382. Empire of AI: Karen Hao On Today’s Hidden Colonialism (Ep 2)

23:02Anita:Well, I mean, one of the things that is a common thread is that when you read about these people in the press, there are certain stories that are, you know, very homespun, you know, like very close family networks sitting around a table, you know, busy, you know, sort of after school, parents who are very involved and then they sort of retreat into a garage. And, you know, you've got some shades of darkness and straight weirdness that are evident. I mean, the weirdness is I know his father died of a heart attack when he was very young, you know, sort of just I think 68-odd he was. And Sam decides to turn his father's ashes into diamonds.

23:39Anita:Now, that may be a story that people are familiar with. And for him, that's a symbol of, you know, sort of an enormous closeness, a way to process loss and grief and how, you know, much of it. But there is a dark story running and, you know, we should be careful about it. But his younger sister has accused him of abuse, you know, just saying that he's an abusive man. He's countersuing. So we maybe leave it at that. But it does sort of go along with the fact that all of these people, all of these sort of emperors of tech, you know, the broligarchy, they are very anxious to put out very wholesome stories about themselves.

24:14Whereas, you know, things are complicated. and in this case in this story very complicated yeah I mean one thing that's so interesting about Altman in particular because I've covered many other tech leaders and what's distinct about him is that even people that have worked with him for a very long time very closely struggle to articulate what he actually stands for like what does he truly believe they're not really sure and And I think this goes to why he has such a split. You know, some people think he's amazing and some people think he's the devil is because he has this remarkable ability to understand people and to persuade them to lend their talent or their money to a particular endeavor.

25:02So he's very good at turning his vision of the future into reality. but some people who do not actually agree with that vision of the future begin to feel that he he is manipulating them and tricking them along the way into serving a cause that they actually don't agree with this is the case of Dario Amadei where Dario left to found Anthropic and to be a competitor of open AI because he felt like Altman had somehow turned his intelligence, Dario's intelligence against the cause that he actually wanted to participate in.

25:41Anita:Right. And anthropic, we should say for anyone who is going, well, who are they? I mean, they're behind things like Claude, for example. Yeah, that's right. Sort of enormously successful AI that's out there. And the story of Silicon Valley is built on, you know, disgruntled workers walking out and making their own thing. 100%. Okay, let's just take a break there, Karen. And when we come back, I'd really love to talk to you about the parallels of quasi-religious missions, if I can put it that way, of AI companies and how they echo the civilizing missions of historical empires.

26:31Anita:recognize cancer cells. That's immunotherapy. One type is CAR T cell therapy. T cells are your body's watchman and doctors can now extract these and reprogram them to spot cancer. CAR T cells, they've started showing this real promise in treating cancers of the blood like leukemia, because unlike most cancers, blood cancer cells, they float around separately and that makes it easier for CAR T cells to seek them out and destroy them. And now Cancer Research UK funded scientists are designing CAR T cells that can tackle solid tumors too, making them tougher so they can survive a more crowded and hostile cancer environment.

27:08Anita:Now these approaches are experimental, but they do start to show where treatment might go next. And over the past 50 years, Cancer Research UK's pioneering work has helped to double cancer survival in the UK. I mean, this shows you how far research can take us. For more information about Cancer Research UK, their research, breakthroughs, and how you can support them, visit cancerresearchuk.org forward slash rest is science.

27:54First day fits, handled. Walk out ready for whatever's next. Visit your nearest Croc store today.

28:02Anita:This episode is brought to you by Facebook. So you were scrolling on Marketplace and there it was, the bike you'd been searching for. You sent a message and it turned out the seller was super chatty, kind of funny, and an avid cyclist. The next thing you know, you're in a cycling crew, well a community cycling group the thing about facebook you might find more than what you're looking for from a browse to a bike ride this summer find more on facebook

28:35Anita:welcome back karen so your your book opens with a line from sam altman um and it says the best way to start a religion is to start a company now was he joking or is this the truest thing he's ever said. I don't think he was joking at all. I think that, I mean, so this is from a blog post that he wrote and Altman was a very prolific blogger in the early days of his career. And so it is quite a treasure trove of the things that he was thinking about. It's not a hundred percent, you know, when you read the blog post, you should understand that Altman communicates to the public very carefully. He's always thinking about how he needs to motivate the audiences that he's speaking to.

29:18So it's not a direct correlation with what he's necessarily thinking or processing at the time. But I think this particular line where he's reflecting on the necessity of building a religion to mobilize the most people around a cause was him hitting upon something that he then used through the rest of his career, which is that you can get people to have a lot more conviction around particular technology and to mobilize more money and more resources than ever before if you can get them to believe in a myth, first and foremost, and a mission.

29:52Anita:A mission. So that's such an important word because that, again, it set bells ringing in my head because I thought, actually, that is exactly East India Company, VOC, you know, all these, the Belgians, for example, they start off with, you know, it's a purely economic venture. You know, they're going out, they're exploring, they're taking land, they're making money, they're sending it back. And then it becomes a civilizing mission. Yes. You know, whether it's a mission to spread Christianity or just civilization. And you're saying that is exactly the thing. And that's not just open AI. I mean, that seems to be something I hear from others, too.

30:29Yeah, absolutely. It's not just open AI. I mean, I don't think Altman was particularly clever in hitting upon this idea. I think a lot of the Silicon Valley elite had all hit upon similar ideas. actually in my book, I detail some emails that were exchanged between Elon Musk, Sam Altman, and Greg Brockman in the early days when they were thinking about co-founding OpenAI. And Musk explicitly articulates in these emails, we need to think carefully about how we craft our mission statement, because that is what's going to give us permission in the public to do all the things that we want to do. So the mission statement is both relevant for people in the public and also for then mobilizing employees.

31:10So it's an internal and outward facing mission. And yeah, that's how Silicon Valley operates. And that's how all of the empires of AI operate. They each have this civilizing mission. And they each claim that they are the good guys and everyone else is the bad guy. And that's why. They're saying they're doing the same thing, but they're at war with

31:31Anita:each other. And the war isn't just technological. I mean, it's now very nakedly political because You've got these two massive AI giants in the form of Anthropic and OpenAI who are pumping tens of millions of dollars into super PACs. Hundreds of millions. Hundreds of millions. Hundreds of millions and escalating. So, you know, super PACs are these organizations which are not meant to be connected to candidates. Candidate has no say on how they spend the money that they are putting into a campaign, but they inject money. So when it comes to these super PACs, I mean, how much money are we talking about, Karen?

32:08There are, by my count, four super PACs that the AI industry has set up. One is very deeply connected to OpenAI, although OpenAI denies it. One is deeply connected to Anthropic, and then there are two connected to Meta. And collectively, they have raised over 200 million, maybe close to 300 million now to pump into the midterm elections. Right. The way that these super PACs works is actually modeled after a previous set of super PACs that were launched by the crypto industry.

32:41Anita:And crypto industry just, you know, you might have had a Bitcoin, other coins do exist, but this is sort of it's money that you can't hold in your hand and put in your pocket. It exists in a digital form, just to make that clear. Yes, exactly. And a lot of the characters that are embedded within the crypto industry are the same characters embedded within the AI industry. So suffice to say that they are both a collection of Silicon Valley elites that have extraordinary amounts of money and then seek to deploy that money in various ways to bend society to their will. And the crypto industry was the first faction of Silicon Valley that decided to much more explicitly play politics and deploy their financial might in elections.

33:22And they spun up a super PAC that similarly deployed around$200 million into the elections. And they were very naked about how they were going to use this money. It was if a candidate is pro crypto, we give them money. If a candidate is anti crypto, we use money to block their candidacy. And this was the brainchild of a man named Chris Lehane, who is now the head of global public policy at OpenAI. And he's considered the master of disaster because he was formerly working for the Clintons and helped usher the Clintons through a series of scandals. And then he hopped over to Silicon Valley, joined Airbnb, then joined the crypto industry and then joined OpenAI.

34:05And so because the crypto super PACs, they claim, were incredibly successful at getting a slate of pro-crypto legislators into office and then in passing pro-crypto legislation at the federal level. The AI industry is now taking that exact playbook. And in fact, one of the main super PACs, which is called Leading the Future, which is the one that is deeply intertwined with OpenAI, it has funding from Greg Brockman, the president. Chris Lehane was also behind it, literally has the same cast of people running the super PAC as the former crypto super PAC. So all of these AI super PACs have stated that the game is they're going to find the pro AI candidates and the anti-AI candidates and boost or tank their campaigns, except for Anthropics super PAC, which interestingly is trying to play this counter narrative game where they are deploying a significant amount of money to, they say, to boost people, candidates that actually want to regulate AI.

35:15because Anthropic's whole narrative is that they're pro AI regulation, but this is a narrative and it should be deeply scrutinized because at the end of the day, the way that Anthropic deploys its money is exactly the same as OpenAI in that they're ultimately supporting the candidates that align with their agenda.

35:31Anita:Well, so, I mean, again, this is something that is truly an echo of what we've seen go before in the nation empire. So with the East India Company, you had people who went and made a lot of money through the company overseas and came back and bought themselves, you know, positions, political positions, you know, rotten boroughs, you know, they could become MPs, they could. And then you have a full chorus of people who are lording rather than questioning the behavior of a corporate entity. And that feels like what you're saying is happening with open AI, although, you know, they're fighting each other.

36:06Anita:They're also fighting for territory, political territory, as well as, you know, sort of technological and financial territory. Can we talk about, if we're going to carry on this empire analogy, I want to know about the people at the other end, because you, very interestingly, just haven't looked at the top of the pyramid, but you've looked at the workers in places like Kenya who make these systems usable. Now, first of all, that's going to surprise people that, you know, these are organizations that have workers in the developing world or the global south who are, I mean, are they benefiting? What kind of work are they doing?

36:40Anita:Tell us a little bit about that, Karen. How are they associated? Yeah, one of the myths of the AI industry that allows them to perpetuate this idea that AI is this magical technology is they say that these systems just learn and do everything that they can by themselves. But actually, anything that an AI system can do was taught to it by humans, by data workers who are manually cleaning, preparing, and content moderating the data that is used to train these models. and you know some of that you know the fact that chat to beauty can chat is because there were tens of thousands of workers that were prompting a raw large language model right after it was trained on just a bunch of data and getting the model to start speaking more conversationally based on examples of this is what human conversation looks like speaker a talks then speaker b then speaker a again so i mean does that what does that actually look like is that sort of like, you know, you'll have a big warehouse full of people on terminals saying, you said this, but actually, this is more human.

37:48Anita:If you're going to do this, do it this way. Is that the kind, I mean, I'm just trying to picture it in my head. And I'm sure people want to picture it. Yeah, yeah. So so it comes in different forms. Sometimes these workers are working in offices, but most of the time, they're actually working from home. And they are remote workers that are distributed in largely impoverished communities. And they're essentially just given tasks from, they often don't know which company that they're working for. They're just given a task by a platform that's sort of like an Uber for data work. And the task will just give them very detailed instructions saying, we would like you to write a transcript of dialogue between a father and a son.

38:30Like imagine, imagine what a conversation would be like. And then they'd literally write almost as if they're a playwright. They write some dialogue and then it's just sent into the ether and they have no idea what that was for. But the company is then receiving all this stuff and then they're putting it into the models saying this is examples of dialogue. This is another example of dialogue. And then there are other workers that they'll send a task and the worker will just get instructions that says, write a question that you have on your mind and you will receive an answer and you should upvote or downvote the answer based on whether you think it was a good answer.

39:06And, you know, when, when, before ChatGPT came out, like the workers who were doing this had no idea that they were doing this for an AI model. There was no conception of having a conversation with a computer. And so the instructions that they received were just very vague and they had no idea what they were actually contributing to.

39:24Anita:And may even have sounded ludicrous, like have an argument with your girlfriend, to tell me how that would sound, you know, that kind of thing. Yeah, yeah, exactly. But I suppose somebody listening to this might say, okay, that's fine. But that's employing people. You know, that's making people wealthy. There's so many people, if there's so many people that are involved in this, particularly from, and I think the places that you've talked about in the book, Chile and, you know, Kenya, for example, they're making loads of money. This is great. You know, what's to complain about? Are they making lots of money?

39:52They're not making loads of money. But But before we get to that, I mean, I've been covering these workers for seven years and the way that the work is structured makes it impossible for these workers to actually live a normal life because they just have accounts on these platforms and the platforms will send them work whenever. It's not a nine to five job. You don't know when the work is going to arrive. And in fact, you compete with the other workers on the platform to claim a task to even get the opportunity to do that task. So I tell the story of this one woman, Oscarina Fuentes Zanaya, and she's a Venezuelan refugee based in Colombia.

40:30And she was the first worker to ever show me how this structure completely disrupted the rhythms of her life because she didn't know when the work would arrive. And then she had to snatch the work the moment that it arrived. She was tethered to her computer day in and day out. There was one time she took a walk outside and she got a notification on her phone that a task arrived. She sprinted back to her apartment in the hopes that she would be able to claim the task before another worker did. And she got home too late. And that task could have paid her enough for her groceries for the week. And she vowed after that to never go on a walk again.

41:09Anita:Oh, God, really? Yeah. And she eventually figured out that the tasks only arrived during the weekdays. So she went on walks, 30-minute walks on the weekends. But I mean, this is deeply, deeply controlling work where you cannot, you know, I spoke with another worker in Kenya. Her name is Winnie. And she would work for 22 hours straight once she received work because she thought if I go to sleep and I wake up, the work might be gone tomorrow. And then I would have missed the opportunity to earn just a few extra dollars to put my kids on the bus for school. Few extra dollars. Okay, I mean, just let's talk about that for a bit, because, you know, these are enormously wealthy companies.

41:53Anita:I mean, they're, you know, richer than creases, eye-watering amounts that they can put hundreds of millions into their political, you know, persuasions as well. You know, how much is Winnie getting paid? How much is your Venezuelan refugee? How much are they making for this chunk of their life? It dramatically varies. I mean, sometimes they're making just$2 an hour. Sometimes they're actually making, you know,$100 for the day. But this is kind of the challenge with this work is that you don't know. You just don't know because it's dictated by the company that gave the task and also the platform, the Uber-like platform that's distributing the task.

42:35The price is set based on what they think is the willingness of the worker to do the work. And so that becomes a problem because they're seeking the most impoverished and most desperate parts of the world to do this work so they can squeeze the worker by paying abysmal amounts of money sometimes. But it's, you know, it's ultimately not just about how much they earn. It is it is about just the complete lack of ability for people to do any kind of financial planning to actually know how much they will end up with at the end of the month. because one of the things that these companies will use as a defense is that now increasingly they are employing workers in developed contexts.

43:20So the U.S. data annotation has become the fourth fastest growing job in the U.S. this year. And they're also employing people in the U.K. across Europe. And these companies will say that we're actually now paying, you know, $50 an hour for these workers or sometimes$200 an hour for these workers. And still, when you speak with those workers, and there was a survey that was conducted of workers in North America that are doing this kind of work, an extraordinary share of them have dealt with homelessness, are chronically ill, and default on their bills at the end of the month. because even if they're being paid, it seems like a flashy number and an hourly wage.

44:04They don't know when it's going to arrive. They end up working very, very, very long hours to just cobble together enough. And at the end with taxes and with their medical bills and with all these other things, they end up still not having the basic living wage left over. So there is both an extortion that happens with the pittance that sometimes workers are paid, But even when they're paid supposedly well, it still does not add up to something livable.

44:33Anita:So the other thing that goes hand in hand with empires of yore, particularly the ones that we talk about, is acquisition of land and extraction of, you know, sort of power, water, you know, whatever the natural resources are. Now, is that happening as well? And I know sometimes it's you yourself have had some difficulty in this arena of, you know, sort of pegging numbers to this. But I mean, do we have an idea of, for example, freshwater? Let's talk about the most basic commodity. Are these things, are they thirsty, these data centers that are needed to run these corporations? Yeah. So the thing about the environmental impact is it stems primarily from the vast computing infrastructures that these companies seek to build.

45:18The reason why they build such vast computing infrastructures is because they base the advancement of their AI models on needing to train it on ever, ever more computing power. And they try to multiply the amount of computing power that they use for each subsequent generation of a model by an order of magnitude each time. And so they are acquiring vast tracts of land to build these facilities. And they need an extraordinary amount of many different types of resources to power and to cool these facilities. So when it comes to fresh water, the reason why these facilities, many of these facilities need fresh water is twofold.

45:57One is because of the sheer amount of energy that these facilities are using. A lot of energy is produced through the consumption of fresh water. This is true in the U.S. It's true in many parts of the world. There is a relationship between energy and water. And the second thing that these facilities need is to cool themselves down. And many facilities also use fresh water to cool. And the reason why it has to be fresh is because the facilities need the water to be clean enough to not leave behind residue that then corrodes the equipment over time and makes it not functional. The honest answer is we don't have good numbers on how much freshwater these data centers use because these companies are completely untransparent about those numbers.

46:46In fact, there was recently reporting out of Europe where investigative journalists from Investigate Europe found that the EU Commission passed a directive to preserve the secrecy of the environmental impact of these data center facilities because companies like Microsoft and other big tech giants lobbied the EU Commission and essentially got the EU Commission to draft this directive almost word for word based on the company's decision. interest saying we can't release these numbers at a data center level because it is proprietary

47:23Anita:information. Yeah. Proprietary and confidential. Proprietary information. Yeah. And so, you know, in my book, I was able to obtain some numbers for some facilities, but they were based on projections of reports that the company gave filed in government filings saying like, we believe that we are going to use this amount. And oftentimes, these projections are not even accurate once the facility is actually built and starts operating. There was a case in Aragon, Spain, where The Guardian found through internal document leaks, that Amazon was actively trying to hide the amount of water use of that facility from the public as they sought to also significantly increase it over time.

48:08Anita:We've got, you know, unfortunately, a finite amount of time. There's so many things I I want to ask you. We could talk for ages. I find this completely fascinating. Can we talk about something that's a physical object that is sort of at the heart of tech? And that's the microchip. And this is where geopolitics I find really interesting because 90 % of the world's most advanced chips are manufactured in Taiwan. Yeah. Right? Now, Taiwan itself is in the crosshairs because China says it belongs to them. America says, hands off, it does not belong to you. It is a flashpoint every time there is a movement in the South China Seas.

48:47Anita:Everybody holds their breaths because could this be the thing that triggers us or tips us into a world international conflict? So, I mean, just how is it that first of all, 90 % of the world's most advanced chips are made in this one island? I mean, how did we come to this? Semiconductors are so incredibly complex. They are honestly an extraordinary technology because they're so complex and because of globalization, the supply chain for producing these things is all over the map. And it's not just the fact that Taiwan produces most of the chips in what they call fabs or like the factories for these chips.

49:23Also, there's a Dutch company called ASML that is the only company in the world that makes a particular kind of machine that then the Taiwanese semiconductor manufacturing company needs to acquire to produce the chips. So there's actually many bottlenecks on the supply chain where there's just one company or one geography that actually has this outsized share of the piece for that particular part. And Taiwanese semiconductor manufacturing company is it's a it's a company that was started many decades ago to produce these technologies. They were able to advance through a lot of R &D, their ability to, in particular, etch the wafers.

50:05So when you build a microchip, you have these silicon wafers, these incredibly thin sheets of metal that you are etching to get the chip to have all of the processing power that it can do. And the reason why our computers have gotten smaller and smaller is because the ability to etch these things has become more and more precise and more and more delicate. And literally, some of the transistors that are on these wafers are now three nanometers apart. So that's, I don't know how much a hair is. It's probably less than a hair of distance. and TSMC, this company, just became particularly good at that.

50:49And other companies that produce these chips are just not as good at that. So that's why this particular company, and therefore the island of Taiwan, has this outsized influence on the most advanced computer chips.

51:04Anita:I have an answer for you. I've just looked it up. A human hair is approximately 80 ,000 to 100 ,000 nanometers. So if you're saying, you know, I didn't even realize that. Yes, you should. You should know. But, you know, this idea of sort of all being concentrated in one place has all of the technology that these huge state actors are interested in. Again, that just makes me think of very recently did a series on nutmeg. And, you know, the kind of, you know, it was in one place that you could get nutmeg. And so you have these huge, the Dutch are after it, the British are after it. It feels like these things echo.

51:40Anita:So, so distinctly, just as we come to the end of this, there are going to be people who are saying, you know what, you lot are so miserable and you don't see that this is the next stage in evolution. You're a bunch of Luddites. You're being really stupid about this. AI is going to make everybody's life better. And these are the people who took the risk. These are the innovators. They're the ones who put their homes on the line. They put everything on the line. And this is the next leap forward. What about the happily ever after in this? Why are you convinced? And I can see from your smile, you're not convinced.

52:14Anita:Why can this not lead to a happily ever after? Well, what I will say, so going back to this analogy that AI is like transportation, I am actually quite bullish about certain types of AI technologies. What we are critiquing and analogizing to empire is a very particular type of AI system and the political economy that it has birthed to produce this system. And that is, you know, those are the large language models. Those are the corporations that are trying to accelerate these large language models and scale them ever, ever more. But there are other types of AI systems, which I call the bicycles of AI, that I want more of.

52:51I want more specialized, efficient, sustainable, democratically governed AI systems that can help advance health care, educational outcomes, actually mitigate climate change instead of accelerate climate change. So I do actually believe in the power of technology to bring us to a better world. But we have to ask which kinds of technologies and who controls those technologies. And ultimately, what I think will unlock broad-based benefit out of AI is if we pick the bicycles of AI instead of the rockets, so much more cost-efficient and much more sustainable systems. And we also design governance structures where anyone who is impacted by a particular AI system actually has a seat at the table in holding the developers of that system accountable to their interests.

53:44That's just not what we have right now. We have a governance structure that echoes empire where just a few people get to make the decisions about how these systems are constructed and deployed in the world.

53:54Anita:And I promise you, this is my final question, although I promised you I could ask you a million more. With the case of the East India Company, when the mutiny happened in 1857, it was deemed it was too big to fail. And you had the government step in. Now, is this sort of landscape that we're in, where AI is the new empire or different empires in that landscape, are they too big to fail now? I love this analogy because it also has helped me think about what could happen next. And I think these companies are trying to make themselves too big to fail because in the same way that the East India Company was actually deeply vulnerable, even as it seemed at the height of its power, I think these companies recognize that they are deeply vulnerable and they do not have actually a long term ability to sustain the sheer amount of spending.

54:46Anita:Vulnerable how? I mean, what are the vulnerabilities that you identify? So OpenAI has committed over a trillion dollars of spending to building out the computing infrastructures that they believe they need to train the next generations of their models. They have only been able to generate tens of billions in revenue. That is an extraordinary balance sheet, and it is not unique to OpenAI. All the major AI players have this kind of balance sheet when it comes to their AI development. They do not have a business model. And so they have been trying to raise money to plug this hole while they figure out a business model from wherever they can.

55:26That's why, you know, you see extraordinary numbers about OpenAion and Anthropik raising massive, massive amounts of capital from private markets. That's why many of them are now preparing to go to IPO. I mean, SpaceX already went to IPO, but OpenAid and Anthropik also have announced their intentions to go to IPO because they've tapped out private markets and now they need to get money from public markets. Meta uses creative financing methods to try and get out an extraordinary amount of debt and also then creative accounting methods to try to hide that debt on their balance sheets as a public company.

56:01So all of these companies are unsustainable in the financial sense of the word and are trying to figure out how to make themselves sustainable financially before it's too late. But they recognize that if they don't figure that out, the entire company falls apart. The entire AI industry falls apart in the same way as the British East India Company. And so this is why I think these companies are dealing in more and more circular financing. They are intertwining their fortunes with one another because they're trying to become too big to fail such that when they potentially wobble financially, the government does step in to bail them out.

56:50Anita:well look this has been quite the ride thank you karen i really appreciate it karen has book um yes i mean you're not going to sleep very well after it but you know it's a great read the book is empire of ai really delighted to have you on thanks very much for listening to empire if you'd like early access ad free episodes or our full reading list for this episode just join the empire club at empirepoduk.com that's empirepoduk.com next time on empire we're going in a very different direction our series takes us to ancient ireland and ancient scotland at the time of picts and druids and vikings till the next time we meet it's goodbye