the book.” And so they were like, “We can arrange interviews for you. You can come back to the office. We’ll set up some conversations.”

And then, as we were going back and forth on this, the board fires Sam Altman. And that’s when things started going kind of south, because the company started becoming very sensitive to scrutiny. They started kicking the can down the road, down the road, down the road. And I kept saying, “Hey, when are we rescheduling this? What’s going on?” And then I get an email saying, “We are not going to participate at all. You are not coming to the office. You’re not doing interviews.” I had actually already booked my tickets, so I was already going to fly to San Francisco to have the interviews.

So then I told them, “That’s fine. I will still engage in the process where I’ll give you extensive requests for comment. I’ll ask through my reporting. I’ll keep you updated on all the things that I’m finding so that you can choose to still comment.” I gave them 40 pages of requests for comment, and I gave them over a month to respond to all of that. So this was when the tweet came out — we were doing all this back and forth. And that’s when Altman tweeted this. And they never responded to a single one of the 40 pages.

Access, Control, and the Media Game

STEVEN BARTLETT: Sam Altman does a lot of interviews. He’s doing a lot of interviews all the time. He’s done every podcast I’ve seen him on, everything from Tucker Carlson to Theo Von to Joe Rogan — podcasts all over the world. I wonder why he won’t do mine.

KAREN HAO: Well, maybe —

STEVEN BARTLETT: I don’t know why. I think I’m fair with everyone. I just ask questions I genuinely care about. I don’t come in with huge preconceptions, or at least I meet people for the first time with an open mind. But I’ve heard through the grapevine that he doesn’t want to do mine.

KAREN HAO: Going back to what you were saying earlier — that with the way that OpenAI and these companies control research, you asked, “Do they also do this with journalists?” The answer is yes. And apparently they also do it with anyone who has a broad mass communications platform. It’s not just about the conversation that you’re going to have with them. It’s about who you also choose to platform.

And there’s this huge problem in technology journalism where companies know that a really big carrot that they can give to technology journalists is access.

STEVEN BARTLETT: Yeah, yeah, yeah.

KAREN HAO: And they will withhold that access at the drop of a hat if they catch wind that you’re speaking to someone that they didn’t want you to speak to.

STEVEN BARTLETT: This is so true. And I don’t think the average person really truly understands this. This kind of sounds like theory as you say it, but I’m not going to name names here, because I don’t think it’s important. But there is a particular person in AI whose team have basically dangled the carrot of them coming here for like 18 months. And I’m like, you don’t have to dangle the carrot. I’m going to speak to whoever I want to, regardless of the carrot or not. And when this person comes, if they want to come, I’ll give them a fair shot. I’ll ask them all genuinely curious questions about what they’re doing, their incentives. I won’t gotcha them. I don’t have a history of ever gotcha-ing anybody, even if I have a different opinion.

But they dangle carrots and they say, “Well, he’s thinking about it. Let’s think about a date and what the strategy is.” And I don’t think they understand — if we just dangle it for long enough, then they will perform in the way that we want them to, and they’ll be pleasant about us. They won’t be critical. They won’t give a f*. They won’t platform —

KAREN HAO: — our critics.

STEVEN BARTLETT: Our critics. And I think a lot of their game is just dangle the carrot forever.

KAREN HAO: Yes. Yeah.

STEVEN BARTLETT: That’s like the optimal outcome — if we just dangle it, if we just tell them, “Yeah, we’re just looking at the schedule.” I think, in the modern world, you just have to go there and give your opinion and allow the clash of ideas in the public forum. Let the viewers decide for themselves what they think.

KAREN HAO: This is such a huge part of their machinery — the way that they use these tactics to massage the public image of these companies and make sure that information they don’t want out, and even opinions they don’t want out there, don’t go out there.

And I feel very lucky now that OpenAI shut the door early on me. At the time, I didn’t feel lucky. I felt like I had screwed myself over. I was like, “Should I have been nicer to them in the profile so that I could maintain access?” Which is a horrible question to ask as a journalist. You’re supposed to report the truth, and you’re always supposed to report in the interest of the public. That is the point of journalism. And in that moment, I was relatively junior in my career. I was like, “Did I misunderstand what journalism is about? Should I have actually been playing the access game?”

But it was too late. I had the door shut to me. And so I had to build my career understanding that the front door was never going to be open.

STEVEN BARTLETT: Yeah.

KAREN HAO: And that actually really strengthened my own ability to just tell it like it is — to be objective and just report what I see are the facts being presented to me, irrespective of whether the company likes it or not. And most often, the company really does not like it. But I can continue to do the work. They don’t need to open the front door for me. I was still able to do more than 300 interviews.

The Firing of Sam Altman

STEVEN BARTLETT: So Sam Altman gets kicked off the OpenAI executive team. Did you find out why that happened?

KAREN HAO: Yeah. There’s a scene-by-scene recounting from — I can’t remember the exact number of sources, so I don’t want to misquote myself, but it was around six or seven people that were directly involved, or had spoken to people directly involved in the decision-making process.

So Ilya Sutskever is seeing serious concerns about the way that Altman’s behavior is leading to bad research outcomes and poor decision making at the company. He then approaches a board member, Helen Toner.

STEVEN BARTLETT: Ilya, for anyone that doesn’t know, is the co-founder — the co-founder of OpenAI we mentioned earlier?

KAREN HAO: Yes. And he kind of does a bit of a sounding board thing with Helen, just because Ilya’s freaking out. He’s been sitting on these concerns for a while and he’s like, “If I tell this to someone, this could also be really bad for me if Altman finds out.” And so he asks for a meeting with Toner. And in that first meeting, he barely says a thing. He’s just dancing around trying to figure out, “Hey, is this someone that I can maybe trust to divulge more information?”

STEVEN BARTLETT: And Toner’s role and responsibilities at OpenAI were —

KAREN HAO: She was a board member.

STEVEN BARTLETT: Just a board member.

The Board’s Decision to Fire Sam Altman

KAREN HAO: Yeah. And specifically an independent board member. So OpenAI, when it was a nonprofit, the board was split between people who had a stake, financial stake in the company, and then people who were fully independent. And this was meant to be a structure that would balance the decision making, to be in the benefit of the public interest rather than to be in the benefit of the for profit entity that OpenAI then created.

And Ilya as a non independent board member, was approaching Toner as an independent board member to try and see whether or not she was potentially seeing or hearing the same things that he was about the effect that Altman was having on the company. This then sets off a series of conversations, first between Ilya and Helen and then between Mira Muradi and some of the board members.

Mira Muradi was at that point the chief technology officer of OpenAI, where these two senior leaders, essentially, through these conversations and through documentation that they’re pulling together, like email, Slack messages and so forth, they convey to the independent board members, three independent board members. “We are very concerned about Altman’s leadership, like he is creating too much instability at the company. And it is like he is the root of the problem.”

They were trying to say to these independent board members like, the problem will not be fixed unless Altman is removed because of the way that he’s pitting teams against each other and creating this environment where people are unable to trust each other anymore and they’re competing rather than collaborating on what’s supposed to be this really, really important technology.

STEVEN BARTLETT: When you say instability, that’s quite a vague term that could mean lots of things. Like instability could mean pushing people hard to work harder. What do you mean by instability in as specific terms as you can possibly say them?

The Chaos Inside OpenAI After ChatGPT’s Launch

KAREN HAO: When ChatGPT came out in the world, OpenAI was wholly unprepared. They didn’t think that they were launching a Gangbusters product. They thought they were releasing a research preview that would help them get the data flywheel going, collect a bunch of data from users, and that would then inform what they thought would be the Gangbusters product, which was a chatbot using GPT-4. And ChatGPT was using GPT-3.5.

And because of that there were servers crashing all the time because they had to scale their infrastructure faster than any company in history. And there were all of these outages. They were trying to also hire faster than any company in history to try and have more personnel there. And they were then sometimes hiring people that they were like, “Actually we made a mistake, we shouldn’t have hired you.” So they were firing people left and right and people were just disappearing off of Slack and that’s how their colleagues would learn that they were no longer at the company.

And so it was, yes, like many fast growing companies, a very chaotic environment and a particularly chaotic environment because it was extra fast. Like they had to accelerate more than any other startup. And on top of that, Mira Muradi and Ilya Sutskever felt that Altman was making it worse. Like he was not actually effectively ameliorating the circumstances of the chaos. He was actually sowing more chaos, getting these teams to be more divided.

And this is where it’s important to understand that the executives and the independent board members, they’re all operating under this idea that they’re building AGI and that AGI could either be devastating or utopic to humanity. And so it’s not — yes, it’s like any other company and no, it’s not like any other company. You cannot have, like, in their view, you cannot have this degree of chaos as the pressure cooker for creating a technology that they, in their conception, could make or break the world.

And so that is basically what the independent board members also begin to reflect on. They have these conversations amongst themselves where they’re like, “Well, based on what we’re hearing about Altman’s behavior, like if this was an Instacart, would that warrant firing him?” And they concluded maybe not. But this is not Instacart. And that’s why they were like, “Well, crap, maybe this does actually rise to the bar where we should consider replacing him, because we are ultimately building a technology that we think could have transformative impacts, either in the positive or negative direction.”

And so that is what happens. It’s like these two executives and then the independent board members also, they were hearing other feedback as well, from their connections within the company with other people in the industry. At one point, Adam D’Angelo, who is one of the independent board members and the CEO of Quora, which is a tech startup in the Valley, he is at a party in San Francisco and he starts to hear some of these rumors that there’s something weird about the way that OpenAI has structured its OpenAI startup fund, which was this fund that the company had created to start investing in other startups.

And he realizes they’d never really seen documentation about how the startup fund had been set up from Altman. And finally they get the documents, and it turns out that OpenAI Startup Fund is not OpenAI’s startup fund, it’s Altman’s startup fund. And this was something like one of several experiences that the independent board members were also having where they’re like, “There’s something not right about the fact that there continuously are inconsistencies between the way that Altman is portraying what is being done versus what is actually being done.”

And so when these two executives approach the board or the independent board members, then they’re like, “Okay, this lines up with also the experiences that we’ve been having.” And at that point, they then have this series of very intense discussions where they’re meeting almost every day, talking about, should we actually really consider removing Altman? And in the end, they conclude, yes, we should. And if we’re going to do it, we need to do it quickly. Because they were very concerned that the moment that Altman found out his persuasive abilities would make it impossible to do.

And so they end up firing Altman without telling anyone. They don’t talk to any stakeholders to get them on the same page. Microsoft gets a call right before they execute the action saying, “We’re going to —”

STEVEN BARTLETT: — fire Altman. And Microsoft, for anyone that doesn’t know, are a lead investor in OpenAI at the time.

KAREN HAO: Yes, one of the only investors in OpenAI at the time. And that is what then devolves the whole thing, because every single person that is affected by this decision is now extremely angry that they were not involved. And that is what then creates this campaign to bring Altman back. And then Altman is reinstalled as CEO.

The Co-Founder’s Verdict on Altman

STEVEN BARTLETT: How does a CEO of a major company get fired by the board? Because board members — there’s a quote in your book on page 357 where you say about Ilya saying, “I don’t think Sam is the guy who should have the finger on the button for AGI.”

Now I asked myself this question. I work with lots of people here. We have 150 people that work in this business. And those people know me best. They see me on camera, they see me off camera. So if they said that we don’t think Stephen is the right person to host the Diary of a CEO, for example, it would take a lot for them to say that. They must have seen some shit off camera for them to go, “We don’t think he’s the right person to be on camera,” or for whatever reason.

And in the case of AI, which is much more consequential than a podcast that is filmed in my old kitchen, it almost sends a chill down one’s body to think that the co-founder of a business has gone to the board and said, “This isn’t the guy to lead this consequence.”

KAREN HAO: And it wasn’t just Ilya. Mira Muradi then also said, “I don’t think Altman is the right guy.”

STEVEN BARTLETT: And then they both left later.

KAREN HAO: So then Altman comes back and lo and behold, Ilya never comes back. So his concerns about the fact that Altman finding out would be bad for him manifested. He ended up not coming back and Mira Muradi then left shortly thereafter.

STEVEN BARTLETT: Quite a lot of these people leave, don’t they? OpenAI?

The Exodus from OpenAI

KAREN HAO: They do. So if you consider one of the origin stories of OpenAI is this dinner that happened at the Rosewood Hotel, which is a very swanky hotel right in the heart of Silicon Valley that was one of Elon Musk’s favorites whenever he was coming up from LA to the Bay Area. And there was this dinner that was there where Altman was intending to recruit the OG team that would start OpenAI. So he’s kind of telling everyone, “You might have a chance to meet Musk because Musk is going to come to this dinner.” And he cold emails Ilya and gets Ilya to then come. And Ilya specifically wants to come because he wants to meet Musk. And he also emails all these other people, including Greg Brockman, Dario Amodei —

STEVEN BARTLETT: These are all people that ended up working at OpenAI.

KAREN HAO: And they all, almost all of them — not every one of them, but almost all of them — end up working at OpenAI and leaving. Almost all of them end up leaving specifically after they clash with Altman.

STEVEN BARTLETT: Ilya, he left and launched a company called Safe Superintelligence. I mean, that’s an indirect if I’ve ever heard one. Do you know what I mean? If someone co-founded this podcast with me and then they left and started a podcast called Safe Podcasting, I’d take that as a slight. I’d have people knocking on their door and asking for their texts.

KAREN HAO: One of the things that is happening here is it is not a coincidence that every single tech billionaire has their own AI company. They want to create AI in their own image and that’s why they keep not getting along. And in fact, it’s not just don’t get along. They end up hating each other after working together and then splinter off into their own organizations.

So after Musk leaves, he starts xAI. After Dario leaves, he starts Anthropic. After Ilya leaves, he starts Safe Superintelligence. After Mira leaves, she starts Thinking Machines Lab. They want to have control over their own vision of this technology. And the best way that they have derived from their experiences of trying to put their vision into the arena is by creating a competitor and then competing with OpenAI and with all the other companies out there.

The Psychology of AI Myth-Making

STEVEN BARTLETT: Do you think some of these AI CEOs realize that they are quite literally summoning the demon, as Elon said 10 years ago, but they don’t really care because being the person that summoned the demon makes you consequential and powerful and historical. Even if the outcome is potentially horrific, even if there’s like a 20% outcome of it being horrific.

I remember, I think it was Dario. He’s the one that said there’s somewhere between a 10% and 25% chance of things going catastrophically wrong. On the scale of human civilization, 25% is a one in four chance. If you put bullets in a four chamber revolver and said, “Stephen, the upside is you could become a multi-gazillionaire and be remembered forever. The downside is that there would be a bullet in your head.” There is no chance that I would take that bet with a 25% potential chance of things going catastrophically wrong.

KAREN HAO: So I have a very long answer to this, because do they know if they’re summoning the demon? It really depends on what we define as summoning the demon. And in this particular case, to go back to what we were saying before, there’s a mythology that the AI industry uses where summoning the demon is an integral part of convincing everyone that therefore they can be the only ones that are developing this technology.

STEVEN BARTLETT: Ah, I got it. So on one end, you’ve got to say, “If we don’t, China will,” and that’s terrible.

KAREN HAO: Yeah.

STEVEN BARTLETT: But if we let anyone else do it other than me, then we’re f*ed as well.

KAREN HAO: Exactly.

STEVEN BARTLETT: So that means that I have to do it and you have to give me money and support.

KAREN HAO: Exactly. So when they’re saying these things, we should understand it not as a genuine prediction based on what they’re seeing, because, first of all, we don’t predict the future, we make it. We should understand this as an act of speech to persuade other people into believing that they should cede more power, more resources to these individuals.

And so do they know that they’re summoning the demon? I mean, they are purposely trying to create this feeling within the public that they are, because it is a crucial part of their power. But do they — if we were to define just, do they realize that the things that they are doing are having already really harmful impacts all around the world on vulnerable people, vulnerable communities, vulnerable countries? That’s where I’m like, maybe yes, maybe no. And they don’t really care, because in the frame of mind — I sometimes use the analogy that the AI world is like Dune.

The Dune Analogy: Myth-Making and Power

STEVEN BARTLETT: Dune, for anyone that doesn’t know Dune —

KAREN HAO: — science fiction epic written by Frank Herbert. And it’s set in this intergalactic era where there are all these houses and they’re fighting each other for spice. So it’s a callback to colonialism and Empire. And they all are trying to control the spice.

But one of the features of this story is that there are these myths that are seeded on the different planets — a religious myth, basically about the coming of the Messiah — that are used as ways to control the people. And Paul Atreides, when he arrives at the planet Arrakis with intention of trying to then fight against the Empire and avenge his father’s death, he steps into a myth that has been seeded on this planet that says that one day there will be a Messiah that comes and saves the planet. So he steps into the role of the Messiah and leans into this idea in order to better control the people and rally them behind him as a leader to help with this quest.

He knows that it’s a myth in the beginning, but because he lives and breathes and embodies it, it kind of starts to blur in his mind whether this is really a myth or whether he’s really the Messiah. And this is what I think happens in the AI world.

On one hand, there are all these executives that actively engage in myth-making because — I have all these internal documents that I write about in the book where they are very keenly aware of how to bring the public along with them by showing them dazzling demonstrations of the technology, by crafting a mission that will sound really good and make people give more leniency to their companies. So they know they’re doing the myth-making.

And also, I think many of them lose themselves in the myth because they have to live and breathe and embody it day in and day out. And so when Dario says he thinks that 10 to 25% of the future could be catastrophic, or whatever the probability is, he is actively engaging in the myth-making, but also he’s losing himself in the myth. I think if you were to ask him, he would be like, “Yes, I genuinely believe that,” because there’s been a blurring of when he’s saying something just to say something, versus when he actually believes what he’s required to believe in order to then continue doing the things that he’s doing.

Cognitive Dissonance and Fundraising

STEVEN BARTLETT: And this is the whole psychology of cognitive dissonance, right, where the brain struggles to hold two conflicting worldviews at the same time. So it’s incentivized, or it endeavors, to dismiss one. So if you wanted to be a healthy person, but also a smoker, and I pointed out that smoking’s bad for you, the first words out of your mouth are going to be, “Yes, but smoking helps me with stress.”

KAREN HAO: Yeah.

STEVEN BARTLETT: “Yes, but I only do it when —” And I think I kind of see that at the moment. Because these companies have to raise extortionate, huge amounts of money to fund their AI research and their building out of all of these data centers. So when they’re out in the public, they’re always fundraising. All of these major companies are fundraising all the time at the moment. So you can’t be fundraising and saying, “I’m going to destroy your children’s future. Potentially, there’s a 25% chance that your children aren’t going to have a great life,” which might be the truth.

KAREN HAO: I mean, that is actually what they say. This is what famously Dario Amodei does.

STEVEN BARTLETT: He does that. But the others — Sam’s not doing that as much anymore.

KAREN HAO: Yes. And it’s because each of them kind of distinguish themselves a little bit as the brand that they need to project.

STEVEN BARTLETT: Do you think any of them have a stronger moral compass than others? Because I think Dario often gets the credit for having more of a backbone and being more conscious of implications.

KAREN HAO: He does get a lot of credit for that.

STEVEN BARTLETT: He’s from Claude and Anthropic.

KAREN HAO: For anyone that doesn’t know. I don’t think it truly matters, the answer to that question. Because to me, even if you were to swap all the CEOs for someone that people would say is better at running these companies, it doesn’t fix the problem that I identify in the book, which is that there is a system of power that has been constructed where these companies and the people running these companies get to make decisions that affect billions of people’s lives around the world. And those billions of people do not get any say in how it goes.

Democracy, Governance, and the AI Power Structure

STEVEN BARTLETT: Those people, they can go to the polls, right? So if the public are sufficiently educated, they can go to the polls and pick a leader that says they’re going to legislate or pass laws or try and pass laws.

KAREN HAO: Yes. But at the speed and pace at which these companies operate and at the sheer scale and size, they’re able to also spend extraordinary amounts of money — hundreds of millions in this upcoming midterms — to try and kill every possible piece of legislation that gets in their way and craft legislation that would codify their advantage.

And so to me, I think sometimes as a society we obsess a little bit with are these leaders good or bad people? And to me the bigger question is, is the governance structure that we’ve created a sound one that allows broad participation, or an anti-democratic one that has consolidated this decision-making power in the hands of the few? Because no person is perfect. I don’t care who is at the top of these companies, they’re not going to have the ability to make decisions on behalf of so many people around the world who live and talk and have a culture and history that are fundamentally different from them without things going wrong.

And so that is why throughout history we’ve moved from empires to democracy. It’s because empire as a structure is inherently unsound. It does not actually maximize the chances of most people in the world being able to live dignified lives.

The China Argument: Devil’s Advocate

STEVEN BARTLETT: I’m going to try and take on their point of view. So this is me playing devil’s advocate. Okay, but Karen, if the US don’t continue to accelerate their research with AI, at some point China’s model is going to become so smart and intelligent that we’re basically going to have to rent it off them. They’ll get the scientific discoveries, they’ll discover the new era of autonomous weapons, and we will be their backyard. And logically, that argument does appear to be pretty true.

KAREN HAO: No, it’s not.

STEVEN BARTLETT: If we scale up — if we just imagine any rate of change with this intelligence — at some point we’re going to come to a weapon that could theoretically disable all of the United States’ electricity, their weapons systems. It would know exactly how to disable the United States from a cyber perspective because it would be that smart. All you’ve got to imagine is any rate of improvement over any period, any sort of long period of time. So this is a theory that might be true.

KAREN HAO: And if it’s true — I mean, yeah, any theory might be true.

STEVEN BARTLETT: But again, going to this point of like, even if it’s a small percentage, it’s worth paying attention to. On the other side of the coin, this is a theory that people talk about — it could be the case that the most intelligent civilization is going to be the superior civilization. Logically, that’s a pretty sound thing to say. No?

Challenging the Assumptions of AI Scaling

KAREN HAO: So there are a lot of fundamentals in this argument that would need to be true in order for this to be a viable argument. And let’s knock them down one by one. So the first one is that these systems are intelligent and that just scaling them is going to bring us more intelligence.

STEVEN BARTLETT: So far, so true.

KAREN HAO: No, it’s actually not. Because first of all, we don’t actually know if these systems are — like, intelligence is not — it’s not quite the right analogy almost. It’s sort of like, a calculator can do math problems faster than a human. Does that make it intelligent?

STEVEN BARTLETT: It has a narrow intelligence because it’s solving a narrow problem, which is like one plus one equals two. But —

KAREN HAO: And these systems, they actually also are quite narrowly intelligent in the sense that even though these companies say that they’re everything machines that can do anything for anyone, they actually can only do some things for some people. This is like the jagged frontier of these AI models. Some of the capabilities are quite good. Other capabilities are not that good.

You know why that happens? It’s because the company can only focus on advancing certain types of capabilities. It can’t literally focus on advancing all types of capabilities. They have to actually set their mind to advancing a certain set — by gathering the data that is needed for that capability, by getting a bunch of human contractors to annotate and train the model to do that exact thing. And so scaling these models is actually a perpendicular question to whether we are actually getting more cyber capabilities specifically and more military capabilities specifically.

STEVEN BARTLETT: I would argue that most of the top people in AI believe that the intelligence is going to continue to scale for some time. A lot of them do, like Geoffrey Hinton does.

The Myth of General AI Intelligence

KAREN HAO: And again, it’s back to his hypothesis about how human intelligence works and what the appropriate model of the brain is. His hypothesis throughout his career has been the brain is a statistical engine. But that’s his hypothesis, and that is not universally agreed upon, especially among people that are not in the AI world.

When you talk with neuroscientists and psychologists, people who actually study human intelligence and the human brain, that is where you start to get a lot of debate and disagreement about this particular view that Hinton has.

And so this is kind of like one of the things is, AI is already being used in the military and has been used in the military for a long time. But specifically accelerating large language models isn’t just the only path for getting military capabilities. Like, the companies would have to choose to specifically pick military capabilities to accelerate, not just general intelligence.

They create this myth that they are actually pushing the frontier of all of the capabilities of the model, but that’s not what’s actually happening internally. And I had hundreds of pages of documents on how they were specifically training models. They picked what capabilities they want to advance. And you know how they pick them? It’s based on which industries would be able to pay them the most money for their services. So they pick finance, law, medicine, healthcare, commerce. It’s not actually intelligent like a baby, where the more that the baby grows up, they start having these general abilities.

STEVEN BARTLETT: I think I have jagged intelligence. I’ll be honest. I wasn’t going to say it, but I know a lot about a little bit.

KAREN HAO: Yeah, but you also have the capability to learn and acquire knowledge by yourself, and you also have the ability to choose what you’re going to learn and acquire by yourself.

STEVEN BARTLETT: It’s not easy and it takes a lot more time than these models. It seems less compute, but.

KAREN HAO: And you can learn how to drive in one place and then immediately know how to drive in another place. These models cannot do that. Every time a self-driving car is shifted to another location, it has to completely retrain on that location. So all the self-driving cars — I mean, we’re sitting in Austin right now and there’s all these self-driving cars that are driving through Austin.

STEVEN BARTLETT: But when one of them learns, they all learn.

KAREN HAO: Well, it’s just because it’s an operating system that has an AI model as part of it, and you’re training the AI model and then you deploy the AI model across all the self-driving platform.

STEVEN BARTLETT: Which is a big advantage, because if one Optimus robot learns one thing in one factory, they all learn it. And imagine that — imagine if humans, if we all learned what all the other humans learned, that would give us such an unbelievable competitive advantage. I mean, one of the ways we did that is through communication.

KAREN HAO: It could not, because they could be learning the wrong thing, which has also happened again and again with these technologies — all of them then learn the wrong thing and they all have the same failure mode. I mean, part of the resilience of human society is that we do have different expertises and we also have different failure modes.

Holding AI to a Higher Standard Than Humans

STEVEN BARTLETT: I think sometimes we hold AI models to a higher standard than we hold humans to. And in a weird way, because I’d hear on stage — we’re in Austin at the moment — and I’d hear people go, “Ah, but those AI models, they hallucinate.” Sometimes I’m like, have you met a human? I hallucinate all the time. I can barely spell or do math.

KAREN HAO: Yes, but it’s once again using this analogy that was specifically picked in the early days of the field as a way to market these technologies. We’re repeatedly using the intelligence analogy and relating these machines to human intelligence as a way to try and gauge whether or not it is good or worthy or capable in society.

STEVEN BARTLETT: I think the output is the thing that really is the most consequential, which is like, okay, it might have a different brain and a different system, but does it arrive at the same capability? Like, is it able to do surgery on someone’s brain? Is it able to drive a car? My car drives itself in Los Angeles — I don’t touch the steering wheel and I can drive for many, many hours. And here in Austin, I just saw the ones the other day where they’ve removed the steering wheel and the pedals, the new Cybercabs. So I go, it doesn’t really matter if it’s using a different system. If it’s navigating through the world as a car and it has a better safety record than human beings, then as far as I’m concerned, intelligence or not, it’s like —

KAREN HAO: Yes, but that was not the original argument that you made, which was that these systems are just generally going to become more intelligent across different things based on the prediction. This is a prediction that you’re making. And this is a prediction that all —

STEVEN BARTLETT: — that Ilya’s making, Dario’s making, Elon’s making, Zuckerberg’s making, Altman’s making, Demis is making.

KAREN HAO: And do you know what the common feature of all of them is? They profit enormously off of this myth.

Scaling Up: Brute Force Intelligence or Financial Strategy?

STEVEN BARTLETT: Elon has recently spearheaded the construction of Colossus, a massive supercomputer in Memphis housing 100,000 GPUs, specifically to scale up their Grok AI models faster than their competitors. It appears that they’ve all converged around this idea that you can brute force your way to greater, more generalized intelligence.

KAREN HAO: They’ve converged around the idea that you can brute force your way into models that they can sell to people for automating certain tasks that are financially lucrative.

STEVEN BARTLETT: And I heard Elon say that if you’re a surgeon, there’s just no point. He was like, “Don’t train to be a surgeon.” He says, in a couple of years time, Optimus and AI generally are going to be better than any surgeon that’s ever lived. Do you think these things are true?

KAREN HAO: Well, I’m pretty sure it was Hinton that famously — slash infamously — said there would be no need for radiologists anymore. He set a deadline that we’ve already passed. I don’t remember how many years ago. Radiology is doing great as a profession.

STEVEN BARTLETT: Do you think it will be in five years?

KAREN HAO: Okay, so this once again goes back to this question of why do we build technology? And why should we specifically be building AI? For me, the whole project of technology development and advancement is not to advance technology for technology’s sake. It’s to help people.

And there have been lots of research that has shown that actually the best outcomes for people in a healthcare setting is for the radiologist to have the AI model in their hands and for the human expert to use the AI model as a tool, as an input into their judgment. And it is that combination that leads to the most accurate and early diagnoses of certain types of cancer that then help improve the prognosis of the patient.

The Self-Driving Car Debate

STEVEN BARTLETT: Do you believe that in the coming years all the cars — pretty much all the cars on the road — will be driving themselves?

KAREN HAO: No.

STEVEN BARTLETT: You don’t think so? How come?

KAREN HAO: Because of the way the technology works. Currently, the way that AI models are primarily developed, they’re statistical engines. You have what’s called a neural network, which is a piece of software that has a bunch of densely connected nodes, like parameters.

STEVEN BARTLETT: Is this what they call parameters?

KAREN HAO: Yeah, pretty much. And you’re just pumping a bunch of data into it and then it’s analyzing the data and finding all these correlations, finding all these patterns, and then it’s through those patterns that the machine is then able to act autonomously.

And so the way that they’re training a self-driving car is they’re recording all this footage and then they have tens of thousands or hundreds of thousands of human contractors that draw literally around every single vehicle in the footage, every single pedestrian, every single traffic light, every single lane marking, and label it exactly as such, so that then it’s fed into an AI model that can identify all of these different components. And then it’s connected to another piece of software that is not AI, that’s saying, okay, if the AI model recognizes a pedestrian, we do not run over the pedestrian. If the AI model recognizes a red traffic light, we stop.

And so the thing about statistical engines is that it’s based on probabilities, it’s not based on deterministic logic. Systems make errors all the time, and it is technically impossible to get them to stop making errors.

STEVEN BARTLETT: Humans make errors way more than systems in this case.

KAREN HAO: Yeah.

STEVEN BARTLETT: Like the safety record — isn’t it like 10 times more safe to be driven in a Tesla with autonomous driving than it is for a human to drive?

KAREN HAO: It depends on the place. It depends on whether the Tesla was trained to specifically navigate the place that you’re driving.

STEVEN BARTLETT: Just get drunk.

KAREN HAO: Because if it’s in Mumbai, in some place in Vietnam, no, it would not be safer. I would much rather be driven by someone that has been driving in that place their whole life. I’m not arguing against the fact that in certain places where the car has been explicitly trained to drive, it has a better safety record than the humans that are driving in that place. But you specifically asked if I think that most cars in the world —

STEVEN BARTLETT: Let’s say the United States, because we’re here.

KAREN HAO: I don’t actually think that it’s imminently on the horizon.

STEVEN BARTLETT: Ten years?

KAREN HAO: No, I don’t think so.

STEVEN BARTLETT: I sat with Dara from Uber and he’s pretty convinced that his 9 million couriers will be replaced by autonomous vehicles.

KAREN HAO: I mean, how long have self-driving cars been invested in? It’s been more than 10 years. And what percentage of cars right now are autonomous on the US roads? Part of it is actually not a technical problem. Part of it is also a social problem — do people even trust getting into these vehicles? Part of it’s also a legal problem, which is if the self-driving car kills someone, which has happened —

STEVEN BARTLETT: Yeah, it has happened.

KAREN HAO: Who is responsible?

STEVEN BARTLETT: So in the case in LA, it was both Tesla and the driver, because the driver dropped their phone. They looked down — this was a couple of years ago, I believe — and they went to grab their phone and they hit someone. And so it went to court and they were held both responsible, both the driver and Tesla. In terms of Tesla, pretty much everyone that gets the car, it comes with autonomy now for pretty much most people, I believe.

KAREN HAO: Partial autonomy.

STEVEN BARTLETT: Yeah, it’s called Full Self-Driving at the moment, where —

KAREN HAO: Yes, it is called Full Self-Driving.

STEVEN BARTLETT: Full Self-Driving, supervised, where you kind of have to be looking in the right direction.

KAREN HAO: But yeah, so it’s partial autonomy.

STEVEN BARTLETT: And here in Austin it’s full autonomy because there’s no steering wheel on the new car, so you can’t drive it anyway. But the Model Y is the undisputed highest-selling car, best-selling car in the world across all brands.

I guess my point here is, these predictions where they say AI is going to completely change transportation and driving, it’s going to completely change things — lawyers aren’t going to have jobs, accountants aren’t going to have jobs. Do you believe that they are true? Do you believe that there’s going to be mass job displacement?

AI and the Future of Employment

KAREN HAO: Okay, so I do think that there is going to be huge impacts on employment and we are already seeing those impacts. It is not simply because the AI models are just automating those jobs away. It is specifically because the models are improving in certain capabilities based on what the companies that are developing them choose to improve them on.

And executives at other companies are then deciding to fire or lay off their workers because they think that AI can replace the worker, irrespective of whether that might be true. And there have been cases — the Klarna CEO who laid off a bunch of people thinking that he would replace everyone with AI, and then it didn’t actually work and he had to ask some people to come back.

The Reality of AI and Employment

STEVEN BARTLETT: I actually DMed him about this. If you’re hearing this, this is because I’ve DMed Sebastian and he’s fine with me sharing this. He said, because I’ve heard his name mentioned a lot. And so when we talked about AI in the past and people mention Sebastian and Klarna as the example, I wanted to clarify with him what the truth was.

He said, “It’s great to hear from you. I think sometimes people struggle with two things can be true at the same time. I think it might be time to come back on your podcast. To your point, this is the media misinterpreting my tweet. We are doubling down on AI more than ever. Klarna is shrinking with almost 100 employees per month due to AI. We used to be 7,400 at the peak a year ago, 5,500. Now we’re 3,300 and by the end of summer, so this was last year, we’ll be 3,000 people. AI handles 70% of our customer service conversations at this moment. This is because we have realized that with AI the production cost of software comes down to almost zero. Just like manufacturing used to be all handcrafted and then the machines came. Code used to be all handcrafted up until a few years ago and now it is machine produced. And ultimately we pay people more than ever for the unique handcrafted man made stuff. Klarna is a bank. People will want to connect to humans, not only machines. They want us to be personable, relatable, even flawed. So we need to make sure, while we are automating replacing with AI in parallel, we make sure we offer a super available human experience.”

KAREN HAO: I’m really glad you read this because I think it touches on some really important nuances to the AI — the impact that AI is going to have on employment. So I think there’s often these binary narratives. It’s like AI is going to come for every job, or people say AI is not actually working and it’s not actually coming for jobs.

And the reality is that it’s coming for jobs. There are definitely jobs that are being automated away because of the capabilities of their models, and there’s also jobs that are being lost because executives are deciding to lay off the workers even if the models don’t match the capabilities — because it’s good enough, they would rather have the good enough model for way cheaper.

STEVEN BARTLETT: Or they made a mistake with hiring, they bloated their team. And it’s a great, convenient thing to say.

KAREN HAO: Exactly. There are many reasons, but clearly we’re already seeing impacts on the job market. The US jobs report that came out earlier this year showed that there has been a decline in hiring, a slowdown in hiring across especially white collar professional industries.

The Anthropic Report and Job Disruption

STEVEN BARTLETT: And you saw Anthropic’s report, didn’t you, this week? The TLDR is it matches what you were saying, where Anthropic looked at exactly how people were using their models and they looked at what people are saying, and they said that there’s been a 40% reduction in entry level jobs in particular.

And then they made this graph which has gone viral over the Internet. The red shows where we are now in terms of capability. And based on how people are currently using the models, they extrapolated out that the blue part will be the disrupted parts. This is the things that they say AI can do right now, but people don’t realize it yet. So if you look at it, it’s like it’s the physical real world human stuff which robots maybe can do someday, like construction or agriculture, that are untouched — but office and admin, finance stuff, math.

KAREN HAO: And notice that these are all of the things that I just named, that they purposely — finance, math, law, media and arts.

STEVEN BARTLETT: That’s me cooked.

KAREN HAO: Yeah, Office and Admin. They do focus a lot on assistant type and managerial work. But the other thing that the Klarna CEO said was that people also want human experiences. So it’s not actually just about the capabilities of the models, it’s also about what people want. Like some things they would turn to AI for and some things they wouldn’t, irrespective of whether or not AI is capable of doing it, but because of a preference that they want human to human interaction.

And so what we’re seeing right now is the thing that happens with every wave of automation, which is that there is a bunch of entry level work that gets automated away and there are also new jobs created. But the jobs that are created are in one of two categories. There are people that get even higher skilled jobs — and what he was saying, like we pay people more for the handcrafted code now. And there’s also the people who get way worse jobs.

So there was this amazing article in New York magazine that was talking about how a lot of people are getting laid off and then they end up working in data annotation, which is the labor that I’ve been referring to throughout this conversation, that companies need in order to teach their models the next thing that the companies are trying to automate. And so a marketer gets laid off and then they go and work for a data annotation firm to train the models on the very job that they were just laid off in, which will then perpetuate more layoffs if that model then develops that skill.

The article was talking about how this has become a huge catch all for a lot of people that are struggling with finding job opportunities right now, including award winning directors in Hollywood that are actually secretly doing this data annotation work to put food on the table.

And so when they talk about there’s going to be mass unemployment and then there’s going to be some new jobs created that we can’t even imagine, I think a lot of these narratives rarely talk about — first of all, why are some jobs going away? It’s not just because of the model capabilities, also because of executive choices and because of the rhetoric that they use if they want to just downsize. But the other thing that is rarely talked about is that a lot of the jobs that are created are way worse than the jobs that were there. And it breaks the career ladder. So it’s the entry level and the middle tier jobs that get gouged out. It’s higher order jobs and then way more lower order jobs that get created. And so how do people continue to progress in their careers? There’s no more rungs on the ladder.

Navigating the AI Transition as an Employer

STEVEN BARTLETT: I actually don’t know the answer to this question. And I’ve been furiously trying to find a good answer to this question. Everything is theory. And for my audience, most of my audience don’t run businesses. A lot of them do, a lot of them aspire to, but they don’t run businesses. So they’re also in the land of theory. They’re hearing lots of different things. Jack Dorsey does his tweet saying he’s halving his headcount because of AI. They don’t know what’s true. They don’t know the internal economics at Jack’s company. And did he bloat the company during the pandemic, and he’s just using this as an excuse to make his share price spike 7 points because his investors now think they’re an AI company — whatever, it’s hard to parse through.

So eventually I go, okay, what am I doing? I have hundreds of team members, probably 70 companies I invest in, maybe five or six that I’m the lead chair of. What am I actually doing on a day to day basis? Right now I also consider myself to be head of recruitment. But in the last month in particular, I have met extremely capable candidates in terms of cultural alignment, hard work, those kinds of things. But I’ve had to take a great deal of pause because when I run the experiment of can I get an AI agent to do that exact same thing, the answer is increasingly yes, especially in a world of open source.

KAREN HAO: And so what I’m curious about now — you confront this decision where you’re seeing, in this short term period you could just choose the AI agent, and in the long term period there is no career ladder. So who are you promoting into these senior roles? How do you resolve it for your own company?

STEVEN BARTLETT: Yeah, it’s a good question. So there’s kind of two ways I’m thinking about it. I think really deep expertise is very, very valuable because if you’re now the orchestrator of potentially AI agents, it’s really about having a deep understanding of the right question to ask. And that’s someone who has deep expertise on something. So I need my CFO, because if she’s going to be orchestrating our team of agents that might be doing financial analysis or whatever else, she needs to understand what to tell them to do in our company. And in turn, financial analysts can’t do that. They need the 50 odd years of experience that Claire has on the other end.

I need Kaz. Kaz is 25. Kaz knows everything about AI agents. He’s a young Japanese kid who’s highly, highly curious. On the weekend, he’s building AI agents to solve problems in my life. I need those two kinds of thinking — highly proficient agent-maxing young kids, or they don’t necessarily need to be young, but really lean in, high curiosity — that’s creating a force multiplier on my business. And then I need deep expertise.

Now, everything else outside of there — another group I’ve thought of is people with extremely great IRL people skills, because we do meet people in real life. We greet you when you arrive here, when we go for lunch with big clients that we have, whether it’s Apple or LinkedIn or whoever it might be, we need to schmooze. And we have teams who are in person in the office. So we do a lot of stuff IRL, and increasingly we’re building communities even for this show. We’re doing community events all around. So we need people that are good at that as well — IRL, bringing people together in real life and organizing stuff. Those are the three groups of people that I’m like, you know, irreplaceable right now.

KAREN HAO: And if you were to — all of the roles that could be done by AI agents, if we were to replace them with AI agents, do you think you would still have these three roles, pools of people to hire and promote into the three critical things that you need in the long term?

The Human Connection Paradox

STEVEN BARTLETT: If things carry on at the current rate of trajectory, one could assert that even those roles would experience pressure. People think of things either statically or linearly or exponentially. If you imagine an exponential rate of improvement, which is what I’ve seen, even a 10% compounding rate of improvement — at some point, I think what remains is actually the IRL, irreplaceably human stuff, human to human. Our Maslovian needs of being in person aren’t going to change. We need connection. Humans get very sick when they don’t have other human beings in their life and strong, deep relationships.

KAREN HAO: 100% agree.

STEVEN BARTLETT: So that stuff is going to matter a whole lot. I have this contrarian weird take that actually — maybe this is the first technology that’s going to deliver on the promise of making us human and connected, because we’re going to be rendered useless of everything else other than what humans are good at. Because all the other technologies said, “Oh, we’re going to make you more connected, connecting the world.” And they disconnected the world and isolated the world. But maybe this is the one. It’s so intelligent now that it doesn’t need us to f* around in spreadsheets anymore.

KAREN HAO: Do you see that actually happening in real time right now? That it’s making us more able to be in person, connected with one another, having deeper social community engagements?

STEVEN BARTLETT: Yes, yes. And I’ll give you some data points.

KAREN HAO: Okay.

The Human Cost of AI: Data Annotation and the Working Class

STEVEN BARTLETT: Data point number one. The Financial Times released a report on social media usage, and what they saw is 2022 is the peak, and it’s plateaued ever since. The generation that’s plateaued the fastest and heading down is the younger generations. The boomers are still off to the races, right on Facebook and stuff.

And then you look at the way Gen Alpha are using social media. They’re not posting as much. They call it posting zero. They’re scrolling sometimes. But they’re in dark social environments like WhatsApp and Snapchat and iMessage. They’re not performing to the world.

They also value IRL experiences much more than any other generation. They’re not getting smashed. Every brand has a run club. We’re seeing run clubs exploding around the world. And we’re seeing this real sort of almost innate realization that technology let us down at some fundamental level, like dating apps let us down, social networking kind of has let us down.

And we’re seeing, I think, maybe a bifurcation of society where a lot of people are going, “f* this. I want to go back to what it is to be a human.” And I would imagine that in such a world where intelligence is so sophisticated that we no longer needed to sit at laptops, I think screen time’s going to continue to fall. I think you go into an office, you’re not going to see people sat at laptops. You’re going to see something completely different.

And then we talk about robots and Optimus robots. Elon says there’ll be 10 billion Optimus robots. Elon has been wrong with timing before. He’s almost never been wrong on the big things completely. His timing has got a bad track record. So I think he’s probably right.

I’ve got some people on the way from Boston Dynamics and these other big companies like Scale AI, and they’re actually bringing the robots here to show it, like folding laundry, doing the dishes. I’m not saying that’s what I would want in my home, but I think factory work is going to completely change. I think a lot of manual labor is going to completely change. And I think we’re going to be forced to do what only we can do.

Sebastian, who’s the CEO of Klarna, has actually just called me. Hello, Sebastian. You all right?

KAREN HAO: Hey, how are you?

STEVEN BARTLETT: I’m good. How are you? It’s been a while. It has been a while since you’re on the show. I was just saying we do need to get you back on. I just had a couple of simple questions because I do a lot of interviews and Klarna’s always mentioned because I think the media has said that you double down on AI, then you reversed because it didn’t work out. So I know I spoke to you a while ago and we exchanged a couple of DMs about it, but that was more than almost a year ago now. So I just wanted to get an update on Klarna’s business, AI agents and all of that, if possible.

First and foremost, we were early on released AI to support our customer service, which had that initial benefit of more calls being dealt with by AI, which customers liked because those calls or chat messages were much, much faster and more qualitative. Since then that has actually expanded slightly.

What we did, however, try to communicate as well is that we believed in a world where AI is cheap and available, the value of human interaction will be regarded as higher. So the future of customer service VIP is a human. We have then hence doubled down on providing more of that.

But at the same point in time, the efficiency gains within the company have continued. I mean, we used to be about 6,000 people and now we are less than 3,000, which is two, three years since we stopped recruiting. And at the same point in time our revenue has doubled. So you can clearly see that AI has allowed us to do more with less people, but we have avoided layoffs and instead relied on natural attrition when people move on to other jobs.

From my perspective, we will continue to not really recruit much. I mean, we recruit a little bit here and there, but we expect that natural attrition of 10, 15% per year to continue on, to become fewer.

I think the big breakthrough was really in November, December last year where even the most skeptical engineers who are very well renowned and appreciated, like the founder of Linux and stuff like that, basically said that coding has now been resolved and hence you don’t need to code anymore. And that was a common sentiment. So I think in coding, that’s definitely engineering work. That has been a tremendous shift in the last six months.

What do all these people go do, Sebastian?

I am optimistic. I mean, I think obviously people will have a lot of opinions about this topic, but I still believe that we are going to move towards a richer society now. In the short term there could be more worry about what happens if people don’t get a job and so forth. But I think in the longer term I am optimistic about what it means for society and humanity.

Thank you so much. Seb, I’ll chat to you soon. Thank you for taking the time. I appreciate you, man. Mate, thanks. All right. All right, bye-bye.

KAREN HAO: Well, I actually had thoughts on something that you said before he called, which is you were saying that the Gen Zers, there’s this trend that they’re actually disconnecting from technology, so they’re becoming more in person. And then there’s this other class of workers that are actually leaning into the technology but then becoming more human because they’re leaning into the technology, because they’re realizing that they should actually just be spending more time doing in-person-to-person interactions rather than staring at a spreadsheet. And so they’re no longer doing the typing and whatever.

I really want to go back to this New York magazine piece that just came out because what you’re describing is true for a very specific category of people, which is often the business owners and leadership within companies that actually can make these decisions on how they spend their time and what they ultimately do with their time. But what the piece talks about is the working class, people who are not business owners that are then having to experience being laid off and then working for the data annotation industry, which is now one of the top jobs on LinkedIn, by the way.

STEVEN BARTLETT: Really?

KAREN HAO: Yeah. So LinkedIn had a report that showed the top 10 jobs with the highest growth in the last year. And data annotation is on that list.

STEVEN BARTLETT: And for anyone that doesn’t know what data annotation is.

KAREN HAO: Yeah. So data annotation is the process of teaching these chatbots or any AI system to do what they ultimately are able to do. So the fact that ChatGPT can chat is because there were tens of thousands or hundreds of thousands of people that were literally typing into a large language model and showing it, “This is how you’re supposed to then respond when a user types in a prompt like this.”

Before they did that work, ChatGPT didn’t exist. You would prompt the model and the model would generate some text that was not in dialogue with the person. It would kind of generate something that was adjacently related.

STEVEN BARTLETT: Is this what they call reinforcement learning, where you give it like a…

KAREN HAO: It’s a part of the process of reinforcement learning. So you do data annotation, which is literally showing lots of different examples of things that you want the model to know. And then reinforcement learning is getting the model to then train on those examples iteratively in a way that then gives the model some of those capabilities.

And what the New York magazine piece highlighted is many, many of the people that are getting laid off now, or are struggling to find work — and these are highly educated people, they’re college graduates, PhD graduates, law degree graduates, doctors, and award-winning directors — are then struggling to find employment in the economy because the economy has been very much restructured by AI. They are then finding themselves serving this industry.

And the industry is designed in a way that is extremely inhumane. Because the companies that use these data annotation services — there are these third-party providers that are data annotation firms — an OpenAI, Grok, a Google, they will hire these firms to then find the workers to perform the data annotation tasks that they need. These third-party firms are incentivized to pit workers against each other because they want this data annotation to happen at speed and as cheaply as possible, so that they can also compete with one another in this middle layer to get the contract from the client.

And so all of these workers that were interviewed for this New York magazine story talk about how they actually no longer have an ability to be human, because they are waiting at their laptop to be pinged on Slack for when a project is going to open up for data annotation. Because they’ve tried job hunting, they literally can’t find anything else. This is the thing that’s going to help them put food on the table for their kids.

And there was this one woman who said, “I have so much anxiety about when the project is going to come, when it’s going to leave, that when the project came, it was right when my kid was coming off of school and I just started tasking furiously because I don’t know when it’s going to go. And I need to earn as much money as possible in this window of opportunity. So then when my kid came home and tried to talk to me, I screamed at my child for distracting me.”

And then she said, “I’ve become a monster and I am not even allowed to go to the bathroom or take care of my kids, let alone myself, because this industry that is absorbing more and more of the workers that are being laid off is mechanizing my life, atomizing my work, devaluing my expertise, and then harvesting it for the perpetuation of this machine that all of these AI executives are saying is then going to come for everyone else’s jobs.”

And so what you were saying about this class of workers — the business owners that get to become more human because there are all of these AI models now doing the tasks that they don’t have to do anymore — it is at the cost of the vast majority of people who are not business owners, that are struggling to find work, getting absorbed into the work of then providing these technologies that the business owners can use. And instead of becoming more human, they feel like their humanity has been squeezed and diminished and they have no ability to have control, agency, and dignity in their lives anymore.

The Human Cost of AI: Jobs, Environment, and Inequality

STEVEN BARTLETT: I think this is a big question that kind of pertains to this graph here, which is, all of these people, if we believe Anthropic’s prediction of who will be disrupted, these people in these industries like arts, media, legal, life and social sciences, architecture and engineering, computer and maths, business and finance and management, and also office and admin. These people, if we believe this, would have to retrain at something else.

And unlike the industrial revolution, where you might get 10, 20 years to retrain because factories take a long time to build, the distribution layer that AI sits on top of is the open Internet. So this is why ChatGPT can go pop and get hundreds of millions of users in no time at all and become the fastest growing company of all time. One of my fears is that this disruption takes place at a speed where we can’t transition.

KAREN HAO: And I think you said that sentence in the passive voice. The transition would happen at a speed, but who is driving that speed? It’s the companies. The companies and their race with one another.

STEVEN BARTLETT: Yeah.

KAREN HAO: And so they are driving the transition to happen at a speed at which it would be really hard to take care of all of the people that would be bulldozed over by.

STEVEN BARTLETT: This is one of the crazy questions that no one can answer for me when I sit with these people that are AI CEOs. I go, so what happens to the people? If you agree that this is going to happen at superspeed, I’ve spoken to the CEO of Uber, Dara, who said very similar things to what you’re saying — that there’ll be data labeling jobs, for example, for the drivers, but they can’t all become data labelers.

And there’s a question around meaning and purpose and fulfillment, and that comes from losing your meaning in life. I also sit here with so many people who talk about how their father lost their job in Iran or some other country and came to the United States and had to be a toilet cleaner. In one particular case, someone was a doctor in Iran, but came to the US and was a toilet cleaner and had to deal with the sense of shame that that particular person felt, and the lack of dignity that that caused, and how that made that person’s self-esteem feel, and the depression and alcoholism that transpired from that. If this happens at a large scale across society, there’s going to be a ton of consequences like that.

The Haves and Have-Nots: AI’s Widening Inequality

KAREN HAO: I mean, this is like the core themes of my work. And the reason why I’m critical of these companies is that they are creating technologies in a way that creates the haves and have-nots in an extreme form. It’s exacerbating the inequality that we already see in the world. The people who have things will have way more riches, they’ll have way more free time, they’ll be allowed to be more human. But the people who don’t have things are being squeezed even more.

And it’s not just from a work perspective. I talk in my book also about the environmental and public health crisis that these companies have created, where they are building these colossal supercomputer facilities in communities all around the world. And they specifically pick some of the most vulnerable communities.

We’re sitting in Texas right now. OpenAI’s largest — one of its largest data center projects — is being built in Abilene, Texas, as part of the Stargate initiative, which was an effort announced at the beginning of Trump’s second administration to spend $500 billion on AI computing infrastructure. This facility, when it’s finished, will consume more than a gigawatt of power, which is over 20% — over 20%. So this is actually a little bit inaccurate. Now, this was something that circulated online for a while, but there are updated numbers.

STEVEN BARTLETT: Just for someone not watching — because they’re listening on Spotify or something — it’s a picture of the size of this facility.

KAREN HAO: So this is not the Abilene, Texas one. This is a Meta facility. So let’s first talk about OpenAI’s facility in Texas. That one would be the size of Central Park, and it would run a million computer chips, and it would require the power of more than 20% of New York City.

Data Centers and Environmental Racism

STEVEN BARTLETT: Do you know, one of the things which I found confusing — so I’d like to alleviate the dissonance — is I thought you were saying earlier that you didn’t think the job disruption promises were real.

KAREN HAO: No, what I was saying is that when we talk about what these executives predict about the future, we need to understand that they are ultimately trying to influence the public in a way that allows them to continue maintaining control over the technology.

STEVEN BARTLETT: But objectively, do you think that the job disruption that they talk about —

KAREN HAO: Yeah, I mean, I see real — well, not necessarily. I don’t want to comment specifically on this chart, but we’ve already seen in job reports that there is a restructuring of the economy happening right now.

But going back to the data center — so this supercomputer facility, it’s a Meta supercomputer facility, being built in Louisiana, and it would be four times the size of the Abilene, Texas one, and use half of the average power demand of New York City. So it’s 1/5 the size of Manhattan. This makes it seem like almost all of Manhattan, but it would be 1/5 the size of Manhattan.

When these facilities go into these communities, what happens? Power utility increases, grid reliability decreases. The facilities also need fresh water to generate the power for powering them, as well as fresh water to cool. And there have been lots of documented stories of communities that are already really constrained in their freshwater resources — they’re under a drought — when a facility comes in, and then the community is actually competing with this facility for fresh water. I talk about one of those communities in my book.

And also, sometimes these facilities, instead of connecting to the grid, have a power plant pop up next to them. So in Memphis, Tennessee, where Musk built Colossus — the supercomputer for training Grok — he used 35 methane gas turbines to power the facility. This is a working-class community, a Black and Brown community, a rural community, that was not even told that they would be the hosts of this facility. They discovered it because they literally smelled what seemed like a gas leak in all of their living rooms. And that’s when they discovered that these methane gas turbines were taking away their right to clean air.

This is a community that’s already been facing a history of environmental racism. They had already had lots of struggles to access their right to clean air. And now there’s this huge supercomputer that’s landed in their midst that is pumping thousands of tons of toxins into their air, exacerbating the asthmatic symptoms of the children, exacerbating the respiratory illnesses of other people. It’s one of the communities that has the highest rates of lung cancer.

STEVEN BARTLETT: And so that supercomputer’s taking their —

KAREN HAO: — jobs, and then they also have supercomputers taking their jobs. So this is what I mean — the haves and have-nots are fundamentally being pulled apart even further.

If you, in this version of Silicon Valley’s future, are in the misfortunate category of being a have-not, we are talking about you now getting a job that is way worse than what you had, because you might be doing data annotation. And you might be treated as a machine rather than as a human, to extract the value of your labor for perpetuating this labor-automating machine that these people are building. You might be competing with these facilities for fresh water resources. They’re also polluting your air. Your bills have increased. The affordability crisis is getting worse. How is that making people able to be more human?

Building Bicycles Instead of Rockets: A Better Path for AI

STEVEN BARTLETT: What do we do about it?

KAREN HAO: Okay, so one of the analogies that I always use is: AI is like the word transportation. Transportation can literally refer to everything from a bicycle to a rocket. And we have nuanced conversations about transportation where we always say we need to transition our transportation towards more sustainable options. We need to transition towards public transport, electric vehicles. And we don’t ever say everyone should get a rocket to serve all of their transportation needs.

Like, we’re in Austin. If you used a rocket to fly from Dallas to Austin, that would just make no sense. It’s just a disproportionate use of resources to get the benefit of getting from point A to point B. This is how we should think about AI.

So all of the models that we’ve been talking about, I like to think of them as the rockets of AI. They use an extraordinary amount of resources and they provide some dramatic benefit to some people, but they’re also exacting an extraordinary cost on a large swath of people because of the costs of developing this technology.

Why don’t we build more bicycles of AI? This is things like DeepMind’s AlphaFold, which is a system that predicts how proteins will fold based on amino acid sequences — really important for accelerating drug discovery, for understanding human disease. And it won the Nobel Prize in Chemistry in 2024. And the reason why it’s a bicycle of AI is because you’re using small, curated data sets. You just have data that has amino acid sequences and protein folding. So that means you need significantly less computational resources to develop the system, which means significantly less energy, which means less emissions, and so on and so forth. And you’re providing enormous benefit to people.

STEVEN BARTLETT: It feels like the horse has left the stable in this regard, because they’ve already taken people’s IP, they’ve taken media, they train on this podcast. We know they do because it shows that they do. I think there’s a button actually in the back end of YouTube now that allows you to just click it, and it says, “We will train on your YouTube channel.” So the horse has kind of left —

KAREN HAO: Here’s the thing. If the horse truly had left the stables, they wouldn’t have to train on anything anymore. Why is it that their appetite for data has actually expanded? It’s because in order to build the next generations of their technologies, in order to have the technologies continue to be relevant and continue to update with the pace of new knowledge creation and society’s evolvement, they need to train again and again and again and again.

And why are they employing actually more and more data annotation workers over time? It’s because they need more and more of that work over time. I mean, I’ve been reporting on data annotation work for over seven years now, and it’s not gone down — it’s increased.

STEVEN BARTLETT: Do you think there’s any chance of it going down? Do you think there’s any chance of this sort of brute-force scaling approach — where you take data, you take computational power, energy, and you have data labelers and build out more and more parameters for the models — do you think there’s any chance it’s going to stop or go in a different direction other than the one it’s going in now?

KAREN HAO: I would love to reframe the question and say: what should we be doing in this moment where it’s not going down, where we do recognize that actually these companies in this moment need continued resources, inputs, and labor to perpetuate what they are doing?

Breaking Up the Empire: Regulation and Resistance

STEVEN BARTLETT: Yeah. Because this sounds like “stop.” And I just feel like “stop” is like a hurdle. With the government in place, they’re supporting these companies like crazy globally. This is happening. So “stop” doesn’t feel —

KAREN HAO: I always say we need to break up the empire and we need to develop alternatives. And we are already seeing a flourishing of incredible grassroots movements that are applying an enormous amount of pressure to the way that the empire is trying to unfold its agenda. 80% of Americans in the most recent poll think that the AI industry needs to be regulated.

STEVEN BARTLETT: Yeah.

KAREN HAO: When was the last time that 80% of Americans were on the same side of an issue?

STEVEN BARTLETT: No. Yeah. When I have these conversations on the podcast, the comment section is clear.

KAREN HAO: Yeah.

STEVEN BARTLETT: There’s no disagreement. There’s no one in there going, “Oh no, I think they should crack on.”

KAREN HAO: Yeah. Dozens of protests against data centers have broken out all around this country, in the US, all around the world.

STEVEN BARTLETT: So what do we do about it?

KAREN HAO: So these are people that are doing something about it. They are actually reasserting their agency and exercising democratic contestation against the ways that the empires are going about their business.

STEVEN BARTLETT: What goal should we be aiming at? So if I said to my audience — Janet, at home — because this is kind of what I see in the comments, hopefully business — it’s like, what can I do?

The Path Forward: Accountability, Alternatives, and Collective Action

KAREN HAO: The goal is not that we completely get rid of this technology. The goal is that these companies need to stop being empires. And the way I define a typical business versus an empire is that the empires are predicated on this idea that they do not have to provide a fair exchange of value with the workers who work for them or the people who use them, or all of the other people that are involved in the supply chain of producing and deploying these technologies. They can exploit and extract and exploit and extract and get more value than what they offer.

Whereas typical businesses, there is a fair exchange. You buy a service, you feel like you got the same amount of value as the service that you provided. But for these data annotation workers, for example, they do not feel in any way that they’re being paid the same value that they provide to these companies.

So for me, the North Star is that we should be pushing back and holding accountable these companies when they operate in an imperial way. And that’s what we’ve seen with all of these people that are now literally protesting in the streets against data centers and having an enormous effect, by the way, actually stalling data center projects and also completely banning data centers from being developed in their localities.

We’re seeing that with artists and writers that are suing these companies for intellectual property infringement and creating a huge public conversation about what is it that we actually — how do we actually want to protect our intellectual property?

It’s like three weeks ago I met Megan Garcia, who is the mother of Sewell Setzer III, who is the 14 year old who died by suicide after being sexually groomed by Character AI’s chatbot. And she, when that happened, obviously was incredibly devastated by what had happened to her son. She also decided to do something about it. She sued the companies. And that lawsuit then sparked many other parents and families who were actually experiencing similar things to sue these companies as well.

That has created an enormous public conversation about what these companies are actually doing when they exploit and they extract. What is the cost to the lives of people around the world, including children?

What Can We Actually Do?

STEVEN BARTLETT: So what do you think my audience should do if they agree with everything written in your book, Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI? If they agree with everything said here, if they agree with everything we’ve discussed today, they’re concerned about their kids, they don’t want everyone to become data labelers, they don’t think that’s a particularly great solution — what can they actually go and do?

KAREN HAO: When I was writing the book, the only discourse that was happening was “this is the best thing since sliced bread.” Because of all of the actions of these people, like saying when they’re not happy with the things that these companies are doing, we now have 80% of Americans that want to regulate this industry.

And so I would say to people, think about all of the ways that your life intersects with the resources that the AI industry needs to perpetuate what they do, and also the spaces that they would need to deploy these technologies to continue having broad-based adoption in their work.

You’re a data donor to these companies. You could withhold that data. And that’s what those artists and writers are doing — they’re suing these companies to try and create mechanisms by which that data would then be withheld. You probably have a data center popping up around you. If you’re at a school environment or a company environment, you’re probably having a discussion in those environments right now about what the AI adoption policy should be.

And these companies — I was talking with some OpenAI employees just the other day and they were telling me that it’s understood internally that the revenue targets for the company are extraordinary and they need things to go flawlessly for it to all work out. And so they would need every single person to adopt this, every single space to adopt this. They would need to be able to build their data centers at the speed that they’re trying to build them.

And so what I would say to every one of your viewers is: let’s not make it go flawlessly if we don’t agree with what they are doing.

STEVEN BARTLETT: Ah, okay, I got you.

KAREN HAO: And then let’s build alternatives. Because the thing is, what I’m saying is not that these technologies don’t have utility. It’s that specifically the political economy that has emerged to support the production of these technologies right now is exacting a lot of harm on people. But we have research that shows that the very same capabilities could be developed with much more efficient methods, with much less resource consumption. And we have a lot of different other AI systems at our disposal that are like the bicycles of AI, that we also know provide extraordinary benefit at very little cost. So let’s break up the empire and let’s forge new paths of AI development that are broadly beneficial to everyone.

Holding Two Truths at Once

STEVEN BARTLETT: It’s strange. I think I’ve trained myself to deal with dichotomies in my head, and this for me is such a dichotomy. Where I, as a CEO and as a founder, as an entrepreneur and someone that loves technology — I think it’s incredible. It’s absolutely incredible, AI. It’s just so amazing and incredible the things it’s enabled me to do and create.

KAREN HAO: Yeah. Because it’s designed to enable people like —

STEVEN BARTLETT: — you. And my car driving in the morning and being safer. Incredible. I think the billion-odd people that use AI tools or ChatGPT or whatever it might be, they’d probably say that it’s added value to their own life. But — and this is the part that people find confusing — you can, and I invest in companies that are heavily using AI. But is it possible to think that is true and also think that there are significant unintended consequences which technology and the history of technology should have taught us to take a moment to pause and talk about?

KAREN HAO: I think this is absolutely — you can have both of these things in your head. And what I’m saying is that this tension doesn’t have to be a tension, because we could actually preserve the utility and benefits of these technologies, but actually develop and design them in a different way that doesn’t have all of these unintended consequences.

STEVEN BARTLETT: Yes. And I think there needs to be a big social conversation, which is why I have so many conversations about AI on the show. There needs to be a big social conversation about being intentional about the social and environmental impact. And that conversation is not being had in government. From what I can see, the conversation takes place in the industry, and actually trying to pull it out of the industry and open people’s minds to it is hopefully what we’ve been doing over the last couple of months with this subject.

KAREN HAO: I think it’s actually been happening everywhere outside of the industry, and for local governments and state-level governments there have been huge conversations about this everywhere. I’ve been on book tour, I’ve been to dozens of cities around the world. People are having these crucial conversations everywhere. I have not gone to a single city —

STEVEN BARTLETT: Yeah, it’s everywhere. Even here in South Bay. Yeah.

KAREN HAO: I haven’t gone to a single city where the room is not packed and people are not wrestling with the same exact questions as every other person in every other room that I’ve been in.

Closing Reflections

STEVEN BARTLETT: Speaking of packed rooms — I know you’ve got to go because you’ve got to talk today. So we’ve got a last question, which is the closing tradition on the show: how would your advice to a friend with a terminal diagnosis differ from what you would do yourself?

KAREN HAO: Differ from what you would do yourself. Oh, my God. I would tell them — enjoy, live life for yourself and take it easy. And I am not taking it easy.

STEVEN BARTLETT: Well, I think it’s a good thing you’re not taking it easy because you’re leading a conversation which is incredibly important. And I think that’s the thing — the conversation is the important thing. And because of algorithms and echo chambers, it’s so rare to have a conversation these days, especially a long-form one.

KAREN HAO: I agree.

STEVEN BARTLETT: Like this. So I think that’s so important. And your book — I think a lot of people would have learned a lot of stuff today, because I sit here and interview AI people all the time, and I’ve learned so much today from reading your book and the extensive, objective perspective that your book takes. You’re able to unravel all of these stories that we sometimes see in tweets and we don’t know if they’re true or not, because you’ve gone and met the people and you’ve done your research. And you’re an incredibly intelligent person, an extremely intelligent person who clearly has humanity’s interests as your North Star, and that shows up in everything you do and everything you say.

So please continue to fight in the way that you are, because it’s an incredibly important one. And it’s people like you that are, I think, galvanizing the world to take the collective action that we’re starting to see everywhere.

Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI by Karen Hao. I’ll link it below. For anyone that wants to read this book, I highly recommend you do. It’s a New York Times bestseller for good reason. Karen, thank you.

KAREN HAO: Thank you so much, Steven.

Related Posts

• FO583 Raj Shamani w/ Sam Altman – AI Revolution, Superintelligence & Future Of Humanity (Transcript) 
• AMD (Advanced Micro Devices) Q3 2026 Earnings Call Transcript
• Apple (AAPL) Q4 2026 Earnings Call Transcript
• Palantir Technologies (PLTR) Q3 2026 Earnings Call Transcript
• Jensen Huang: What Quality Will Help You Succeed in the Age of AI? (Transcript)