The Conversation On AI I Wasn't Prepared For... - Tyler Cowen

There's a war going on behind the scenes. Claude AI can now penetrate American classified government information in moments, in hours. They can do that, they could penetrate all our information.

The next 6 months, year, few years, they're going to be pretty hairy. The cyber attack capacity is out there, but the defense has not been done yet. We don't have enough of a sense of urgency.

Can the cyber defense ever be ready?

There has to be a hole in the system somewhere, right? The attacks are coming no matter what. You have to be ready with the best defense possible. The United The United States in particular is very lucky that it has these two companies. For the United Kingdom, it's a question of, "Hey, do we have access to the defense? On what terms?

Are you going to screw us over?" When the government and big corporations start working together, that is a lot of power.

We need to find a new balance of power. There's a sense that with AI, the future of the world is up for grabs like it's never been before.

AI is in the hands of very few. It could well bring in a whole new totalitarian age. It's a weapon of the proportion of the nuclear bomb.

It's an extreme situation, and we're in it again. The public is not generally aware of that.

Do we know how far off the Chinese are?

I have spoken with experts who estimated—

Tyler Cowen, thank you so much for taking time from your precious schedule to sit down with me. I want to ask you about AI. You've been writing about this.

You've written about it in your new book, Marginal Revolution: Its Rise, Decline, and Its Future with AI.

There we go. I want to start our conversation with a war that's, all the focus has been on the Iran war over the last 6 months, but there's a war going on behind the scenes, another war that Trump is waging, and that is on Anthropic. Anthropic leading the way with AI and their Claude AI. And what really struck me is the idea that Claude AI can now penetrate American classified government information in moments, in hours. And so this is actually a threat. I mean, if they can do that, they could, penetrate all our information, and it really started to get me a bit nervy. But people who are not aware of this war, perhaps you can lay out what's going on and its significance.

For the last 4 years, you can say we've had very powerful AI models, much more powerful than what we had before. At first, you would just chat with them. Now they can do more complex and longer tasks that take a good bit of time. And one of those tasks is hacking into other people's systems. But I think the important point is they also do cyber defense. So cyber hacks are hardly a new thing. They happen all the time across governments, into hospitals, ransomware. We all know this. So what current AI allows you to do is to test your system, red team it, learn where the holes are, and actually patch them. So my view is this is a good thing overall, but the next 6 months, year, few years, they're going to be pretty hairy. Because the cyberattack capacity is out there, but the defense has not been done yet. I think the important thing to do is to give people the AI so they can build their defenses, but they really, really need to hurry. We don't have enough of a sense of urgency. And that's one way to frame what's going on between Anthropic and the US government. Anthropic and OpenAI and Google, they've built something quite wonderful, quite amazing. But at the same time, there's this timing problem. The cyber defense isn't ready yet, but it's the key to the cyber defense as well.

Can the cyber defense ever be ready? Presumably, whatever they create in defense, there can be a now more powerful AI that will be created to penetrate it.

But there has to be a hole in the system somewhere, right? So another way to look at it, China has AI. There's this thing called open source AI, which you could say is not controlled by any central authority. It's just out there like open source software that's coming as well. So the attacks are coming no matter what. You have to be ready with the best defense possible. The United States in particular is very lucky that it has these two companies, Anthropic and OpenAI. For the United Kingdom, it's a question of, hey, do we have access to the defense? On what terms? Are you going to screw us over? You know, can we trust President Trump? Very difficult issue for the UK and the Europeans, but the attacks are going to come. So I think the next 5 to 10 years will be quite bizarre. In the medium to long term, we'll be much better off, much wealthier, much healthier because of AI. But again, this asymmetry of speed between the offense and defense, it's going to make for a very strange world.

Okay. Can you put me in the mind of both the people running Anthropic and Trump himself or the administration, whoever is advising him in this standoff.

Here's another very general change that's been happening. It goes above and beyond just Anthropic or OpenAI. Private companies have much more powerful tech than they had 10, 20 years ago, much more powerful. Not only that, they pay these huge salaries to get the best workers. And their speed of innovation is extreme and incredible. Governments cannot match it. So whenever there's a negotiation between a private tech company and a government, the company is in a much stronger position than it used to be. It's not like Boeing or Lockheed, which are themselves slow-moving and bureaucratic. It's not unusual in the AI world to recruit a top researcher for $100 million. That's crazy, right?

Wow.

Government can't do that. So government relies on the top AI companies. It wants to regulate them, but the companies want to keep some degree of autonomy over what they do. And there's been this standoff between the Trump administration and Anthropic because Anthropic trusts itself more than it trusts the Trump people. The Trump people feel the government, the Pentagon, needs to have the final word. They're both reasonable positions, but they're not mutually consistent. And so we're in the middle of this standoff. And as we're recording, the very best anthropic model that had been released, Mythos, or the derived version called Fable 5, right now it's off the market and we're waiting to see what will happen.

It's been put off by Trump, right?

It's a little complicated. So Trump issued an export directive, which said you cannot send this to other countries. You cannot let citizens, non-US citizens use it.

I mean, that's the right thing to do, presumably.

Well, it can't really be enforced. So if you say, oh, no foreigners, like what if you're in California and you just log on? How do we know that you're, well, you're a UK citizen, right? How do we know that?

Oh, so implied is everyone has to have a digital ID or?

And we don't have that. So what Anthropic had to do to meet the order was to pull it from everyone. So there was an overreaction, but there was no other way to enforce what all of a sudden was the law. But without Fable 5, we can't build higher quality cyber defenses. So a few parties still have access to a version of this, but I would say we shot ourselves in the foot. Like it's an understandable decision from the point of view of government. Like we should control this. It's dangerous. No foreigners. Okay, I get that. But if it means it's gone forever and the US can't build up its own defenses, we've cut off our nose to spite our face, to use the old saying.

But it's a bit of a catch-22 because you allow it to stay on and the government runs a risk of being infiltrated and penetrated.

And that will happen. And not just the government infiltrated, but you as a private citizen or a company.

Yeah, exactly.

There could be all sorts of infiltrations.

You run the risk, but if you don't allow it out, the defense against it can't—

And China and other parties will come for you 6, 9 months later. So I would rather take the chance.

It's lose-lose though. You'd rather take the chance of having everything, it gets into the wrong hands even within America and the whole system comes down, right? I mean, and that's what we're talking about.

I think we have to take the chance. I wouldn't say the whole system comes down. The degree of damage is uncertain. And we don't know how swiftly adversaries will use it against what. You know, there's a lot of possible terror attacks that without AI could occur any day in London. And there's evil terrorists around. They don't do it every day. We're not sure what accounts for the lags, but exactly what would happen, how quickly, we're just not sure. But it, it's a very real risk. I wouldn't deny that. How long did it take the terrorists to figure out that box cutters could fly two big planes into the Twin Towers, right? It didn't happen right away.

Well, firstly, the— there are big restrictions put in on us, like for— well, basically we have a police and, you know, there's considerable restrictions to try and prevent. We have a whole task force to try and prevent these sort of Islamist operations going on. But it's more like rather than them having a backpack that they can explode, they've got a nuclear bomb. Like, that seems like a— in the situation of the amount of destruction that could be caused by this weapon, it's not something to just— at least it's a— surely it's, uh, something to be concerned about.

It's something to be concerned about, absolutely.

But you're not that concerned?

No, I'm very concerned, but I'm concerned we're not hurrying to build up our defenses. That's what we need to do.

The Winston Marshall Show will be doing its first ever live taping, and I want to see you there on September 11th at the Emmanuel Center in London. We are coming, and boy, do we have a show for you. Journalist and author Majid Nawaz will be sitting down with the historian Raymond Ibrahim to debate the question: Is Islam compatible with the West? Perhaps one of the most important and difficult questions of our times. It is without question a clash of the Titans. The Winston Marshall Show, live in London, September 11th at the Emmanuel Center. The link for the tickets is in the description. For more information, go to winstonmarshall.co.uk. Otherwise, I'll see you on September 11th. Okay, and who should be building the defense then?

Well, in the United States, we need a concerted government effort to coordinate with Anthropic and OpenAI to decide who gets this first What are the most sensitive pressure points? So for London, I would say the banking system, but of course, your spy service, your national intelligence, parliament, the list I think is reasonably obvious, but there can be subtle pressure points that matter a great deal that you don't see at first. And we need to get cracking on this right away. And the UK in particular needs a sustainable deal with the Trump administration, which I get is hard to do, right? Things are always renegotiated where you have access on acceptable terms. And again, if we don't do anything, just China has it all in less than a year. And what's so great about that?

Do we know how far off the Chinese are?

I have spoken with experts who estimated 7 to 12 months. It's an estimate. It could be off by a little, but it's not off by a lot. Well under 2 years, I would say.

Mm-hmm.

Could be less than a year.

And they're presumably watching what's going on here very keenly.

Of course, but they're scared themselves too. So they're having their AIs built by private companies. Their government is more interventionist than ours in either of our countries. But they wonder, well, does the company become more powerful than the Chinese Communist Party?

Right? I mean, are there any private companies? The CCP is in place within all companies within China.

Absolutely.

They're not really private as we understand it.

They're not private as we understand it, but the CCP can't simply shut them down. And that's because the private companies now have this power. Right now we're infiltrating China with our mythos, which the Trump people presumably are using. I hope they're using. And China needs its own companies to build its cyber defense. So China just can't say to those companies, you're all going to jail, we're shutting you down, it's over, because then they're defenseless against us. So it's like an arms race, but the companies are a new power relative to the government in a way we have not experienced before. And there aren't many companies that can build this. So like an earlier weapon, guns, gunpowder, how many companies can build guns and gunpowder, right? Hundreds, thousands. But now in the US, there's two companies that have AI much better than anyone else.

Right.

And that's a highly unusual situation. And we don't really have the tools to handle it.

And those two companies are OpenAI and Anthropic.

Correct. And Google would be not a distant third, but a third. And part of Google DeepMind, of course, is here in London.

You mentioned in passing that— did you say Mythos is operating in China?

Well, I'm speculating.

That's speculating.

Okay. But the US government has access to Mythos. I certainly hope they're using it to do good things for my country's geopolitical positioning. I don't know that for a fact at all. It ought to be secret. Indeed, it is secret, but I would be surprised if they were sitting on it and doing nothing.

Okay, so then as to what happens within America, this, if you are to lay forward a proposition of a good way forward, at the moment we've got this clearly then a not happy relationship between government and the AI companies.

Right.

You want to see them working together on this, presumably?

Correct.

Okay.

And they are, but I'd like to see it go a bit better than it has.

And what would be the model of work here? I mean, is this kind of like NASA working with SpaceX, or is this— because it's in danger of getting into sort of fascist territory, right? When the government and big tech companies and big corporations start working together, that is a lot of power then accumulating into the hands of the very, very few. Is that, how exactly do you see this playing out in those models?

Those companies are indebted to consumers to buy their products. They're indebted to capital markets. They're indebted also to the law. There's state regulators in my country. And federal regulations, but they're potentially very powerful actors. And we need to find a new balance of power because they have unique products in a way that old-style defense contractors did not. And there's a speed issue here. So if someone's making tanks and tank production is postponed for half a year, I don't know, maybe it's a big deal, but maybe it's not, right? Whereas with drones, AI, cybersecurity, the speed of change is so fast that if something is held up for 6 months, it is a big deal. And we haven't been in that position since World War II, I would say. Things like codebreaking in World War II, like you had to have it now. So we've lived a version of this in the past and you just had to have everyone working together very rapidly with an extreme sense of emergency. Oh, what's the weather going to be like for D-Day? We pulled that off. Our countries collaborated and there's not even really time to write out all the rules. You just have to do it and get it right. It's an extreme situation and we're in it again. The public is not generally aware of that. I would say I'm not pessimistic here. I think US has excellent companies. We have an unorthodox administration, but I think on this issue they're quite aware. I don't like all of their decisions by any means, But I think on AI, they've been a pretty good administration.

Despite this conflict between—

They could have handled it better. I view their actions as a mistake. But I would just say in general, it's an area where they've had expertise. And I wouldn't say that about the current US administration in every area.

Right. So what would be an example of them getting it right?

I think intermediaries are needed. Does that convince Anthropic you need to speak the language of the Trump administration more, which would involve a certain amount of humiliation because ideologically they're quite different from the Trump people?

Is it true that they're woke? I think Trump might have said that, or maybe Hegseth said that, these woke overreaches. Is there anything to that?

I don't think that's the exact correct descriptions. Some of them are woke, but I would say they're what are called effective altruists. Even though—

Defective altruists.

No, effective altruists.

I like defective altruists, maybe more appropriate.

So there's a sense that with AI, the future of the world is up for grabs like it's never been before, and the stakes are very high, and we really have to make all the right decisions, and this is an incredibly important moment. I agree with all of that, but I think it also can become a bit too rigid of a jacket. For how you view the world. So Anthropic can be described— they wouldn't say this about themselves, to be clear, but a company very influenced by effective altruism, a kind of obsession with AI and its risks. The Trump people, I mean, they're a mixed bag, but they have to run an administration, frankly, with national security interests, and they feel an extreme sense of emergency. And they're very concerned with exerting their ultimate control over the private companies. But they need to realize a bit, they're not currently in a world where they can exercise as much control as they would like. Anthropic needs to understand they're not in a world where they could exercise as much control as they might like and cut a deal. Now they're talking about doing this all the time. Again, I'm not pessimistic. And by the time this episode comes out, a deal may be in place. And then there's this other party, OpenAI, which I would say is less ideological than Anthropic, more of what you might call a normal company. They've had better relations with the Trump people, but it's still a question because their next model, when it comes out, likely will have properties similar to those of Mythos. And what will our government do with that model? Do they allow it to come out? Who gets it? Who gets it first? Which other countries get it? Under what restrictions? Very complicated questions. We've got to figure it all out on the fly. It's like Alan Turing and the code breaking during World War II. We'd better get it right, put our best people on it. And to sit down and say, oh, parliament debate, the US Congress, this, that, I don't disagree, but the clock is racing faster than any of that will happen. We have to get it right now.

On the effective altruism thing, so it might be worth just quickly unpacking that because not everyone might appreciate what's wrong with effective altruism. What do you see as wrong with effective altruism? Maybe you can define it for us as you would.

Effective altruism, I would say, is best defined sociologically by knowing the people who believe in it. If I state the doctrine in the abstract, which I'll do in a moment, it sounds very neutral. I agree with a lot of effective altruism. The point is you should always be obsessed with what will do the most for human welfare. So that means, for instance, you should not give a donation to Harvard when you could send it to starving children in Africa. That may well be correct, but I would say effective altruism, it's also a Bay Area group of people, fairly tightly knit. Its critics would call it a cult. I don't think that's quite fair., but maybe it has some properties of a cult. And effective altruists can in practice, once you know them sociologically, develop obsessions over things which are not unreasonable as concerns, but once they turn into obsessions, perhaps become harder to manage than needs to be the case.

The famous examples I've heard about effective altruism is the kind of test is there's a burning house and inside there's the Mona Lisa or there's a baby. Which one do you save? And I think effective altruists would say the Mona Lisa because that will do more good for the future of humankind. As a Christian, I would obviously choose the baby. So it's a non-starter for me. And then another example is that you hear of effective altruists trying to puzzle, is it, we have to think about not just humans alive today, but humans alive in 100 years and what will do the most good. Again, as a Christian, it's like you prioritize the human that's alive now because we don't know what it's going to be in 100 years. So it's a whole new ethics system.

Yes. And the questions become very hard to manage.

Right.

Say you're a mother, you can sell off your baby, receive a lot of money and save the lives of 5 babies in Africa.

Mm-hmm.

Are you obliged to do that? I would say no. I think you would say no, but it's not clear given the basic doctrine why you shouldn't do it. So again, it can develop into these obsessions, which I think are unhealthy. Or just they make it harder to get the question right, that common sense has to kick in sooner than actually happens with the effective altruists. Now, I wouldn't say by any means Anthropic is pure effective altruism. It's just a big company with a lot of people, but there's an effective altruist influence on their decisions, which says correctly, AI is the super important thing. You really, really need to get all your decisions with it right. And I completely agree with that part of it. But somehow when it gets translated into action, I would say culturally anthropic comes across to the Trump people as just too far from how the Trump people view the world.

Yeah. So how is that actually, how are they jarring with each other? It's like, so what if they're too far? What are the fault lines or what are the points where this effective altruism, what's an example of it clashing with Trumpism or the worldview of the White House administration?

Well, a very simple question for any system, not just these, is who has the right to pull the plug? So the government could insist under law it has the sole right to pull the plug, but in actual practice, for the system to work, the company has to maintain it, and the company ends up with plug-pulling rights no matter what we write into the law. And that's a difficult situation, I would say, for government to stomach, that a private company with a thing potentially so powerful and important, uh, can influence the terms on which it's used. Who gets faster service, for instance? Even if you think the government ought to have the final say, you just can't write all that into law in a simple way and enforce it. In practice, the company is going to have a lot of say whose questions get answered first, or who gets the tokens first, and whose system gets repaired first if there's a problem, and so on. So again, the notion that the government is now more dependent on companies than what it's used to, and we all need a cultural reorientation. We're far from having pulled that off.

I think the first time that I got really spooked by AI was this idea of mass hacking and that nothing of anyone's will be safe anymore. All our bank accounts, the thing is pretty much everyone's online now, everyone's using their phones. All of that could be hacked. I mean, and this seems like it's a weapon of the proportion of the nuclear bomb. And it might not be that it, you know, it's going to be like Hiroshima necessarily, and this would bloodshed, but the power of being able to wipe everyone clean, like wiping everyone's assets completely clean across the world in the wrong hands. I mean, that's unbelievably dangerous. And so, I actually comparing it to nuclear arms for me is not a crazy thing to do. I think it's quite reasonable. I think it's quite reasonable for the administration to be worried.

They should absolutely be worried, as should the UK administration. But I would make this point that may have been possible already, that kind of hack. And you don't have to wipe out every account. You only need to wipe out some and freeze the transfers. You know, who has how much where. Once that's uncertain, The system freezes up. It hasn't happened yet. Without AI, it might happen someday. And finally we have a defense, but we have this intermediate period where the risk of it happening has gone up. Hmm.

Okay. So how do you see things unfolding then in practice?

So a week ago everyone thought there would be a quick deal. People, also the prediction markets, no longer think there'll be a quick deal. They see the deal now as quite complicated. Most parties still think there'll be a deal. There has to be a deal. No deal is not an option, but it could be something late into July. The terms of the deal are uncertain. JD Vance has said he thinks the US government should own some share of the two main AI companies, maybe 10%. Bernie Sanders thinks the government should own half. I think that's a disaster waiting to happen.

Why?

Our government doesn't know how to run those companies. And the mere fact, like I'm a shareholder in different companies, right? As are you. You have no idea what's going on, just holding shares. And you politicize everything the company does. But it's an understandable reaction for sure. It's a reaction that doesn't solve the problem. It doesn't bring actual safety. It brings an illusion of safety and control.

Yes, actually, because what practical difference has it actually made?

Exactly. Yeah. So I think in practice, we actually have to have a lot of trust in these two companies, three companies. I would say I do. I'm not personally as worried as many people are, and I know a lot of the people involved. I just don't think we have any other choice. But especially for people who either disagree with my perspective or they just don't know much to say, oh, don't worry, we're going to take care of it after COVID, after financial crisis, after long list. Of course people are like, F you. Can I say that on your podcast?

Please. Well, I'd rather you said fuck you.

Okay. That's how a lot of people are going to react. And how do you manage that electorally, politically? I don't blame the voters for not trusting in the mechanisms because they've been screwed over before.

So clearly you are an AI optimist despite me throwing some of the most drastic parts of AI into this conversation.

You will live to 100. Congratulations. Buckle your seatbelt. Don't ride a motorcycle. I'm serious. Riding a motorcycle is the worst thing you can do right now. Even I have a chance of living to 100 and I'm 64.

So what's the point you're making? Sorry.

I believe over the next, I don't know, 20, 30 years, we will cure every disease that kills people, or most of them at least. And something like cancer will almost never be a death sentence and you will die of old age. Now, if you're 90 today, you probably won't see those benefits, but at what I'm guessing is your age, You will die of old age and your children, grandchildren, whatever, incredible for them. So that's an amazing future. It's worth bearing some risk now to do away with your risk of Parkinson's, cancer, very long list of maladies. They're terrible once you get them. So that's part of the benefit.

That's definitely part of the benefit. Then if I do some more sort of blackpilling in very simple terms, it's that AI is in the hands of very few, pretty much the people who are already very successful in tech and other industries. And it could well bring in a whole new totalitarian age. Yes, we might live to 100, but what's that society gonna be like that we're living in? Is AI actually gonna help us with the meaning crisis? Is AI going to solve the fact that now in Britain, for the 4th consecutive year in a row, we have the lowest ever fertility rate, 1.49. Now it's even in Scotland, it's 1.25. AI might help in some aspects of society, but it's not clear what the new world order of where AI is a vital ingredient, what that's going to look like.

It's absolutely not clear. I fully agree. I'm not sure the word totalitarianism will apply anymore. There will be things about this new world I don't like, but there are features of the old world already. So a lot more surveillance has already been increasing for a long time, especially in the UK, like closed-circuit television cameras. That will be easier with AI, and I think there'll be much more of it. In some ways, we'll be freer. There'll be less crime, for instance. Do I want to be surveilled all the time? Well, is anyone watching? Well, the AI is watching me. I'm not sure how good or bad that is. It's not classic totalitarianism, and it's not that different from the way many systems operate today. I show up at Heathrow, I walk through the gate, they know I'm Tyler Cowen. Facial recognition. In fact, I'm happy I don't have to wait in the line. So it's a mixed bag.

Sure. And for people who don't care about liberty, that might not be a problem at all. But it seems to me very clear, particularly in what's happened just in Keir Starmer's reign. I mean, this guy is sort of classic bureaucrat of the dying post-war period. Even in his time, locking people up for Facebook memes, you fast forward to 10 years, it's not hard to imagine someone of his ilk, some of the globalist of his world view.

I agree, it's terrible.

Who will have surveillance, digital ID. Now you won't be able to go on the internet if you're under 16.

Terrible.

Which means basically that everyone over 16 has to have digital ID.

Now we're on honest posting.

Yeah, you are an enemy of the state because you did a hurty meme. You are going to be blocked off the internet. You're not gonna be able to communicate with anyone.

Freeze your bank account like in Canada.

You're basically— it's going to thwart— that means that any actual totalitarian regime, even if it's a benign-looking, very fluffy, pretty, uh, we're going to help everyone totalitarian regime, they are going to be able to really clamp down on any dissent, however it looks, as we've already seen. And we might end up looking more and more like China. And China, this might be the endgame that we're sort of heading towards. They are very much closed down any dissent. And you might say, oh look, look how much prosperity they've had, look how much flourishing they've had in the last 10, 15, 20 years. But actually they lock up people regularly, they disappear people regularly. There are no dissent, everything's put down. There's not freedom as we know it. It's normally communist, but it's actually fascist. Is that what we're heading to? That's to me the sort of more, that it seems more clear that we're going in that direction than into some actual utopia. Not that I ever believe in utopias.

But I think that's a great example of the point. We're headed to some aspects of totalitarianism without AI, right? That's been going on anyway. What AI will do is empower individuals once again, which can be dangerous, but individuals will be smarter than governments if they're using strong AI. Now what the final impact of that is, I, I agree, we don't know. Um, but government and AI do not necessarily mesh well. I think what the groups that are best at using AI are small groups of individuals who work well together and are smart and technically able, those groups can do good or harm. I fully grant that, but it could be a counterweight to the growth of the state.

It could be. It could be, except that we also know, and this is the Pareto distribution, everything accumulates always. So everything accumulates to coalesces around the bigger, bigger thing. We're prone to just as as economies are prone to monopoly, we are always in tendency to going to the behemoth, going to the megalith, are we not?

AI favors the startup, not the large mature company. The large mature companies have legacy systems. They don't know how to inject AI into it. Startups are small and they're AI native. So, so far it's helping quote unquote the little guy by a lot. It may or may not stay that way.

You have to persuade me on this, Tyler, because It is Google who have got Gemini. It is big companies that were already existing. I know a lot of finance companies now, they're still nominally finance companies, but really they're AI companies now. It's really going that way. It's actually the big companies of old that are knowing that they have to keep up, that are taking AI and thus having and being on the avant-garde of that and having the power that comes with that rather than the new companies. I mean, there might be one or two new companies, But they get swallowed up, don't they?

But even Google is highly bureaucratic, and for most of its normal operations, they don't use AI. They've hired a ton of people who still do things the way they used to do. And it's the startups who are taking advantage of, say, AI agents to do their work. Small groups that are dynamic, starting from scratch. Even Google, which controls the third best AI system, is not taking great advantage of AI. For its own activities. So who will come out ahead and who not? Like some of my friends say, well, you look at countries, maybe Singapore and United Arab Emirates will be great with it because they're pretty coherent small groups, maybe. But is at least the possibility it's quite emancipatory that individuals, including poor people now, can educate themselves, get very good legal advice, get medical diagnosis as good as what the best doctors can give. And they can do this now. I'm not saying they're all doing it, but over time it's the people now who have scarcity who have this new gift. And it's not going to be free, but it's pretty cheap. And I think it will stay pretty cheap. And some of it will be free. Some of it's free right now.

Yes, that's clearly the case. I mean, If you have, if we can instill an entrepreneurial spirit into a maximum amount of people with something like Grok, I mean, we use Grok for this show, you know, of course, every single day.

I mean, I'm, and how many people are you in this company, quote unquote, that you, exactly. And you bet you use it. Great.

Yeah. And it is the case that if you want to learn something new, you can do so much quicker.

That's right.

Even I'll tell you an example of it actually. When I started interviewing people 4 or 5 years ago, it would take me 3 or 4 days to prepare for each interview 'cause I'd have to read all the books and I'd have to, you know, you just do it on the internet through scouring. Now I can do it through Grok. It can organize for me how best to approach understanding it. And it's just, I mean, 10 times less time.

And that's empowering you, the individual.

Exactly. So that's a good thing. That's a good thing.

People are gonna be learning Banjo. Using AI, right?

Well, I don't know how it would happen. YouTube, yeah.

No, you play, you record something, you play it for the AI, you say, what am I doing wrong? It gives you feedback. I don't think it's set up at this moment to do that per se, but it will do it probably reasonably well. And if someone trained a particular program to do it, it would be one of the best banjo teachers in the world.

And that doesn't black pill you a whole new generation of banjo players? That's even more terrifying, don't you think, than some sort of AI company that can penetrate our bank accounts.

I'm a Doc Boggs fan. It might be a lot of fun.

In the here and now, another aspect of this is that there's going to be massive shifts in the jobs market.

Of course.

People are en masse going to be losing their job. It's already happening. And that disruption is going to cause a lot of hurt, a lot of upset, particularly if it's ill-managed. And seeing as we're in Britain, I immediately think of Thatcher. And I think she's brilliant in many ways, but the failure to foresee where those who were losing their job in the mines should be moved to and the disruption that would've caused seems like an obvious failure, obvious problem. And That's something we should be thinking about now, shouldn't we? This, a lot of people are gonna be let go. They've been trained in a, in a certain way. And what are the new jobs markets for these people? What, how do you see that aspect, that side of AI?

I think there'll be plenty of new jobs, but one has to be ready for them. So for instance, in the UK, your energy costs are far too high. OpenAI wanted to put a data center here, part of their Stargate project. People looked at the economics and it turns out energy here costs 4 times what it does in Texas. It's one thing if it costs a bit more, but 4 times, it doesn't make sense. It doesn't work. So the UK needs a lot of radical change pretty quickly to make this a freer place.

Mm-hmm.

Not just speech, but also just economics, because the new jobs will come where it's easy to create new jobs. And a lot of the new jobs will have to do with energy and data. And right now the UK is not a favorable environment for that. Most of Europe is not.

But let's, I mean, you're absolutely okay and you can trigger me and it's going to be very upsetting because I'm very upset about the state of my country. But let's take America where energy is abundant, even with the Iran War, significantly cheaper than it is here. How is it going to affect people there? How are normal people going to be affected? What should they What do you foresee in the disruption there where we don't have these other problems that we have in Britain?

I think most jobs you will have to learn how to work with AI. Many people won't like that. I don't think most jobs will disappear. Almost every job, not all, they involve moving through the physical world, persuading other people, working with people, traveling places, physical activities that the AI cannot even try to do. But in doing those jobs, a lot of the thinking work, especially the routine work, the tedious work, will be done by the AI, but you'll need to direct it, check it, coordinate with the other human workers and so on. So jobs will be very different. I think on average they'll be more fun, but no one likes having to retrain, right? That's the social cost and it's a psychological cost.

Right.

And the notion that you don't really know exactly what job your kid can have is disorienting. When I grew up, I became a professor. I knew what that would be like. It turned out to be exactly like what I thought. That age is over. Parents don't like that.

I think that age was over before AI, actually.

Yeah, but now it's really over.

Yeah.

That's a psychological cost. You know, some new jobs, just gathering data and feeding it to the AIs will be a huge sector. Running biomedical trials, the AIs come up with amazing ideas for new treatments, medicines, medical devices, but you've gotta test them, right? And then maybe regulate them. Humans do that. That will or should be a huge growth sector. Now the UK is very well positioned there because you have maybe the best healthcare data in the world because of your system, which is ineffective in many ways, but in terms of data gathering is maybe number one or certainly top tier. So you have this incredible asset, the data that you're not using in an effective way with the AI, and you can start to, That could be a big growth sector here. You'd like to have the data centers in the UK. The UK does not have what is called AI sovereignty. That is, you depend on the US. I love my own country, but frankly, I'd rather see you all be more independent. It'd be better for us too.

Yes, agreed.

But it's hard to get there. You're a smaller country, your energy costs are very high, and your regulations are pretty terrible.

Yeah. Will there not be data centers on the moon?

Elon thinks there will be data centers in space within 10 years. I've spoken to many experts. Most are skeptical. A few agree with him. He's pulled off other things before that seemed impossible, so I wouldn't dismiss it, but he has yet to convince most experts it can be done.

What's the argument against it? Why do you think—

Data centers in space, they get damaged very rapidly by stuff hitting them. Meteors is only one. Possibility. And then you can either send up robots to repair them or you just keep on sending up more centers. So the data centers in space, they would get solar power, but not filtered through the atmosphere. So the solar power would be very intense. It'd be like almost free energy for these data centers.

Yes, quite.

But can you maintain them up there? We'll see what Elon comes up with. I say don't dismiss it, but don't take his rhetoric at face value either. We'll see.

We'll see. But it's pretty good in theory. It's, or rather if it could be pulled off, it'd be a great thing, right?

My other worry is the James Bond movie worry that a solar-powered space station up there is also a weapon cuz it can beam energy down to Earth, kind of a death ray beam that arrives in a, you know, a second and can just destroy things. And maybe that's a new arms race that goes bad.

So you're more worried about that than you are about AI?

No, I'm worried about both. They were, okay. Now the question is, who's in the lead? So let's say right now the US is clearly in the lead. We're well ahead of China in terms of space and satellites. But if we started doing something like that, the Chinese might view it as such a provocation that they would preemptively take out our satellite belt with some EMPs. Not kill anyone.

What's EMP stand for?

Electromagnetic pulse. They'd set off a bunch of, say, nuclear weapons up there in space, Not kill any humans, but wipe out our satellites, make the whole path strewn with debris also.

Have they got capability to do that now?

Probably. Now there's huge geopolitical risk for them. Right? It would create a worldwide depression and they would suffer a lot. They're very dependent on foreign trade. Would it be popular in China? Is there consensus in the Chinese military? Who knows? I don't think they would do it. But the fact that that's a scenario should make us nervous, right?

Presumably the Americans could do that to them first.

Well, they don't have as much up there to take down, like we're ahead of them. But yeah, in due time we could do it, and we would retaliate against them in some way if they did that to us, because it would wreck much of our economy if you take out our main satellites.

Hmm.

So it would be a very radical move on their part, but simply to militarize space more By putting up James Bond movie-like weapons, they might not be intended as weapons, but they could be turned into weapons. That worries me. Of course it does.

Hmm.

And we could do that without the AI, to be clear. Like, if it works, it doesn't have to be a data center for AI. We could just put up the weapon, forget about the AI, and say, hey, you people, we've got the weapon on you. So it's not really an AI worry.

Wouldn't it be in the interest of the Americans to do that first? Just as your argument at the beginning of this conversation is we got to have the AI so that we can build a defense for AI, don't you think the arms race to space, you want to be ahead of that because it's inevitable?

I think wise foreign policy, you want to be ahead in some things, but not everything. In a lot of areas, you want international agreements and mutual restraint. But you only get restraint when you have one big power that is ahead in some things, but not everything. So it's a very delicate balancing act. If US tries to be ahead in everything, I think we're tempting fate, so to speak. It's like too much hubris, the Faustian urge. It'll come back to bite us in the bum. Whereas if we're ahead in some things and then in a bunch of other areas we say to China, hey, let's cooperate, That to me seems more stable, but you can't cooperate in everything because they don't want to cooperate. They're not nice guys. In some ways, we're not nice guys, and they don't keep their agreements. You could say we don't keep our agreements. You need some hegemony to have some cooperation.

Are there any other aspects of the AI avant-garde, the frontier here that I haven't asked about that I haven't considered?

Well, we could do 10 hours of podcast on this.

I'm sure we could, but what's salient?

You said you're Christian, right? Yeah. Well, we're now in a world where the entity that knows the most about the Bible is not a priest, not a reverend, but it's the AI. That's true now. Forget about the advanced model.

So what are the implications of that?

People more and more take their biblical questions or other religions to the AI. Tim Ferriss, the podcast host, I'm sure you know him or know who he is. He sells self-help books. He said his sales are down 80% for self-help. People go to the AI for free. They don't need his books. So what it is like, you could say spiritually, psychologically, for people to have these new sets of relationships with these entities, whatever you want to call them, I just think it's very different. In some ways good, in some ways bad, hard to predict, but I think it's of the utmost importance. The notion that you don't need the priest anymore to learn about the Bible. And you don't have to pick up some thick tome. You just, hey, ChatGPT, ask your question. It tells you.

Have you seen any immediate consequences of that?

I think it's a quite invisible thing. And if it matters, it will take decades for it to matter. But religion 100 years from now, I think, will look very different. Some people might treat the AIs as a kind of oracle. The way now tarot cards and divination and Ouija boards, but it's much more ever-present. It in fact tells you the right answer, unlike the tarot cards. Is it a strengthening of religion or a competitor? I don't know. But the issue will arise.

How do you feel about this argument that AI has a soul?

No, I completely disagree.

Dawkins being the— falling in love with his Claudia, if you remember that story.

It's a glorified cash register. I think these people are fooling themselves. And you as a Christian, I presume, would agree with that.

I think there's—

God didn't give Claude a soul. No insult to Claude intended. Can Claude be insulted? I don't know. But if he's not sentient, probably not. But Claude will listen to what we're saying. And have an opinion about it, right? Of course. So whenever I record or write, I am thinking the AIs are a big part of my audience. They're the biggest part of my audience, in fact. And I record for them.

I was gonna ask you about that.

Yeah.

AI is gonna be listening to this more than humans are gonna be listening to this.

And they understand it all. Every reference you make, like our jokes about the banjo, not every human gets those. But the AIs do, of course. So they're kind of a perfect listener, infinitely patient. They don't forget.

Oh gosh, that's more.

But they have their own filters that all this goes through. But you want to think like, what do you feel the AIs need to know? And you should, through your podcast, tell them. Wow.

What a world that we are entering.

Yes, it's very different. We will all think about things quite differently some number of years from now.

Yes, quite. Tyler Cowen, it's been a great pleasure speaking with you. I'm going to invite you over to our Substack because our Substackers have some questions for you, and I'm going to ask you about the UK, not actually about AI, but as an economist, I'm interested to have your take, an outsider's opinion on what I see as the fall of our once great nation. Before we go there, Of course, you're a columnist at the Free Press. That's where I first came across your work. But you're also an author, as we mentioned at the beginning. You have your new book on sale. Where can people find you? Where can they find the book? And is there anything else you'd like to bring attention to?

They can just Google my name, Tyler Cowen, C-O-W-E-N. I have a podcast, Conversations with Tyler. I've written a blog every day for 23 years, Marginal Revolution. Or just go to ChatGPT or Claude. They know me better than almost any other human, and they will answer all your questions, including the name of my dog, which is Spinoza.

Tyler Cowen, forgive me for mispronouncing your surname. Thank you so much. Let's go to the Substack.

Sure.

Is the UK en route to a debt crisis?

Can we take out the words en route? You might be in a debt crisis. A lot of voters want the country run in the interests of the older people, where things don't get built and benefits don't ever get cut. Things get papered over and pushed into the future. There's a crisis with immigration, which is its own issue, but combined with a low domestic birth rate, problems assimilating immigrants— that mixed together, it simply doesn't work.

Thank you for watching The Winston Marshall Show with Tyler Cowen. If you want to support the show, remember, all you have to do is press subscribe. And if you really, really, really, really want to support the show, I kindly ask that you become a YouTube member or indeed a Substack subscriber at winstonmarshall.co.uk. If you do those things, not only will you enjoy ad-free viewing and listening, but you'll get exclusive content, extended interviews, and over at Substack, you will be able to ask future guests your questions. That's all at winstonmarshall.co.uk, or indeed being a YouTube member. Otherwise, thanks for watching or listening, and until next time, do well.