Loading summary
Transcript414 lines
- [00:00]
Keshav Murugesh
Today I'm chatting with Tameh Besaroglu and Ege Erdo. They were previously running Epoch AI and are now launching Mechanize, which is a company dedicated to automating all work. One of the interesting points you made recently, Tame, is that the whole idea of the intelligence explosion is mistaken or misleading. Why don't you explain what you were talking about there?
- [00:22]
Tameh Besaroglu
Yeah, I think it's not a very useful concept. It's kind of like calling the Intelligence Industrial Revolution a horsepower explosion. Like, sure, during the Industrial Revolution, we saw this drastic acceleration in raw physical power, but there are many other things that were maybe equally important in explaining the acceleration of growth and technological change that we saw during the Industrial Revolution.
- [00:43]
Keshav Murugesh
What is a way to characterize the broader set of things that the horsepower perspective would miss about the Industrial Revolution?
- [00:50]
Tameh Besaroglu
So I think in the case of the Industrial Revolution, it was a bunch of these complementary changes to many different sectors in the economy. So you had agriculture, you had transportation, you had law and finance, you had urbanization and moving from rural areas into cities. There were just many different innovations that kind of happened simultaneously that gave rise to this change in the way of economically organizing our society. It wasn't just that we had more horsepower. I mean, that was part of it. But that's not the kind of central thing to focus on when thinking about the Industrial Revolution. And I think similarly for the development of AI. Sure, we'll get a lot of very smart AI systems, but that will be one part among very many different moving parts that explain why we expect to get this transition and this acceleration and growth and technological change.
- [01:46]
Keshav Murugesh
Yeah, I want to better understand how you think about that broader transformation before we do. The other really interesting part of your worldview is that you have longer timelines to get to AGI than most of the people in San Francisco who think about AI. When do you expect a drop in remote worker replacement?
- [02:05]
Ege Erdo
Yeah, maybe for me that would be around like 2045 or.
- [02:10]
Keshav Murugesh
Wow. Wait, and you.
- [02:11]
Tameh Besaroglu
I'm a little bit more bullish. I mean, it depends what you mean by drop in remote worker and whether it's able to do like literally everything that can be done remotely or do most things.
- [02:20]
Ege Erdo
I'm saying literally everything for literally everything.
- [02:23]
Tameh Besaroglu
Just shade, I guess, predictions by five years or by 20% or something.
- [02:28]
Keshav Murugesh
Why? Because we've seen so much progress over even the last few years. We've gone from chatgpt like two years ago to now. We have models that can literally do reasoning, are better coders than me, and I studied software Engineering in college. I mean, I did become a podcaster. I'm not saying I'm like the best coder in the world, but if you made this much progress in the last two years, why would it take another 30 to get to full automation of human brains?
- [02:58]
Ege Erdo
Right.
- [02:59]
Keshav Murugesh
I said that wrong. You know what I'm saying? Full automation of remote work.
- [03:01]
Ege Erdo
Yeah. So I think a lot of people have this intuition that progress has been very fast. They just look at the trend lines and just extrapolate. Obviously it's going to happen in, I don't know, 2027 or 2030 or whatever. It is very bullish. And obviously that's not a thing you can literally do. There isn't a trend you can literally extrapolate of when do we get the full automation? Because if you look at the fraction of the economy that has actually been automated, it's very like by AI, it's very small. So if you just extrapolate that trend, which is something say, Robin Hanson likes to do, you're going to say, well, it's going to take centuries or something. Now, we don't agree with that view, but I think one way of thinking about this is like, how many big things are there? How many core capabilities, components are there that the AI systems need to be good at in order to have this very broad economic impact? Maybe 10x acceleration in growth or something. How many things have you gotten over the past 10 years? 15 years? And we also have this compute centric.
- [04:05]
Tameh Besaroglu
So just to double click on that. I mean, I think what EGE is referring to is like, if you look at the past 10 years of AI progress, we've gone through about nine or 10 orders of magnitude of compute, and we got various capabilities that were unlocked. So you had, in the early period, people were kind of solving gameplay on specific games, on very complex games. And that happened 2015 to maybe 2020 and go and chess and Dota and other games. And then, and then you had maybe sophisticated language capabilities that were unlocked with these large language models and maybe kind of advanced abstract reasoning and coding and maybe math. That was maybe another big such capability that got unlocked. And so maybe there are a couple of these big unlocks that happened over the past 10 years. But it takes that happened on the order of once every three years or so, or maybe one every three orders of magnitude of compute scaling. And then you might ask the question, how many more such competencies might we need to unlock in order to be able to have an AI system that can match the capabilities of humans across the board maybe specifically just on remote work tasks. And so then you might ask, well, maybe you need kind of coherence over very long horizons, or you need kind of agency and autonomy, or maybe you need multimodal kind of full multimodal kind of understanding, just like a human would. And then you ask the question, okay, how long might that take? And so you can think about, well, just in terms of calendar years, the previous unlocks took about, you get one every three years or so. But of course that previous period coincided with this rapid scale up of the amount of compute that we use for training. So we went through maybe nine or ten orders of magnitude since Alexnet compared to the biggest models we have today. And we're getting to a level where it's becoming harder and harder to scale up compute. And we've done some extrapolations and some analysis looking at specific constraints like energy or GPU production. And based on that it looks like we might have maybe three or four orders of magnitude of scaling left. And then you're really spending a pretty sizable fraction or a non trivial fraction of world output on just building up data centers, energy infrastructure fabs, which is.
- [06:40]
Keshav Murugesh
Already like 2% of GDP, right?
- [06:42]
Tameh Besaroglu
I mean currently it's less than 2%. Yeah.
- [06:44]
Ege Erdo
But currently most of it is actually not going through towards AI chips. Even most TSMC capacity currently is going towards mobile phone chips or something like that. Right.
- [06:52]
Keshav Murugesh
Even Leading Edge is going like, it's like five years.
- [06:54]
Tameh Besaroglu
Even Leading Edge is pretty, pretty small. But yeah. So that suggests that we might need a lot more compute scaling to get these additional capabilities to be unlocked. And then there's a question of do we really have that as a, do we have that in us as an economy to be able to sustain that scaling?
- [07:15]
Keshav Murugesh
But it seems like you have this intuition that there's just a lot left to intelligence when you play these models, it's like they're almost there. It's like you forget you're often talking to an AI.
- [07:26]
Ege Erdo
What do you mean they're almost there? I don't know. I can't ask Claude to pick up this cup and put it over there.
- [07:31]
Keshav Murugesh
For remote work, you know?
- [07:32]
Ege Erdo
Okay, but even for remote work I can't ask Claude to. I think the current computer use systems can't even book a flight properly.
- [07:39]
Keshav Murugesh
How much of an update would it be if by the end of 2026 they could book a flight?
- [07:43]
Ege Erdo
I probably think by the end of this year they're going to be able to do that. But that's very like nobody gets A job where they're paid to like book flights for like, like that's not a task.
- [07:55]
Tameh Besaroglu
Some people, I mean, if it's literally just book flight, you know, and without, you know.
- [08:01]
Ege Erdo
But I think that's an important point because a lot of people like look at jobs in the economy and then they're like, oh, like that person. Like their job is to just do X, but then that's not true. Like that's something they do in their job. But it's probably, if you look at the fraction of their time on the job that they spend on doing that is a very small fraction of what they should do. It's just this popular conception. People have travel agents, they just book hotels and flights, but that's not actually most of their job. So automating that actually wouldn't automate their job and it wouldn't have that much of an impact on the economy. So I think this is actually an important thing, that important worldview difference that separates us from people who are much more bullish because they think jobs in the economy are much simpler in some sense and they're going to take much fewer competences to actually fully automate.
- [08:48]
Keshav Murugesh
So our friend Leopold has this perspective of unhobblings where the way to characterize it might be like they're basically like baby AGIs already. And then there's because of the constraints we artificially impose upon them by for example, only training them on text and not giving them the training data that is necessary for them to understand a slack environment or a Gmail environment or previously before inference time scaling, not giving them the chance to meditate upon what they're saying and really think it through, and not giving them the context about like what is actually involved in this job, only giving them this piecemeal a couple minutes worth of context in the prompt, we're holding back what is fundamentally a little intelligence from being as productive as it could be. Which implies that unhobbling just seem easier to solve for than entirely new capabilities of intelligence. What do you make of that framework?
- [09:46]
Tameh Besaroglu
I mean, I guess you could have made similar points five years ago and say you look at AlphaZero and there's this mini AGI there, and if only you unhobbled it by training it on text and giving it all your context and so on, that just wouldn't really have worked. I think you do really need to rethink how you train these models in order to get these capabilities.
- [10:10]
Keshav Murugesh
But I think the surprising thing over the last few years has been that you can start off with this pre trained corpus of the Internet and it's actually quite easy. ChatGPT is an example of this unhobbling where 1% of additional compute spent on getting it to talk in a chatbot like fashion with post training is enough to make it competent, really competent at that capability. So why not think that agency, I mean reasoning is another example where it seems like the amount of compute that is spent on RL right now in these models is a small fraction of total compute. Again, reasoning seems complicated and then you just do 1% of compute and it gets you that. Why not think that computer use or long term agency on computer use is a similar thing?
- [10:56]
Tameh Besaroglu
So when you say reasoning is easy and it only took this much compute and it wasn't very much, and maybe you look at the sheer number of tokens and it wasn't very much and so it looks easy. Well, that's kind of true from our position today. But I think if you ask someone build a reasoning model in 2015, then it would have looked insurmountable. You would have had to train a model on tens of thousands of GPUs, you would have had to solve that problem and each order of magnitude of scaling from where they were would pose new challenges that they would need to solve. Need to produce kind of Internet scale or tens of trillions of tokens of data. In order to actually train a model that kind of has the knowledge that you can then unlock and access by way of training it to be a reasoning model, you need to maybe make the model more efficient at kind of doing inference and maybe distill it. Because if it's very slow, then you have a reasoning model that's not particularly useful. So you also need to make various innovations to get the model to be distilled so that you can train it more quickly. Because these rollouts take very long. It actually becomes a product that's valuable if it's a couple tokens a second as a reasoning model that would have been very difficult to work with. So in some sense it looks easy from our point of view standing on this huge stack of technology that we've built up over the past five years or so. But at the time it would have been very hard. And so my claim would be something like I think the agency part might be easy in a similar sense that in five years or three years time or whatever, we will look at what unlocked agency and it'll look fairly simple. But the amount of work that in terms of these complementary innovations that enable the model to be able to learn how to become a competent agent. That might have just been very difficult and taken years of innovation and a bunch of improvements in kind of hardware and scaling and various other things.
- [12:54]
Keshav Murugesh
Yeah, I feel like what's dissimilar between 2015 and now in 2015, if you were trying to solve reasoning, you just didn't have a base to start on. I don't know, maybe you tried formal proof methods or something but there was no leg to stand on where now actually you have the thing, you have the pre trained base model, you have these techniques of scaffolding, of post trading of rl. And so it seems like you are skeptical that you think that those will look to the future as say AlphaGo looks to us now in terms of the basis of a broader intelligence. And I'm curious if you have intuitions on why not think that language models as we have them now are like we got the big missing piece right. And now we're just plugging things on top of it.
- [13:51]
Ege Erdo
Well, I mean, I guess what is the reason for believing that? I mean you could have looked at AlphaGo or AlphaGo 0 AlphaZero. Those seemed very impressive at the time. You were just learning to play this game with no human knowledge. You're just learning to play it from scratch. And I think at the time it did impress a lot of people. But then people tried to apply it to math, they tried to apply it to other domains and it didn't work very well. They weren't able to get competent agents at math. So it's very possible that these models, at least the way we have them right now, you're going to try to do the same thing people did for reasoning but for agency. It's not going to work very well. And then you're not going to.
- [14:32]
Keshav Murugesh
You're saying by the end of 2026 we will have agentic computer use.
- [14:37]
Tameh Besaroglu
I think I guess said you'd be able to book a flight which is very different from having full agentic computer use like a human.
- [14:44]
Keshav Murugesh
It's because the other things you need to do on a computer is just made up of things like booking a flight.
- [14:51]
Ege Erdo
Sure. But they are not disconnected tasks. That's like saying everything you do in the world is just like you just move parts of your body and then you move your mouth and your tongue and then you throw it in your head. But that's a very. Yeah. Individually those things are simple. But then how do you put them together? Right.
- [15:10]
Keshav Murugesh
Yeah. Okay, so there's like two pieces of evidence that you can have that are quite dissimilar. One, the meter eval, which we've been talking about privately, which shows that the task length over certain kinds of tasks. I can already see you're getting ready. Has been double the AI's ability to do the kind of thing that it takes a human 10 minutes to do, or an hour to do, or four hours to do. The length of time for corresponding human tasks. It seems like these models seem to be doubling their task length every seven months. So the idea being that by 2030, if you extrapolate this curve, they could be doing tasks that take humans one month to do or one year to do. And then this long term coherency in executing our task is fundamentally what intelligence is. So this curve suggests that we're getting there. The other piece of evidence, I kind of feel like my own mind works this way of I get distracted easily and it's kind of hard to keep a long term plan in my head at the same time. And I'm slightly better at it than these models, but they don't seem that dissimilar to me. I mean, even I would have guessed reasoning is just a really complicated thing. And then it seems like, oh, it's just something like learning 10 tokens worth of mcts of wait, let's go back. Let's think about this another way. Chain of thought alone just gets you this boost. And it just seems like intelligence is simpler than we thought. Maybe agency is also simpler in this way.
- [16:39]
Ege Erdo
Yeah, I mean, I would say that reasoning did seem. I mean, I think there's a reason to expect complex reasoning to not be as difficult as people might have thought even in advance. Because a lot of the tasks that AI solved very early on were tasks of various kinds of complex reasoning. So it wasn't the kind of reasoning that goes into when a human solves a math problem. But if you look at the major AI milestones since 1950, a lot of them are for complex reasoning. Like a chess is, you can say a complex reasoning task. Go is, you could say a complex reasoning task.
- [17:14]
Keshav Murugesh
I think there are also examples of long term agency. Like winning at Starcraft is an example of being agentic over a meaningful period of time.
- [17:24]
Ege Erdo
That's right. So the problem in that case is that it's a very specific narrow environment. You can say that playing Go or playing chess, that also requires a certain amount of agency. And that's true, but it's a very narrow task. So that's like saying if you construct a software system that is able to react to Very specific, very particular kind of images or very specific video feeds or whatever. Then you're getting close to general sensory, motor skill, automation. But the general skill is something that's very different. And I think we're seeing that we still are very far, it seems like, from an AI model that can take a generic game of Steam. Let's say you just download a game released this year, you don't know how to play this game and then you just have to play it. Right. And most games are actually not that difficult for a human.
- [18:21]
Keshav Murugesh
Like, I mean, what about Claude Pace Pokemon? I don't think it was trained on Pokemon.
- [18:25]
Ege Erdo
Right. So that's an interesting example. First of all, I find the example very interesting because yeah, it was not trained explicitly. Like it wasn't. They didn't do some RL on like playing Pokemon Red. But obviously the model knows how it's supposed to play Pokemon Red because there's tons of material about Pokemon Red on the Internet. In fact, if you, if you were playing Pokemon Red and you got stuck somewhere, you didn't know what to do, you could probably go to Claude and ask it, Claude, like I'm stuck in Mount Moon and what am I supposed to do? And then it could probably give you a fairly decent answer. But that doesn't stop it from getting stuck in Mount Moon for 48 hours. So that's a very interesting thing where it has explicit knowledge, but then when it's actually playing the game, it doesn't behave in a way which reflects that it has that knowledge.
- [19:10]
Keshav Murugesh
All it's going to do is like plug, you know, plug the explicit knowledge to its actual. Right.
- [19:13]
Ege Erdo
But is that easy?
- [19:16]
Keshav Murugesh
I just, like, I'm not sure I understand why. Okay, if you can leverage your knowledge from pre training about these games in order to be somewhat competent in them, I feel like that is some evidence of. Okay, they're going to be using, they're going to be leveraging a different base of skills.
- [19:32]
Ege Erdo
Yes.
- [19:33]
Keshav Murugesh
But that, with that same leverage they're going to have like a similar repertoire of abilities. Right. If you've read everything about whatever skill that every human has ever seen, I.
- [19:44]
Ege Erdo
Mean, a lot of the skills that people have that we don't have very good training data for them.
- [19:48]
Keshav Murugesh
That's right. That's right. What would you want to see over the next few years that would make you think, oh no, I'm actually wrong. And this was like the last unlock and it was like, now just a matter of ironing out the kinks and then we get the thing that will kick off the, dare I say, intelligence explosion.
- [20:05]
Tameh Besaroglu
Yeah. So I think something that would reveal its ability to do very long context things, use multimodal capabilities in a meaningful way and integrate that with reasoning and other types of systems and also agency and being able to take action over a long horizon and accomplish some tasks that takes very long for humans to do. Not just in specific software environments, but just very broadly. Say, downloading an arbitrary game from Steam and something that's never seen before doesn't really have much training data. Maybe it was released after its training cutoff and so there's no tutorials, or maybe there's no earlier versions of the game that has been discussed on the Internet. And then accomplishing that game and actually playing that game to the end and accomplishing these various milestones that are challenging for humans, that would be a substantial update. I mean, there are other things that would update me too. Like OpenAI making a lot more revenue than it's currently doing.
- [21:11]
Keshav Murugesh
Is the 100 billion in revenue that would, according to their contract, mark them as AGI enough.
- [21:16]
Tameh Besaroglu
I think that's not a huge update to me if that were to happen. So I think the update would come. If it was in fact $500 billion in revenue or something like that, then I would certainly update quite a lot. But 100 billion, that seems pretty likely to me. I would assign that maybe, I don't know, 40% chance or something by the end.
- [21:37]
Keshav Murugesh
I mean, what is this like if you've got a system that is in just producer surplus terms worth 100 billion. The difference between this and AlphaZero is AlphaZero is never going to make $100 billion in the marketplace. Right. So just the what is intelligence? It's like something able to usefully accomplish its goals, or your goals. If people are willing to pay $100 billion for it, that's pretty good evidence that it's like accomplishing some goals.
- [22:05]
Ege Erdo
Sure.
- [22:05]
Tameh Besaroglu
I mean people pay $100 billion for all sorts of things. Right. That itself is not a very strong piece of evidence that it's going to be transformative.
- [22:13]
Ege Erdo
I think people pay trillions of dollars for oil. Oil is not. I don't know, it seems like a very basic point, but the fact that people pay you a lot of money for something doesn't mean it's going to transform the world economy. If only we manage to unhobble it. That's a very different claim. Right.
- [22:28]
Keshav Murugesh
Look, a ton of B2B software companies start off by building self serve consumer grade products and that's fine. At first. Eventually though, you have to go after Enterprise. The most successful and durable software companies of the last decade have all made this transition. But getting enterprise ready is hard. Single sign on role based access controls and comprehensive audit logs are all actually quite complex and tedious to build, and they're ripe for bugs and annoying edge cases. These features take a ton of engineering time and capital, which you should be spending on the core product. For example, One of Slack's PMs said that they spent $30 million building these features and they were only half done. That's where WorkOS comes in. WorkOS has helped Vercel, Plaid, Vanta, OpenAI and hundreds of others become enterprise ready with APIs to integrate all of these common features. If you want to learn more, go to workos.com and tell them that I sent you. Okay, so then this brings us to the intelligence explosion. Because what people will say is we don't need to automate literally everything that is needed for automating remote work, let alone all human labor in general. We just need to automate the things which are necessary to fully close the R and D cycle needed to make smarter intelligences. And if you do this, you get a very rapid intelligence explosion. And the end product of that explosion is not only an AGI, but something that is superhuman potentially. These things are extremely good at coding and they're good at the kinds of things that you would think and reasoning, and it seems like the kinds of things that would be necessary to automate R and D at AI labs. What do you make of that logic?
- [24:14]
Ege Erdo
I mean, I think if you look at their capability profile, it is if you compare it to a random job in the economy. I agree they are better at doing sort of coding tasks that will be involved in R and D compared to like a random job in the economy. But in absolute terms, I don't think they're that good. I think they are good at things that maybe impress us. About human coders. If you wanted to see what makes a person a really impressive coder, you might look at their competitive programming performance. In fact, companies often hire people based on if they are relatively junior, based on their performance on these kinds of problems. But that is just impressive in the human distribution. So if you look in absolute terms at what are the skills you need to actually automate the process of being a researcher, then what fraction of those skills do the AI systems actually have, even in coding? Like a lot of coding is you have a very large code base you have to work with. The instructions are very kind of vague. There isn't. For example, you mentioned a meter eval in which, because they needed to make it an eval, all the tasks have to be kind of compact and closed and have clear evaluation metrics. Like, here's a model, get its loss on this data set as low as possible, or whatever. Or like, here's another model and its embedding matrix has been scrambled. Just fix it to recover most of its original performance, et cetera. Those are not problems that you actually work on in AI, R&D. They're like very artificial problems. Now, if a human was good at doing those problems, you would infer, I think, logically, that that human is likely to actually be a good researcher. But if an AI is able to do them, the AI lacks so many other competences that a human would have, not just a researcher, just an ordinary human that we don't think about in the process of research. So our view would be automating research is first of all more difficult than people give it credit for. I think you need more skills to do it and definitely more than models that are displaying right now. And on top of that, even if you did automate the process of research, we think a lot of the software progress has been driven not by cognitive efforts, though that has played a part, but it has been driven by compute scaling. We just have more GPUs. You can do more experiments to figure out more things. Your experiments can be done at larger scales. And that is just a very important driver. If you're 10 years ago, 15 years ago, you're trying to figure out what software innovations are going to be important in 10 or 15 years, you would have had a very difficult time. In fact, you probably wouldn't even conceive of the right kind of innovations to be looking at because you would be so far removed from the context of that time with much more abundant compute and all the things that people would have learned by that point. So these are two components of our view. Research is harder than people think and depends a lot on compute AI.
- [27:16]
Keshav Murugesh
Sorry, can you put a finer point on what is the kind of thing, what is an example of the kind of task which is very dissimilar from train a classifier or debug a classifier that is relevant to AI, R&D.
- [27:31]
Tameh Besaroglu
I think it's like examples might be introducing novel. Having novel innovations that are very useful for unlocking innovations in the future. So that might be introducing some novel way of thinking about a problem or introducing. So maybe a good example might be in mathematics, where we have these reasoning models that are extremely good at solving math problems.
- [27:57]
Ege Erdo
I mean, very short horizon, maybe not.
- [28:01]
Tameh Besaroglu
Extremely good, but certainly better than I can and better than maybe most undergrads can. And so they can do that very well, but they're not very good at coming up with novel conceptual schemes that are useful for making progress in mathematics. So it's able to solve these problems that you can kind of neatly excise out of some very messy context. And it's able to make a lot of progress there. But within some much messier context, it's kind of not very good at figuring out what directions are especially useful for you to build things or kind of make incremental progress on that enables you to have a big kind of innovation later down the line. So thinking about both this larger context as well as maybe much longer horizon, much fuzzier things that you're optimizing for, I think it's much worse at those types of things, right?
- [28:55]
Ege Erdo
So I think one interesting thing is if you just look at these reasoning models, they know so much, especially the larger ones, because they know in literal terms more than any human does in some sense. And well, we have unlocked these reasoning capabilities on top of that knowledge. And I think that is actually what is enabling them to solve a lot of these problems. But if you actually look at the way they approach problems, the reason what they do looks impressive to us is because we have so much less knowledge and the model is approaching the problems in a fundamentally different way compared to a human would. A human would have much more limited knowledge and they would usually have to be much more creative in solving problems because they have this lack of knowledge. While the model knows so much, you'd ask it some obscure math question where you need some specific theorem from 1850 or something, and then it would just know that if it's like a large model, so that makes the difficulty profile very different. And if you look at the way they approach problems, the reasoning models, they are usually not creative. They are very effectively able to leverage the knowledge they have, which is extremely vast. And that makes them very effective in a bunch of ways. But you might ask the question, has a reasoning model ever come up with a math concept that even seems like slightly interesting to a human mathematician? And I've never seen that.
- [30:20]
Keshav Murugesh
I mean, they've been around for all of six months, but that's a long timeline.
- [30:23]
Tameh Besaroglu
I mean, that's.
- [30:25]
Ege Erdo
Just think about it.
- [30:26]
Tameh Besaroglu
One mathematician might have been able to do a bunch of work over that time, and they have produced orders of magnitude Fewer tokens on math.
- [30:34]
Keshav Murugesh
That's right, that's right, yeah.
- [30:35]
Ege Erdo
And then I just want to emphasize this because just think about the sheer scale of knowledge that these models have. It's enormous from a human point of view. So it is actually quite remarkable that there is no interesting recombination. No interesting, oh, this thing in this field looks kind of like this thing in this other field. There's no innovation that comes out of that. And it doesn't have to be a big math concept. It could be just a small thing that maybe you could add to a Sunday magazine on math that people used to have, but there isn't even an example of that.
- [31:09]
Tameh Besaroglu
I think it's useful for us to explain a very important framework for our thinking about what AI is good at and what AI is lagging in, which is this idea of Moravex paradox, that things that seem very hard for humans, AI systems tend to make much faster progress on, whereas things that look a bunch easier for us, kind of AI systems that totally struggle or often totally incapable of doing that thing. And so this kind of abstract reasoning, playing chess, playing Go, maybe playing Jeopardy, doing kind of advanced math and solving.
- [31:48]
Ege Erdo
Math problems, there are even stronger examples like multiplying 100 digit numbers in your head, which is the one that Kol solved first out of almost any other problem, or following very complex sort of symbolic logic arguments, like deductory arguments. People actually struggle with that a lot. Like how do premises logically fall from conclusions? People have a very hard time with that. Very easy for formal proof systems.
- [32:14]
Tameh Besaroglu
An insight that is related and is quite important here is that the kind of very. The task that humans seem to struggle on and AI systems seem to make much faster progress on are things that emerged fairly recently in evolutionary time. So advanced language use emerged in humans maybe 100,000 years ago. And certainly playing chess and Go and so on are very recent innovations. And so evolution has had much less time to optimize for them, in part because they're very new, but also in part because when they emerged there was a lot less pressure because it conferred kind of small fitness gains to humans. And so evolution didn't optimize for these things very strongly. And so it's not surprising that on these specific tasks that humans find very impressive when other humans are able to do it, that AI systems are able to make a lot of fast progress. In humans, these things are often very strongly correlated with other kind of competencies, like being good at just achieving your goals or being a good coder is often very Strongly correlated with solving kind of coding problems, or being a good engineer is often correlated with solving competitive coding problems. But in AI systems, the correlation isn't quite as strong. And even within AI systems, it's the case that the strongest systems on competitive programming are not even the ones that are best at actually helping you code. So, like, you know, O3 minis high seems to be maybe the best at solving competitive code problems, but it isn't the best at actually helping you write code.
- [33:54]
Ege Erdo
And maybe it isn't getting most of the enterprise revenue from places like Cursor or whatever. Like, let's just. Claude. Right.
- [34:00]
Tameh Besaroglu
But an important insight here is that the things that we find very impressive when humans are able to do it, we should expect that AI systems are able to make a lot more progress on that, but we shouldn't update too strongly about just their general competence or something, because we should recognize that this is a very narrow subset of relevant tasks that humans do in order to be a competent, economically valuable agent.
- [34:26]
Keshav Murugesh
Yeah. First of all, I actually just really appreciate that there is an AI organization out there where. Because there's other people who take the compute perspective seriously, think or try to think empirically about scaling laws and data and whatever. And it's striking how often that, like, taking that perspective seriously leads people to just be like, okay, 2027 AGI. Which might be correct, but it is just interesting to get like, no. We've also looked at the exact same arguments, the same papers, the same numbers, and we've come to a totally different conclusion on all these arguments. I think this is all fascinating. Okay, so I asked Dario this exact question two years ago when I interviewed him, and it went viral over Twitter.
- [35:12]
Ege Erdo
Did he say AGI in two years?
- [35:14]
Keshav Murugesh
But Dario's always had short timelines.
- [35:16]
Ege Erdo
Okay, but we are two years later.
- [35:18]
Keshav Murugesh
Did he say two? I think he actually did say two years.
- [35:20]
Ege Erdo
Or did he say three years?
- [35:21]
Tameh Besaroglu
So we have one more year. One more year. Better work hard.
- [35:26]
Keshav Murugesh
But he's. I mean, I think he's like. He in particular has not been that well calibrated. He's like, in 2018, he had like.
- [35:34]
Tameh Besaroglu
I remember talking to, like, a very senior person who's now at anthropic in 2017, and then he told various people that they shouldn't do a PhD because by the time they completed it.
- [35:46]
Keshav Murugesh
That's right. That's right.
- [35:47]
Tameh Besaroglu
Everyone would be automated. Yeah, yeah.
- [35:49]
Keshav Murugesh
So anyways, I asked him this exact same question. Right. Because he had short timelines, which is that if a human knew the amount of Things these models know they would be finding all these different connections. And in fact I was asking Scott about this the other day when I interviewed him, Scott Alexander and he said look, humans also don't have this kind of logical omniscience. And I'm not saying we're omniscient, but we have examples of humans finding these kinds of connections. This is not an uncommon thing. Right. I think his response to that was that these things are just not trained in order to find these kinds of connections. But if you like, their view is that it would not take that much extra computer in order to build some RL environment in which they're incentivized to find these connections. Next token prediction just isn't incentivizing them to do this. But the RL required to do this would not be that or set up some sort of scaffolds. I think actually Google DeepMind did do some similar scaffold to make new discoveries and I didn't look into how impressive the new discovery was. They claimed that some new discovery was made by an LLM as a result. On the Moravax paradox thing. This is actually a super interesting way to think about AI progress. But I would also say there that if you compare animals to humans, long term intelligent planning, like an animal is not going to help you book a flight. Either an animal or like an animal is not going to do remote work for you or even do the kinds of things. I think what separates humans from other animals is that we can hold long term plan, we can come up with a plan and execute on it. Whereas other animals, I often had to go by instinct or within the kinds of environments that they have evolutionary knowledge of rather than like I'm put in the middle of the savanna or I'm put in the middle of the desert or I'm put in the middle of the tundra and I'll learn how to make use of the tools and whatever there like. I actually think there's like a huge discontinuity between humans and animals in their ability to survive in different environments just based on their knowledge. And so it's like a recently optimized thing as well. And then I'd be like, okay, well AIs will, we got it soon. AIs will optimize it for fast.
- [37:50]
Ege Erdo
Right? So I would say if you're comparing animals to humans, it's kind of a different thing. I think animals, if you could put the competences that the animals have into AI systems that might just already get you to AGI Already I think the reason why there is such a big discontinuity between animals and humans is, is because animals have to rely entirely on natural world data basically to train themselves. Imagine that the only thing as a human that you saw was nobody talked to you, you didn't read anything. You just had to learn by experience, maybe to some extent by imitating other people, but not like you have no explicit communication. Well, it would be very inefficient. What's actually happening is that you have this, I think some other people have made this point as well is that evolution is sort of this outer optimizer that's improving the software efficiency of the brain in a bunch of ways. There's some genetic knowledge that you inherit, not that much because there isn't that much space in the genome. And then you have this lifetime learning which is you don't actually see that much data during lifetime learning. A lot of this is redundant and so on. So what changed, seems to have changed with humans compared to other animals is that humans became able to have culture and they have language which enables them to have a much more efficient training data modality compared to animals. They also have I think, stronger ways in which they tend to imitate other humans and learn from their skills. So that also enables this knowledge to be passed on. I think animals are pretty bad at that compared to humans. So basically as a human you are just being trained on much more efficient data and that creates further insights to be then efficient at learning from it. And then that creates this feedback loop where the selection pressure gets much more intense. So I think that's roughly what happened with humans. But a lot of the capabilities that you need to be a good worker in the human economy animals already have. So they are able to, they have quite sophisticated sensory motor skills. I think they are actually able to do like animals are actually able to pursue long term goals, but ones that.
- [40:04]
Keshav Murugesh
They have been instilled by evolution. I think a lion will find a gazelle and that is a complicated thing to do and requires stalking and blah blah, blah.
- [40:12]
Ege Erdo
When you say it's been instilled by evolution there isn't that much information in.
- [40:15]
Keshav Murugesh
The genome but it just like I think if you put the lion in the Sahara and you're like, go find lizards instead.
- [40:22]
Ege Erdo
Okay, so suppose you put a human and they haven't seen the relevant training data.
- [40:27]
Keshav Murugesh
I think they'd like they do slightly better.
- [40:30]
Ege Erdo
Slightly better, but not that much better. I think a lot of the. Again, didn't you recently have an interview with Joseph Henrik, yeah, so he would probably tell you that, well, okay, and.
- [40:41]
Keshav Murugesh
I think what you're making is actually a very interesting and subtle point. That is an interesting implication. So often people point to, they say that ASI will be this huge discontinuity because while we have this huge discontinuity in the animal to human transition, where something, not that much change between pre human primates and humans genetically, but it resulted in this humongous change in capabilities. And so they say, well, why not expect something similar between human level intelligence and superhuman intelligence. And the point you're making is that, or at least one implication of the point you're making is that actually it wasn't that we just gained this like incredible intelligence because of biological constraints. Animals have just been like held back in this really weird way that no AI system has been arbitrarily held back of not being able to communicate with other copies or with other knowledge sources. And so since AIs are not held back artificially in this way, there's not going to be a point where we should take away that hobbling and then now they're like, now they explode now. Actually, I think I would disagree with that. The implication that I made, I would actually disagree with. I'm a sort of unsteearable chain of thought. But because as we wrote a blog post together about AI corporations where we discuss actually there will be a similar unhobbling with future AIs which is not about the intelligence, but a similar level of bandwidth and communication and collaboration with other AIs, which is a similar magnitude of change from non human animals to humans in terms of their social collaboration that AIs will have with each other because of their ability to copy all their knowledge exactly, to merge, to distill themselves, scale.
- [42:28]
Tameh Besaroglu
Maybe. Before we talk about that, I think just a very important point to make here, which I think underlies some of this disagreement that we have with others about both this argument from the transition from kind of non human animals to humans is this focus on intelligence and reasoning and R and D, which is enabled by that intelligence as being just enormously important. And so if you think that you get this very important difference from this transition from primates, non human primates to, to humans, then you think that in some sense you get this enormously important unlock from fairly small scaling in say, brain size or something. And so then you might think, well, AI could be if we scale beyond the size of training runs, the amount of training compute that the human brain uses, which is maybe on the order of 1,824 flop or whatever, which we've recently surpassed, then maybe surpassing it just a little bit more, enables us to unlock very sophisticated intelligence in the same way that humans have much more sophisticated intelligence compared to non human primates. And I think part of our disagreement is that intelligence is kind of important, but just having a lot more intelligence and reasoning and good reasoning isn't something that will kind of accelerate technological change and economic growth very substantially. It isn't the case that the world today is just totally bottlenecked by not having enough good reasoning. And that's not really what's bottlenecking the world's ability to grow much more substantially. I think we might have some disagreement about this particular argument, but I think what's also really important is just that we have a different view as to how this acceleration happens. That it's not just having a bunch of really good reasoners that give you this technology that then accelerates things very drastically, because that alone is not sufficient. You need kind of complementary innovations in other industries. You need the economy as a whole growing and supporting the development of these various technologies. You need the various supply chains to be upgraded. You might need demand for the various products that are being built. And so we have this view where actually this very broad upgrading of your technology and your economy is important, rather than just having very good reasoners and very good reasoning tokens that gives us this acceleration.
- [45:05]
Keshav Murugesh
All right, so this brings us back to the intelligence explosion. Here is the argument for the intelligence explosion. Look, you're right that certain kinds of things might take longer to come about. But this core loop of software R and D that's required, if you just look at what kinds of progress is needed to make a more general intelligence, you might be right that it needs more experimental compute. But we're just getting, as you guys have documented, we're just getting a shit ton more compute every single year for the next few years. So you can imagine intelligence was over the next few years where in 2027 there'll be like 10x more compute than there is now for AI. And you'll have this effect where the AIs that are doing software R and D are finding ways to make running copies of them more efficient, which has two effects. One, you're increasing the population of AIs who are doing this research so more of them in parallel can find these different optimizations. And a subtle point that they'd often make here is software R and D and AI is not just ALIA type coming up with new transformer like architectures to your point, it actually is a lot of you gotta like. I mean I'm not an AI researcher, but I assume there's like from the lowest level libraries to the kernels, to making RL environments, to finding the best optimizer, to there's just so much to do. And in parallel you can be doing all these things or finding optimizations across them. And so you have two effects going back to this one is if you look at the original GPT4 compared to the current GPT4O, I think it's like how much cheaper is it to run?
- [46:59]
Tameh Besaroglu
It's like 100.
- [47:00]
Keshav Murugesh
Yeah.
- [47:00]
Tameh Besaroglu
So you have some times for the same capability or something.
- [47:03]
Keshav Murugesh
So they're finding ways in which to run more copies of them at like 100x cheaper or something. Which means that the population of them is increasing and the higher population is then helping you find more efficiencies. Not only does that mean you have more researchers, but to the extent that what's the complementary input is experimental compute, it's not the compute itself, it's the experiments. And the more efficient it is to run a copy or to develop a copy, the more parallel experiments you can run. Because now you can do a GPT 4 scale training run for much cheaper than you could do it in 2024 or 2023. And so for that reason also this software only singularity sees more researcher copies who can run experiments for cheaper. They initially are maybe handicapped in certain ways that you mentioned, but through this process they are rapidly becoming much more capable. What is wrong with this logic?
- [47:57]
Tameh Besaroglu
So I think the logic, the logic seems fine. I think this is a decent way to think about this problem. But I think that it's useful to draw on a bunch of work that say economists have done for studying the returns to R and D. And what happens if you 10x your inputs, the number of researchers, what happens to innovation or the rate of innovation? And there they point out these kind of two effects where as you do more innovation then you get to kind of stand on top of the shoulders of giants and you get the benefit from past discoveries and it makes you as a scientist more productive. But then there's also kind of diminishing returns that the low hanging fruit has been picked and it becomes harder to make progress. And overall you can summarize those estimates as thinking about the kind of returns to research effort. And we've looked into the returns to research effort in software specifically and we look at a bunch of domains in traditional software or linear integer solvers or SAT solvers, but also in AI like computer vision and RL and language modeling. And there, if this model is true, that all you need is just cognitive effort, it seems like the estimates are a bit ambiguous about whether this results in this acceleration or whether it results in just merely exponential growth. And then you might also think about, well, it isn't just your research effort that you have to scale up to make these innovations because you might have complementary input. So as you mentioned, experiments are the thing that might kind of bottleneck you. And I think there's a lot of evidence that in fact these experiments and scaling up hardware is just very important for getting progress in the algorithms and the architecture and so on. So in AI this is true for software in general, where if you look at progress in software, it often matches very closely the rate of progress we see in hardware. So for traditional software we see about a 30% roughly increase per year, which kind of basically matches. And in AI we've seen the same until you get to the deep learning era and then you get this acceleration which in fact coincides with the acceleration we've seen compute scaling, which gives you a hint that actually the compute scaling might have been very important. Other pieces of evidence besides this coincidental rate of progress, other kind of pieces of evidence are the fact that innovation in algorithms and architectures are often concentrated in GPU rich labs and not in the GPU poor parts of the world like academia or maybe smaller research institutes. That also suggests that having a lot of hardware is very important. If you look at specific innovations that seem very important, the big innovations over the past five years, many of them have some kind of scaling or hardware related motivation. So you might look at the transformer itself was about how to harness more parallel compute. Things like flash attention was literally about how to implement the attention mechanism more efficiently, or things like the chinchilla scaling law. And so many of these big innovations were just about how to harness your compute more effectively. That also tells you that actually the scaling of compute might be very important. And I think there's just like many pieces of evidence that points towards this complementarity picture. So I would say that not only even if you assume that experiments are not particularly important, the evidence we have both from estimates of AI and other software, although the data is not great, suggests that maybe you don't get this kind of hyperbolic faster than exponential super growth in the overall algorithmic efficiency of systems.
- [51:56]
Keshav Murugesh
I'm not sure I buy the argument that because these two things, compute and EA progress have risen so concomitantly that this is a sort of causal relationship. So broadly, the industry as a whole has been getting more compute and as a result making more progress. But if you look at the top players, there's been multiple examples of a company with much less compute, but a more coherent vision, more concentrated research effort being able to beat an incumbent who has much more compute. So OpenAI initially beating Google DeepMind, and if you remember, there was these emails that were released between Elon and Sam and so forth, where we got to start this company because they've got this bottleneck on the compute. And look how much more compute Google DeepMind has. And then OpenAI made a lot of progress. Similarly now with OpenAI versus Anthropic and so forth. And then I think just generally your argument is just too outside view when we just do know a lot about, like, what is like, you're just like this very macroeconomic argument that I'm like, well, why don't we just ask the AI researchers?
- [53:01]
Tameh Besaroglu
I mean, AI researchers will often kind of overstate the extent to which just cognitive effort and doing research is important for driving these innovations, because that's often kind of convenient or useful. They will say the insight was derived from some kind of nice idea about statistical mechanics or some nice equation in physics that says that we should do it this way. But often that's kind of an ad hoc story that they tell to make it a bit more compelling to the reviewers.
- [53:37]
Keshav Murugesh
So Daniel mentioned this survey he did where he, Daniel Koko Kotalo, he asked a bunch of AI researchers, if you had 1/30 the amount of compute I need to do 1/30 because AIs will be, as opposed to, they think, 30 times faster. If you had 1/30 the amount of compute, how much would your progress slow down? And they say, I make a third of the amount of progress I normally do. So that suggests a pretty good substitution effect of if you get 1/10 of compute, your progress only goes down one third. I was talking to an AI researcher the other day who's just one of these cracked people, gets paid millions and tens of millions of dollars a year probably. And we asked him, how much do these AI models help you in domains you already are, how much does these AI models help you in AI research? And he said, in domains that I'm already quite familiar with, where just closer to autocomplete, it saves me four to eight hours a week. And then he said, but in domains where I'm actually less familiar, where it's like, I need to draw new connections. I need to understand how these different parts relate to each other and so forth. It saves me close to 24 to 36 hours a week, right? And then that's like current models. And I'm just like, he didn't get more computed, but it still saved him like a shit ton more time. Just like draw that forward. It's like, that's a crazy implication or crazy trend, right?
- [54:58]
Ege Erdo
I mean, I guess, have we seen, like, I'm skeptical of the claims that we have actually seen that much of an acceleration in the process of R and D. Like, these claims seem to me like they're not borne out by the actual data I'm seeing. So I'm not sure how much to trust them.
- [55:17]
Keshav Murugesh
I mean, on the general intuition that cognitive effort alone can give you a lot of AI progress, right? It seems like a big important thing the labs do is this science of deep learning, like scaling laws is just. I mean, it ultimately netted out an experiment, but the experiment is motivated by cognitive effort.
- [55:36]
Ege Erdo
So for what is worth, when you say that A and B are complementary, you're not saying like. Like, just as you can't get a lot of progress, just as A can bottleneck you, B can also bottleneck you. So when you say you need compute and experiments and data, but you also need cognitive effort, that doesn't mean the lab who has the most compute is going to win, right? That's a very simple point. Either one can be the bottleneck. I mean, if you just have a really dysfunctional culture and you don't actually prioritize using your compute very well and you just waste it, well, then you're not going to make a lot of progress, right? So it doesn't contradict the picture that someone with a much better vision, a much better team, much better prioritization, can make better use of their compute if someone else was just bottlenecked heavily on that part of the equation. The question here is, once you get these automated AI researchers and you start this software singularity, your software efficiency is going to improve by many orders of magnitude, while your compute stock, at least in sort of the short run, is going to remain fairly fixed. So how many ooms of improvement can you get before you become bottlenecked by the second part of the equation? And once you actually factor that in, how much progress should you expect? That's the kind of question I think it's hard for people to have good intuitions about this because people usually don't run the experiments. So you don't get to see at a company level or at an industry level what would have happened if the entire industry had 30 times less compute. Maybe as an individual, what would happen if you had three times less compute? You might have a better idea about that, but that's a very local experiment and you might be benefiting a lot from spillovers from other people who actually have more compute. So because this experiment was never run, it's sort of hard to get direct evidence about the strength of complementarity.
- [57:28]
Keshav Murugesh
What is your probability of, if we live in the world where we get AGI in 2027, that there is a software only singularity quite high, because you're conditioning on the.
- [57:38]
Tameh Besaroglu
Then you're conditioning on compute not being very large. So it must be that you get a bunch of software progress, right?
- [57:46]
Keshav Murugesh
You just have a bunch of leverage from algorithmic progress in that world. Okay, that's right. So then maybe the. Because I was thinking these are independent questions.
- [57:54]
Tameh Besaroglu
I think a call out that I want to make is I know that some labs do have multiple pre training teams and they give people different amounts of resources for doing the training and different amounts of cognitive effort, different size of teams. But none of that I think has been published. And I would love to see the results of some of those experiments. I think even that won't update you very strongly just because it is often just very inefficient to do this very imbalanced scaling of your factor inputs. And in order to really get an estimate of how strong these complementarities are, you need to observe these very imbalanced scale ups. And so that rarely happens. And so I think the data that bears on this is just really quite poor. And then the intuitions that people have also don't seem clearly relevant to the thing that matters about what happens if you do this very imbalanced scaling. And where does this net out?
- [58:54]
Keshav Murugesh
One question I have, which I like, that it would be really interesting if somebody can provide an example of is maybe through history there was some point at which because of a war or some other kind of supply shock, you had to ramp up production or ramp up some key output that people really cared about. While for some weird historical reason, many of the key inputs were not accessible to a ramp up, but you could ramp up one key input. I'm talking in very abstract terms, but you see what I'm saying, right? You need to make more bombers, but you ran out of aluminum and you just need to figure out something else to do and how successful these efforts have been or whether you just keep getting bottlenecked.
- [59:35]
Ege Erdo
Well, I think that is not quite the right way to do it because I think if you're talking about materials, then I think there's a lot of sense in which different materials can be substitutable for one another in different ways. You can use aluminum. I mean aluminum is a great metal for making aircraft because it's sort of light and durable and so on. But you can imagine that you make aircraft with worse metals and then it just takes more fuel and it's less efficient to fly. So there's a sense in which you can compensate and just cost more. I think it's much harder if you're talking about something like complementarity between labor and capital or complementary between remote work and in person work or skilled or unskilled work. There are input pairs for which I would expect it to be much more difficult. Or for example, you're looking at the complementarity between the quality of leadership of an army and its number of soldiers. Right. I mean there is some effect there, but if you just scale up, you just have excellent leadership. But your army only has 100 people, you're not going to get very far.
- [60:40]
Keshav Murugesh
King Leonidas and Thermopylae, well, they lost, right? It would be funny if we're building models of software only singularity and we're like what exactly happened in Thermopylae? It's somehow relevant.
- [60:54]
Ege Erdo
I mean I can actually talk about that, but we probably shouldn't.
- [60:57]
Keshav Murugesh
Okay, sure. By the way, so the audience should know. My most popular guest by far is Sarah Payne. Not only is she my most popular guest, she's my most popular four guest because all four of those episodes that I've done with her are from a viewer minute adjusted basis. I host the Sarah Payne podcast where I occasionally talk about AI and anyways, we did this three part lecture series where we're talking about like one of them was about India, Pakistan, wars through history. One of them was about was it the Japan, like Japanese culture before World War II. The third one was about the Chinese civil war. And for all of them, my tutor, my history tutor was Ege and it just like why does he know so much about like fucking random like 20th century conflicts? But he did and he suggested a bunch of the good questions I asked her. We'll get into that. Actually. What's going on there?
- [61:57]
Ege Erdo
I don't know. I mean I don't really have a good question. I think it's interesting. I mean I read a bunch of stuff but it's kind of a boring answer. I don't know. Imagine you ask a top AI researcher what's going on?
- [62:09]
Tameh Besaroglu
How are you so good?
- [62:09]
Ege Erdo
And then they will probably give you a boring answer. I don't know. I did this.
- [62:14]
Keshav Murugesh
That itself is interesting that often these kinds of questions elicit boring answers. It tells you about the nature of the skills. How'd you find him?
- [62:23]
Tameh Besaroglu
We connected on a discord for Metaka, which is this forecasting platform. And I was a graduate student at Cambridge at the time, doing research in economics. And I was having conversations with my peers there, and I was occasionally having conversations with ege and I was like, this guy knows a lot more about economics. And at the time he was a computer science undergrad in Ankara, and he knows more about economics and about these big trends in economic growth and economic history than almost any of my peers at university. And so what the hell is up with that? So we started having frequent collaborations and ended up hiring EGE for Epoch because clearly makes sense for him to work on these types of questions.
- [63:18]
Keshav Murugesh
And it seems like at Epoch, you've just collected this group of Internet misfits and weirdos. How did you start Epoch and then how did you accomplish this?
- [63:27]
Tameh Besaroglu
Yeah, so I was at MIT doing more research, and I was pretty unhappy with the bureaucracy there, where it was very hard for me to scale projects up, hire people. And I was pretty excited about a bunch of work that my PI wasn't excited about because it's maybe hard to publish or it doesn't confer the same prestige. And so I was chatting with Jaime Sevilla, one of the co founders, and we just collaborated on projects and then thought we should just start our own org because we can just hire people and work on the projects we were excited about. And then I just hired a bunch of the insightful misfits that like.
- [64:12]
Keshav Murugesh
But did you, like, was the thesis like, oh, there's a bunch of underutilized Internet misfits and therefore this org was successful, or you started the org and then you're like, I think it's more of the latter.
- [64:21]
Tameh Besaroglu
So it was more like, we can make a bunch of progress, because clearly academia and industry is kind of dropping the ball on a bunch of important questions that academia is unable to publish interesting papers on. Industry is not really focused on producing useful insights. And so it seemed like very good for us to just do that. And also the timing was very good. So we started just before ChatGPT and we wanted to have much more grounded discussions of the future of AI. And I was frustrated with the quality of discussion that was happening on the Internet about the future of AI. And I mean, to some extent or to a very large extent, I still am. And that's like a large part of what motivates me to do this. It's just born out of frustration with bad thinking and arguments about where AI is going to go.
- [65:15]
Keshav Murugesh
The part about my job that I enjoy the least is the post production. I have to rewatch the episode multiple times, make all these difficult judgment calls and I've been trying to automate all this work with LLM scripts and I found that Google's Gemini 2.5 Pro is the best model I've tried for these tools. So much of the post production requires understanding the delivery, the context, all these other things that you don't get from a text only transcript. Unlike other models I've tested, I can actually just shove in the four hour Raw audio file into Gemini because of its multimodal capabilities and it can generate super high quality transcripts, identify great snippets for clips, clips, a bunch more other things. I've actually made a repo with all these tools and I've linked the GitHub in the description below in case you might find it helpful. I actually use 2.5 Pro in order to write the code for these scripts. It's actually quite interesting to read its reasoning traces as it's thinking through your gnarly list of requests and tasks. Gemini 2.5 Pro is now available in preview with higher rate limits. You can try it out at aistudio Google. Thanks to Google for sponsoring this episode. And now back to EGE and tame. Okay, so let me ask you about this. I can poke you from the so just to set the scene for the audience, we're going to talk about the possibility of this explosive economic growth and like greater than 30% economic growth rates. So I want to poke you both from perspective of maybe suggesting that this isn't aggressive enough in the right kind of way because maybe it's too broad. And then I'll poke you in from the perspective of the more normal perspective that hey, this is fucking crazy.
- [66:54]
Ege Erdo
I imagine it would be difficult for you to do the second thing.
- [66:59]
Keshav Murugesh
No, I mean, I think it might be fucking crazy. Let's see the big question I have about this broad automation. I get what you're saying about the industrial revolution, but in this case we can just make this automatic argument that you get this intelligence and then what you do next is you go to the desert and you build this shenzhen of robot factories which are building more robot factories which are building. If you need to do experiments and you build biolabs and you build chemistry labs and whatever.
- [67:30]
Ege Erdo
Shenzhen at the desert. I agree that looks much more plausible than a software only singularity.
- [67:36]
Keshav Murugesh
But why, but the way you're framing it, it sounds like McDonald's and Home Depot and fucking whatever are growing at 30% a year as well, and not just like the aliens level view of the economy. Is it that there's a robot economy in the desert that's growing at 10,000% a year and everything else is the same old, same old? Or is it like, I mean there.
- [67:59]
Ege Erdo
Is a question about what would be possible or physically possible and what would be the thing that would actually be efficient.
- [68:07]
Keshav Murugesh
Right.
- [68:08]
Ege Erdo
So it might be the case. And again, once you're scaling up the hardware part of the equation as well as the software part, then I think the case for this feedback loop gets a lot stronger. If you scale up data collection as well, I think it gets even stronger. Like real world data collection by deployment and so on. But building Shenzhen in the desert, that's a pretty. If you think about the pipeline. So so far we have relied first on, we're relying on the entire semiconductor supply chain. That industry depends on tons of inputs and materials and whatever it gets from probably tons of random places in the world. And creating that infrastructure, like doubling or tripling whatever that infrastructure, like the entire thing, that's very hard work, right? So probably you couldn't even do it even if you just have Shenzhen. Like that would be even more expensive than that. On top of that, so far we have been drawing heavily on the fact that we have built up this huge stock of data over the past 30 years or something on the Internet. Imagine you were trying to train a state of the art model, but you only have 100 billion tokens to train on. That would be very difficult. So in a certain sense, our entire economy has produced this huge amount of data on the Internet that we are now using to train the models. It's plausible that in the future, when you need to get new competencies added to these systems, the most efficient way to do that will be to try to leverage similar kind of modalities of data, which will also require this. You would want to deploy the systems broadly because that's going to give you more data and maybe you can do the, maybe you can get where you want to be without that, but it would just be less efficient if you're starting from scratch compared to if you're collecting A lot of data. I think this is actually a motivation for why Labs want their LLMs to be deployed widely. Because sometimes when you talk to ChatGPT, it's going to give you two responses and it's going to say, well, which one was good? Or it's going to give you one response and it's going to ask you if was this good or not? Well, why are they doing that? That's a way in which they are getting user data through this extremely broad deployment. So I think you should just imagine that thing to continue to be efficient and continue to increase in the future, because it just makes sense. And then there's a separate question of, well, suppose you didn't do any of that. Suppose you just tried to imagine the most rudimentary, the most narrowest possible kind of infrastructure build out and deployment that would be sufficient to get this positive feedback loop that leads to much more efficient AIs. I agree that loop could in principle be much smaller than the entire world. I think it probably couldn't be as small as Transcendent desert, but it could be much smaller than the entire world. But then there's a separate question of would you actually do that? Would that be efficient? I think some people have the intuition that there are just these extremely strong constraints, maybe regulatory constraints, maybe social political constraints to doing this broad deployment. They just think it's going to be very hard. So I think that's part of the reason why they imagine these more narrow scenarios where they think it's going to be easier. I think that's overstated. I think people's intuitions for how hard this kind of deployment is comes from cases where the deployment of the technology wouldn't be that valuable. So it might come from housing. We have a lot of regulations on housing. Maybe it comes from nuclear power, maybe it comes from supersonic flights. I mean, those are all technologies that would be useful if they were maybe less regulated, but they wouldn't double economic output.
- [71:54]
Tameh Besaroglu
I think the core point here is just the value of AI automation and deployment is just extremely large, even just for workers. At least the ones that at least after finding there might be some kind of displacement and there might be some transition that you need to do in order to find a job that works for you. But otherwise the wages could still be very high for a while at least. And on top of that, the gains from owning capital might be very enormous. And in fact, a large share of the US population would benefit from housing, for example. They benefit, they own housing, they have 401ks, those would do enormously better when you have this process of broad automation and AI deployment. And so I think there could just be a very deep support for some of this, even when it's totally changing the nature of labor markets and the skills and occupations that are in demand.
- [72:55]
Ege Erdo
So I would just say it's complicated. I think what the political reaction to it will be when this starts actually happening. I think the easy thing to say is that, yeah, this will become a big issue and then it will be maybe controversial or something. But what is the actual nature of the reaction in different countries? I think that's kind of hard to forecast. I think the default view is, well, people are going to become unemployed, so it will just be very unpopular. I think that's very far from obviously, and I just expect heterogeneity in how different countries respond. And some of them are going to be more liberal about this and going to allow broader deployments, and those countries probably end up doing better. Just like during the Industrial Revolution, some countries were just ahead of others. I mean, eventually almost the entire world adopted the sort of norms and culture and values of the Industrial Revolution in various ways.
- [73:44]
Tameh Besaroglu
And actually you say they might be more liberal about it, but they might actually be less liberal. They might be less liberal in many ways. And in fact, that might be more functional. In this world in which you have broad AI deployment, we might adopt the kind of values and norms that get developed in say, the UAE or something, which is maybe focused a lot more on making an environment that is very conducive for AI deployment. We might start emulating and adopting various norms like that. And they might not be kind of classical liberal norms, but norms that are just more conducive to AI being functional and producing a lot of value.
- [74:27]
Ege Erdo
This is not meant to be like a strong prediction. It's just an illustrative. It might just be that the freedom to deploy AI in the economy and build out lots of physical things at scale, maybe that ends up being more important in the future. Maybe that is still missing something. Maybe there's other things that are also important. But I would just the generic prediction that you, you should expect variance and some countries do better than others. I think that's much easier to predict than the specific countries that end up doing better.
- [74:55]
Keshav Murugesh
Yeah. Or the norms that that country will necessarily have. So, I mean, one thing I'm confused about is if you look at the world of today versus the world of 1750, the big differences is just like we've got crazy tech that they didn't have back then. We've got these cameras and we've got these screens and we've got rockets and so forth. And that just seems like the result of technological growth and R and D and so forth, capital accumulation. Well, explain that to me because you're just talking about this infrastructure build out and blah, blah, blah. But why won't they just fucking invent the kinds of shit that humans would have invented by 2000, 2050?
- [75:38]
Ege Erdo
Producing this stuff takes a lot of infrastructure buildup.
- [75:40]
Keshav Murugesh
It's not. But that infrastructure is built out once you make the technology, right?
- [75:45]
Tameh Besaroglu
I don't think that's right. There isn't this temporal difference where it's first you do the invention and then often there's this interplay between the actual capital buildup and the innovation.
- [75:58]
Ege Erdo
Learning curves are about this, right? Fundamentally what has driven the increase in the efficiency of solar panels over the past 20, 30 years.
- [76:05]
Tameh Besaroglu
It isn't just people had the idea of 2025 solar panels. No one ever had. Nobody 20 years ago had the sketch for the 2025 solar panel. It's this kind of interplay between having ideas, building, learning, producing and other complementary.
- [76:25]
Ege Erdo
Inputs also becoming more efficient at the same time. Like you might get better materials. Like for example, the fact that aluminum becomes something that's like for example, the fact that smelting processes got a lot better towards the end of the 19th century, so it became a lot easier to work with metal. Maybe that was a crucial reason why aircraft technology later became more popular. So it's not like someone came up with the idea of, oh, you can just use something that just has wings and has a lot of thrust and then that might be able to fly. That basic idea is not that difficult. But then, well, how do you make it like actually a viable thing?
- [77:03]
Keshav Murugesh
That's right.
- [77:03]
Ege Erdo
Well, that's much more difficult.
- [77:04]
Keshav Murugesh
Have you seen the meme where two beavers are talking to each other and they're looking at the Hoover Dam and one of them's like, well, I didn't build that, but it's based on an idea of mine. That's right. The point you're making is that this invention focused look on tech history, underplays the work that goes into making specific innovations practicable and to deploy them widely.
- [77:34]
Ege Erdo
It's just hard, I think it's just hard to even suppose you wanted to write a history of this. You want to write a history of how was the light bulb developed or something. It's just really hard because to understand why specific things happen at Specific times, you probably need to understand so much about the economic conditions of the time. Like, for example, Edison spent a ton of time experimenting with different filaments to be used in the light bulb. The basic idea is very simple. You make something hot and it glows, but then what filament actually works well for that in a product? What is durable? What has the highest ratio of light output versus heat so that you have less waste and it's more efficient. And then even after you have the products, then you're faced with a problem. Well, I mean, it's like 1880 or something. And then US homes don't have electricity, so then nobody can use it. So now you have to build power plants and build power lines to the houses so that people have electricity in their homes so that they can actually use this new light bulb that you created. So he did that, but then people present it as if it's like, okay, he just came out with the idea like, it's a light bulb.
- [78:45]
Keshav Murugesh
Well, I guess the thing people would say is, like, you're right about how technology would progress if we were humans deploying for the human world. But what you're not counting is there's just going to be this AI economy where maybe they need to do this kind of innovation and learning by doing when they're figuring out, I want to make more robots because they're helpful. And so we're going to build more robot factories, we'll learn, and then we'll make better robots or whatever. But. But, like, that is just, like, geographically, that is a small part of the world that's happening in, or. You understand what I'm saying? Like, it's not like. And then they walk into your building, and then you do a business transaction with Lunar Society Podcast, llc, and then. You know what I mean?
- [79:30]
Ege Erdo
Yeah. I mean, for what it's worth, like, if you look at the total surface area of the world, it might well be the case that the place that initially experiences this very fast growth is like a small percentage of the surface area of the world. But I think that was the same for an industrial revolution, was not different.
- [79:46]
Keshav Murugesh
Yeah. So. But then I'm just like, what concretely does this explosive growth look like? If I look at this heat map of growth rates on the globe, is there just going to be, like, one area is, like, blinding hot, and that's like the desert factories with all these experiments and, like. Yeah.
- [80:04]
Ege Erdo
So I would say our idea is that it's going to be broader than that, but probably initially. So eventually it would be probably most of the world. But as I said, because of this heterogeneity, because I think some countries are going to be faster in adoption than others, maybe some cities will be more faster adoption than others. And that will mean that there is differentials. And some countries might have much faster growth than other countries. But I would expect that at a jurisdiction level it would be more homogenous. So for example, I expect the primary obstacles to come from things like regulation. And so I would just imagine it's being more delineated by regulatory jurisdiction boundaries than anything else.
- [80:48]
Keshav Murugesh
Got it. So you may be right that this infrastructure buildout and capital deepening and whatever is necessary for a technology to become.
- [80:56]
Ege Erdo
Practical or even to be discovered, there's an aspect of it where you discover certain things by scaling up learning, by doing writing, learning curve. And there's this separate aspect where you get to suppose that you become wealthier. Well, you can invest that increased wealth in, yeah, you use it to accumulate more capital, but you also can invest it in R and D and other ways.
- [81:21]
Tameh Besaroglu
You get Einstein out of the patent office. You need some amount of resources for that to make sense. And you need the economy to be of a certain scale. You also need demand for the product you're building. So you could have the idea, but if the economy is just too small that there isn't enough demand for you to be specializing and producing the semiconductor or whatever because there isn't enough demand for it, then it doesn't make sense. So you want the economy like a much larger scale of an economy is useful in very many ways in delivering complementary innovations. And discovery is happening through serendipity producing like having there be consumers that would actually pay enough for you to recover your fixed costs of doing all the experimentation and the invention. You need the supply chains to exist to deliver the germanium crystals that you need to do grow in order to come up with the semiconductor. You need a large labor force to be able to help you do all the experiments and so on.
- [82:20]
Keshav Murugesh
I think that the point you're illustrating is like, look, could you have just figured out that there was a Big Bang by first principles reasoning? Maybe. But what actually happened is we had World War II and we discovered radio communications in order to fight and effectively communicate during the war. And then that technology helped us build radio telescopes. And then we discovered cosmic microwave background. And then we had to come up with an explanation for cosmic microwave. And then we discovered like the Big Bang as a result of like World War II communication. Exactly.
- [82:47]
Tameh Besaroglu
People underemphasize that like, giant effort that goes into this kind of buildup of all the relevant capital and all the relevant supply chains and the technology. I mean, earlier you were making a similar comment when you were saying, oh, reasoning models, actually, in hindsight they look pretty simple. But then you're kind of ignoring this giant kind of upgrading of the technology stack that happened that took five to 10 years prior to that. And. And so I think people just underemphasize the support that is had from the overall upgrading of your technology of the supply chains of various sectors that are important for that. And people focus on just specific individuals of like Einstein had this genius insight and he was the kind of very pivotal thing in the causal chain that resulted in these discoveries. Or Newton was just extremely important for discovering calculus without thinking about, well, there was this kind of all these other factors that produced lenses, that produced telescopes, that got the right data and that made people ask questions about dynamics and so on that motivated some of these questions. And those are also extremely important for science, scientific and technological innovation.
- [84:06]
Keshav Murugesh
Yeah, you know, conquests. What is it? One of conquest laws is the more you understand about a topic, the more conservative you become about that topic. And so there might be a similar law here where the more you understand about an industry, the more sort of. Obviously I'm just a commentator or whatever or a podcaster, but I understand AI better than any other industry. I understand. And there I have the sense from talking to people like you that, oh, so much went into getting AI to the point where it is today. Whereas when I talk to journalists about AI, they're like, okay, who is the crucial person we need to cover? And they're like, should we get in touch with Geoffrey Hinton? Should we get in touch with Ilya? And I just have this. You're kind of missing the picture. But then you should have that same attitude towards things. You Maybe the more similar phenomenon is German amnesia. We should have a similar attitude towards other industries that it's much more complicated.
- [85:01]
Ege Erdo
Right. I mean, so Robin Hanson has this abstraction of seeing things in near mode versus far mode.
- [85:06]
Keshav Murugesh
Right.
- [85:06]
Ege Erdo
And I think if you don't know a lot about the topic, then you see it sort of in far mode and you sort of simplify when things, you know, you see a lot more detail. Like in general, I think the thing I would say, and the reason I also believe that just like abstract reasoning and like sort of deductive reasoning or even Bayesian reasoning by itself is not sufficient or like is not as powerful as many other people think is because I Think there's just this enormous amount of richness and detail in the real world that you just can't reason about it. You need to see it. And obviously that is not an obstacle to AI being incredibly transformative, because as I said, you can scale your data collection, you can scale experiments. You do both in the AI industry itself and just more broadly in the economy. So you just discover more things. More economic activity means we have more exposed surface area to have more discoveries. All of these are things that have happened in our past, so there's no reason that they couldn't speed up. The fundamental thing is that there's no reason fundamentally why economic growth can't be much faster than it is today. It's probably about as fast right now just because humans are such an important bottleneck. They both supply the labor, they play crucial roles in the process of discovery of various kinds of productivity growth. There's just strong complementarity to some extent with capital that you can't substitute machines and so on for humans very well. So the growth of the economy and growth of productivity just ends up being bottlenecked by the growth of human population.
- [86:37]
Keshav Murugesh
Publicly available data is running out. So major AI labs like Meta, Google, DeepMind and OpenAI partner with Scale to push the boundaries of what's possible. Through Scale's data foundry, major labs get access to high quality data to fuel post training, including advanced reasoning capabilities. As AI races forward, we must also strengthen human sovereignty. Skio's research team SEAL provides practical AI safety frameworks, evaluates frontier AI system safety via public leaderboards, and creates foundations for integrating advanced AI into society. Most recently, in collaboration with the center for AI Safety, Scale published humanity's last examination, a groundbreaking new AI benchmark evaluating expert level reasoning and knowledge capabilities across a wide range of fields. If you're an AI researcher or engineer and you want to learn more about how Skale's data foundry and research lab can help you go beyond the current frontier of capabilities, go to scale.comdwarKesh. let me ask a tangential question. What has happened in China over the last 50 years? Would you describe that as like in principle, the same kind of explosive growth that you expect from AI? Because there's like a lot of labor that makes the marginal product of capital really high, which allows you to have like 10% plus economic growth rates? Is that basically in principle from AI?
- [88:02]
Ege Erdo
So I would say in some ways it's similar, in some ways it's not the way. Probably the most important way which is not similar is that in China you see this relative like you see a massive amount of capital accumulation, substantial amount of adoption of new technologies, and probably also human capital accumulation to some extent. But you're not seeing a huge scale up in the labor force effect. Labor force. While for AI you should expect to see a scale up in the labor force as well. Not in the human workforce, but in the AI workforce.
- [88:34]
Keshav Murugesh
I think you did kind of like maybe not consecutive increases in the labor force increase, but you did.
- [88:40]
Tameh Besaroglu
The key thing here is just the simultaneous scaling of both these things.
- [88:44]
Keshav Murugesh
Got it.
- [88:44]
Tameh Besaroglu
And so you might ask the question of isn't it basically half of what's going to happen with AI that you scale up capital accumulation in China? But actually that's really not. If you get both of these things to scale, that gives you just much faster growth and a very different picture.
- [89:04]
Ege Erdo
But at the same time, if you're just asking what would 30% growth per year look like in terms of. If you're just going to have an intuition for how transformative that would be in concrete terms, then I think looking at China is not such a bad case, especially in the 2000s or maybe late 90s, that gives you a good. That's even slower than what we're forecasting.
- [89:24]
Tameh Besaroglu
Right. I think also looking at the Industrial Revolution is pretty good.
- [89:26]
Ege Erdo
Well, the Industrial Revolution is very slow.
- [89:29]
Tameh Besaroglu
But just in terms of the types of the kind of the margins along which we made progress in terms of products. So what didn't happen, the thing that didn't happen during the Industrial Revolution is we just produced a lot more of things that people were producing prior to the Industrial Revolution, like producing a lot more crops and maybe a lot more kind of pre Industrial revolution style houses or whatever on farms. Instead. What we got is along pretty much every main sector of the economy, we just had many different products that are totally different from what was being consumed prior to that. So in transportation, in food, I mean.
- [90:13]
Ege Erdo
Healthcare is a very big deal. Antibiotics.
- [90:16]
Keshav Murugesh
So another question, because I'm not sure I understand how you're defining the learning by doing versus explicit R and D. Because there's the way for taxes that companies say what they call R and D. But then there's the intuitive understanding of R and D. So if you think about how AI is boosting tfp, you could say that right now if you just had replaced the TSMC process engineers with AIs and they're finding different ways in which to improve that process and improve efficiencies, improve yield, I would kind of call that R and D on the Other hand, you emphasize this, the other part of tfv, which is like better management and learning by doing that kind of stuff.
- [90:59]
Ege Erdo
Learning by doing could be you could.
- [91:00]
Keshav Murugesh
I mean, how much omfar are you going to like you're going to get to the fucking Dyson sphere by better management.
- [91:06]
Ege Erdo
But that's not the argument. Right? The point is that there are all these different things, that some of them are maybe more complementary than others. The point is not that you can get to a Dyson sphere by just scaling labor and capital. That's not the point. Like you need to scale everything at once. So just as you can't get to a license sphere by just scaling labor and capital, you also can't get to it by just scaling tfp. That doesn't work.
- [91:31]
Tameh Besaroglu
I think there's a very important distinction between what is necessary to scale to get to this Dyson sphere world and what is important. Like in some sense producing food is necessary, but of course producing food doesn't get you to Dyson sphere. Right. So I think R and D is necessary, but on its own isn't sufficient. And scaling up the economy is also necessary. On its own, it's not sufficient. And then you can ask the question what is the relative importance of each?
- [92:00]
Keshav Murugesh
Yeah.
- [92:00]
Ege Erdo
So I think our view here is very much the same. It is very connected to our view about the software R and D thing where we're just saying there are these bottlenecks, so you need to scale everything at once. This is just a general view, but. But I think people misunderstand us sometimes as saying that R and D is not important. No, that's not what we're saying. We're saying it is important. It is less important in relative terms than some other things, none of which are by themselves sufficient to enable this growth. So the question is, how do you do the credit attribution? One way in economics is standard to do that is to look at the elasticities of outputs to the different factors. Capital is less important than labor because the output elasticity of labor elasticity output is like 0.6, while for capital is like 0.3. But neither are by themselves sufficient. If you just scaled one of them and the other remained fixed, then neither would be sufficient to indefinitely scale outputs.
- [93:00]
Keshav Murugesh
One question that Daniel posed to me is because I made this perspective about everything being interconnected when you were talking about like another example people often bring up is what would it take to build the iPhone in the year 1000? And it's like unclear how you could actually do that without just like replicating every intermediate technology or most intermediate technologies. And then he made the point like, okay, fine, whatever. Nanobots. Like nanobots is not a crux here. The crux, at least to the thing he cares about, which is human control, is just by when can the robot economy or the AI economy, whether it's a result of capital deepening or whether it's a result of R and D, by when will they have. Just like the robots and they have more sort of like cumulative physical power. Right.
- [93:50]
Ege Erdo
But he's imagining like a separate thing called the AI economy. Why would you imagine that? That seems like a. I think it's probably downstream of his views about the software only singularity. But again, like, those are views that we don't share.
- [94:03]
Tameh Besaroglu
It's just much more efficient for AI to operate in our economy and benefit from the existing supply chains and existing markets, rather than set up shop on some island somewhere and do its own thing.
- [94:16]
Ege Erdo
Yeah. And then it's not even clear. For example, people might have the intuition, I brought this up before, the distinction between what is the minimum possible amount of build out that would be necessary to get this feedback loop up and running and what would be the most efficient way to do it, which are not the same question. But then people have this view that, oh, the most efficient thing, in principle, we can't do that.
- [94:36]
Keshav Murugesh
Because I think the example he might give is when the Conquistadors arrived to the New World or when the East India Trading Company arrived to India, they did integrate into the existing economy in many cases. It depends on how you define integrate. But the Spanish relied heavily on New World labor in order to do silver mining and whatever East India Trading Company was. Like the ratio of British people to Indian people was just not that high. Right. So they just had to rely on the existing labor force. But they were still able to take over because of. I don't know what the analogous thing here is, but you see what I'm saying? And so he's concerned about by when will they. Even if they're ordering components off of Alibaba or whatever. Sorry, I'm being tried. You see what I'm saying? Even if they're integrated in those supply chains, by when are they in a position where, because this part of the economy has been growing much faster, they could take over the government if they wanted to. That's right. Yeah.
- [95:42]
Ege Erdo
Okay. So I think that is eventually you expect the AI systems to be driving most of the economy. And I don't think that's. Unless there are some very strange coincidences where humans are able to somehow uplift themselves and able to become competitive with the AIs by stopping being biological humans or whatever, which seems very unlikely early on, then AI is just going to be much more powerful. And I agree that in that world, if the AI is just somehow coordinated and decided, okay, we should just take over or something, just somehow coordinated to have that goal, then they could probably do it. But okay, that's also probably true in our world. In our world, if the US wanted to invade Sentinel island, then probably they could do it. I don't think anyone could stop them. But what does it actually mean? I mean, there is this dramatic power imbalance, but that doesn't mean, that doesn't tell you what's going to happen. Right. Why doesn't the US just invade Guatemala or something? Why don't they do that? Seems like they could easily do it.
- [96:53]
Keshav Murugesh
Because the value to the US of land. Yeah. So basically it just seems. I agree that might be true for AIs because most of the shit is in space. And once you get like, you want to do the capital deepening on Mars and like the surface area of the sun instead of like, instead of New.
- [97:13]
Ege Erdo
York City, so it's deeper than that. There's also the fact that if the AIs are going to be integrators into our economy, so basically they start out as like a smaller part of our economy or our workforce, and over time they grow. And over time they might, they become the vast majority of the actual sort of work power in the economy, but they are growing in this existing framework where we have norms and rules for better coordination and then undermining those things has a cost. So if getting the things that is making the humans sort of wealthier than they used to be before and more comfortable, yeah, you would probably be better off if you could just take that from them. But the benefit to you, if you already are getting almost all of the income in the economy will be fairly small.
- [98:03]
Keshav Murugesh
I mean, I feel like the Sentinel thing is not the. There's one reference class that includes that, but historically there's a huge reference class that includes East India Trading Company. Could have just kept trading with the Mughals. They just took over. Right. They could have kept trading with the 50 different nation states in pre colonial India. But. Yeah, that's right.
- [98:23]
Ege Erdo
I mean, that's what they were initially doing and then whatever. I'm not going to.
- [98:28]
Keshav Murugesh
That is the reference class of.
- [98:30]
Ege Erdo
I agree. So if the question is if they are entirely. If they have some totally different values and then they represent most of the economy. Then would they take over? I still don't know because I'm not sure to what extent the class of all AI is a natural, like, class. It's sort of like, why don't the young people in the economy?
- [98:55]
Keshav Murugesh
I think so. I agree that sometimes these kinds of class arguments are misused. For example, when Marxists are like, why don't this class uprise against the others? Dana made the interesting argument that if you look at the history of the Conquistadors, when Cortes was making his way through the New World, he had to actually go back and fight off a Spanish fleet that had been sent to arrest him.
- [99:18]
Ege Erdo
That's right.
- [99:18]
Keshav Murugesh
And then go back. Right. So you can have this fight within this, like, conquering AIs. And then that still nets out in the Native Americans getting disempowered. But with AIs in particular, they're just like copies of each other in many other ways. They just have a lot. They have lower transaction costs when they trade with each other or interact with each other. There's other reasons to expect them to just be more compatible, coordinating with each other than coordinating with the human world.
- [99:48]
Ege Erdo
Sure. But then I guess I'm still not seeing the. I mean, if the question is just that, is it possible for that to happen, which is like a weaker claim, then yeah, I mean, it seems possible. But there are, I think, a lot of arguments just pushing back against it. Probably. Actually, the biggest one is the fact that AI preferences are just not like, just look at the AIs we have today. Can you imagine them doing that? I think people just don't put a lot of weight on that because they think once we have enough optimization pressure, and once they become super intelligent, they're just going to become misaligned. But I just don't see the evidence for that.
- [100:23]
Keshav Murugesh
No, I mean, I think that's actually. I agree. There's some evidence that they're like, good boys.
- [100:28]
Ege Erdo
No, there's more than some evidence.
- [100:30]
Keshav Murugesh
No, but there's also some evidence. There was a new OpenAI paper where in chain of thought, reward hacking is such a strong basin that if you were like, hey, let's go solve this coding problem in the train of thought, they'll just be like, okay, let's hack this and then figure out how to hack it.
- [100:48]
Ege Erdo
So imagine that you gave students at a school a test, and then the answer key was like, Right.
- [100:53]
Keshav Murugesh
But the reference class of humans does include Cortez and the East Indian Trading Company.
- [100:58]
Ege Erdo
Sure.
- [100:58]
Tameh Besaroglu
So I think one Issue here is that I think people are doing this very kind of partial equilibrium analysis or something where they're thinking about just this raw abilities of AI systems in a world where AI systems are kind of dominant and human civilization has done very little in terms of integrating itself. And the AI is integrating itself into the human world, maybe making, insofar as it's poor at communicating and coordinating with AIs, addressing those deficiencies and improving that, insofar as that's posing a risk or creating inefficiencies because it's unable to benefit from coordinating and trading, then it should have this enormous incentive to address that. Insofar as there is a lot of value to be gained from dominating and taking over humans. Like what you might get is a more negotiated settlement. If that's indeed the case, then a war would just be inefficient. And so you would want to negotiate some settlement that results in some outcomes that are mutually beneficial compared to the.
- [102:06]
Keshav Murugesh
Counterfactual, not compared to. There was a mutually beneficial trade that was made between the Qing Dynasty and the British in the Opium wars, right? But there was like it was maybe better than pre industrial China going to war with the British Empire, but it wasn't better than never having interacted with the British Empire in the first place.
- [102:28]
Tameh Besaroglu
So I think one mistake that I feel people make is they have this very naive analysis of what creates conflict. And I think Matthew has written a bit about this, a colleague of ours, where they say there's misalignment and so that that then creates conflict. But that's actually not what the literature on what causes conflict says creates conflict. It's not just misalignment. It's also other issues like not being able, having bad understanding of the relative strengths of your armies versus theirs, or maybe having these very strong commitments that you think some grounds are sacred. And so you're not willing to do any trade in order to give up some of that in order to gain something else. And so then you have to posit some additional things other than just the base value misalignment part.
- [103:28]
Keshav Murugesh
I think you're making a good argument against like humans take up the spears and the machetes and go to war against the AI data centers. Because maybe there's not this asymmetric information that often leads to conflicts in history, but this argument does not address at all the risk of just takeover, which can be the result of a peaceful end negotiation. Or human society is just like, look, we're totally outmatched. And let's just Take these meager concessions rather than going.
- [103:57]
Tameh Besaroglu
But insofar as it's more peaceful, then I think it's much less of a thing to worry about. I think there could be this trend where we indeed have this gradual process where AI is much more important in the world economy and actually deciding and determining what happens in the world. But this could be beneficial for humans where we're getting access to this much larger economy and much more advanced technological stock.
- [104:30]
Ege Erdo
Yeah, so I think it's important to be clear about what is the thing that you're actually worried about. I think some people just say that, oh, humans are going to lose control of the future. We're not going to be the ones who are making the important decision. We, however, conceive that's also kind of nebulous. Okay, so is that something to worry about? If you just think biological humans should remain in charge of all important decisions forever, then I agree the development of AI seems like kind of a problem for that. But in fact, other things also seem like. I know, a problem for that. I just don't expect to generically be true like a million years from now. Even if you don't develop AI, biological humans, the way we recognize them today, are still making all the important decisions and they have something like the culture that we would recognize from ourselves today. I'll be pretty surprised by that. So I think Robin Hanssen has again talked about this where he said a bunch of the things that people fear about AI are just things they fear about change and fast change. So the thing that's different is that AI has a prospect of accelerating a much of this change so that it happens in a narrower period.
- [105:36]
Keshav Murugesh
I think there's not the arg. I think it's not just the kind of change that would have happened from, let's say, genetically modifying humans and blah, blah, blah, is instead happening in a compressed amount of time. I think the worry comes more from it's not just that change compressed, it's a very different vector of change.
- [105:54]
Ege Erdo
Yeah, but what is the argument for this? Yeah, I have never seen a good argument for this.
- [105:59]
Tameh Besaroglu
You should expect a bunch of change. If you accelerate just human change as well, you might expect different values to become much more dominant. You might expect people that don't discount the future as much to be much more influential because they save more and they make good investments.
- [106:16]
Ege Erdo
It gives them more control, higher risk tolerance.
- [106:18]
Tameh Besaroglu
Higher risk tolerance because they more willing to make bets that maximize expected value and so get much more influence. So just generically accelerating human change would also result in a lot of things being lost that you might care about.
- [106:36]
Keshav Murugesh
I think the argument is that maybe the speed of the change determines what fraction of the existing population or stakeholders or whatever have some causal influence on the future. And maybe the thing you care about is like, look, there's going to be change, but it's not just going to be like one guy presses a button that's like the software singularity extreme.
- [107:03]
Ege Erdo
Right.
- [107:03]
Keshav Murugesh
And it's more like over time, norms change and so forth.
- [107:08]
Ege Erdo
So if you're looking at the software singularity picture, I agree that picture looks different. And again, I'm coming back to this because obviously Daniel and maybe Scott, to some extent they probably have this view that the software only singularity is more plausible. And then one person could just be in a position to. We could end up in a situation where their idiosyncratic preferences or something end up being influential. Yeah, even in. I agree that makes the situation look different from if you just have this broader process of automation. But even in that world, I think a lot of people have this view about things like value lock in, or they think this moment is like a pivotal moment in history. And then we're just going to have someone is going to get this AI, which is very powerful because say, of its software on my singularity, and then they're just going to lock in some values and then those values are just going to be stable for millions of years. And I think that just looks very unlike anything that has happened in the past. So I'm kind of confused why people think it's very plausible. I think people have the argument that they see the future again, in my view, in sort of far mode, they think there's going to be one AI, it's going to have some kind of utility function. That utility function is going to be very stable over time, so it's not going to change. There won't be this messiness of lack of coordination between different AIs or different over time values drifting for various reasons, maybe because they become less functional in an environment, maybe because of other reasons. So they just don't imagine that. They think, well, I mean, utility functions, we can preserve them forever. We have this technology to do that, so it's just going to happen. And I'm like, well, that seems like such a weak argument to me. I'm not sure.
- [108:51]
Tameh Besaroglu
Actually, if you look at just like often the idea is because this is digital, you can preserve the information better and copy it with higher fidelity and so on. But actually, if you look, even if you look at just like information on the Internet, you have this thing called link rot, which happens very quickly. And actually information that's digital isn't preserved for very long at all.
- [109:16]
Keshav Murugesh
And the point that Matthew was making is that also this fact that the information is digital has led to, not maybe led to, but at least been associated with faster cultural change.
- [109:25]
Tameh Besaroglu
Cultural change, exactly.
- [109:26]
Ege Erdo
I mean, basically technological changes can create incentives for cultural change just as they make preserve.
- [109:32]
Keshav Murugesh
Like, I mean, I think one of the. There's two key arguments that I've heard. One is that we will soon reach something called technological maturity.
- [109:40]
Ege Erdo
Yeah.
- [109:41]
Keshav Murugesh
And once you. One of the key ways in which society has been changing recently is that it's not like maybe actually its culture would have changed even more. Actually, no, no, I think this argument is wrong that you're making because we do know that language actually changed a lot more. Like, we can read everything that was written after like 1800s when literacy became more common, but it's actually even just go back a couple hundred years after that and you're reading Old English and it's hard to understand. And that is a result of literacy and the codification of language.
- [110:09]
Ege Erdo
Well, that information was better preserved. What about other kinds of cultural practices?
- [110:13]
Keshav Murugesh
But I think the argument would be that maybe they would have actually changed more if that change was a result of technological change in general, not the result of information being digitized. And maybe that culture would have actually changed more if information wasn't as well preserved or technology had continued to proceed. And the argument is, in the future we're going to reach some point at which you've done. All the tech ideas have just gotten way too hard to find. And you need entire galaxies worth of. You need to make a CERN that's the size of a galaxy to progress physics an inch forward. And at that point this growth in technology just churning over civilization goes away. And then you just have the digital thing, which does mean that a lock in is more plausible.
- [111:01]
Tameh Besaroglu
So the technological maturity thing, I agree. That results in this slowdown in change and growth and so on. And certain things might get more locked in relative to what preceded it. But then what do we do today about that? What could you do to have a kind of positive impact, by our lights? And so asking that question, I mean, Robin Hanson had this question of what could someone do in the 1500s to have a positive impact on the world today from their point of view, knowing all they knew back then? I think this question is even worse than that because I think the amount of change that happens between today and technological maturity is just orders of magnitude greater than whatever change happened between the 1500s and today. So it's an even worse position than someone in the 1500s thinking about what they could do to have an impact, positive impact in expectation, predictably positive today. And so I think it's just pretty hopeless. I don't know if we could do anything or find any candidate set of actions that would just make things better. Post lock in.
- [112:05]
Ege Erdo
I mean, that's assuming lock in isn't even going to happen, which is not.
- [112:08]
Keshav Murugesh
In the 1700s, a bunch of British abolitionists were making the case against slavery. And I think the world has. We could just live in a slave society. I don't think there's any in principle reason why we couldn't have been a slave society to this day or more. The world couldn't have slavery. And I think what happened is just like the convincing of British people that slavery is wrong. The British Empire put all its might into abolishing slavery and making that a norm. I think another example is Christianity and the fact that like Jesus has these ideals. You could talk about these ideals. I think the world is a more Christian place.
- [112:45]
Ege Erdo
Oh, it is a more Christian place, sure.
- [112:46]
Keshav Murugesh
And also is more of the kind of place. I'm not saying Jesus Christ would endorse every single thing that happens in the world today. I'm just saying he endorses this timeline more than one in which he doesn't exist and doesn't preach at all.
- [113:00]
Ege Erdo
I don't know, actually. I mean, I'm not sure if that's true. It seems like a hard question, but.
- [113:04]
Keshav Murugesh
I think some from the Christian perspective.
- [113:07]
Ege Erdo
Favorable cultural values, counterfactual.
- [113:10]
Keshav Murugesh
I agree that is always true. I just think the world does have people who read the Bible and are like, I'm inspired by these ideals to do certain things. And it just seems like that's more likely to lead to.
- [113:21]
Ege Erdo
That is what I would call a legacy effect or something. I mean, you can say the same thing about languages. Like, some cultures might just become more prominent and their languages might be spoken more or some symbols might become more prominent. But then there are things like, like how do cities look and how do cars look and what do people spend most of their time doing in their day and what do they spend their money on? And those questions seem much more determined by how your values change as circumstances change.
- [113:49]
Keshav Murugesh
So that may be true, but I'm in the position with regards to the future where I'm like, I expect a lot of things to Be different. And I'm okay with them being different. I care much more about the equivalent of slavery, which in this case is literally slavery. Just to put a finer point on it. The thing I really care about is there's going to be trillions of digital beings. I want it to be the case that they're not tortured and put into conditions in which they don't want to work and whatever, or I don't want galaxies worth of suffering.
- [114:21]
Tameh Besaroglu
Okay.
- [114:21]
Keshav Murugesh
But that seems closer to British abolitionists being like, let's put our empire's might against fighting slavery.
- [114:25]
Ege Erdo
I agree, but I would distinguish between, between the case of Christianity and the case of end of slavery, because I think the end of slavery. I agree. You can imagine a society technologically, it's feasible to have slavery, but I think that's not the relevant thing which brought it to an end. The relevant thing is that the change in values associated with the Industrial revolution made it so that slavery just became an inefficient thing to sustain in a bunch of ways. And a lot of countries at different times phased out different things you could call slavery. So, for example, Russia abolished serfdom in the 1860s. They were not under British pressure to do so. Britain couldn't force Russia to do that. They just did that on their own. There were various ways in which people in Europe were tied to their land and they couldn't move, they couldn't go somewhere else. Those movement restrictions were lifted because they were inefficient. There were ways in which it used to be like the kind of labor that needed to be done in the colonies to grow sugar or to grow various crops. It was very hard labor. It was not the kind of thing that probably you could have paid people to do because they just wouldn't want to do it because the health hazards and so on were very great, which is why they needed people to. To force people to do them. And that kind of work over time became less prevalent in the economy. So again, that reduces the economic incentive to do it. I agree. You could still do it.
- [115:59]
Keshav Murugesh
But I would emphasize it's like the way you're painting the counterfactual is like, oh, but then in that world, they would have just phased out the remnants of slavery. But there's a lot of historical examples where there's not necessarily hard, only hard labor. Like Roman slavery.
- [116:14]
Ege Erdo
Yes, everyone's different.
- [116:16]
Keshav Murugesh
And I interviewed a historian about it recently. The episode hasn't come out, but he wrote a book about, like, the Scope. I think it was like 20% of Roman people under Roman control were slaves. And this was not just agricultural slavery. This was like every like. And his point was that it was this division of the maturity of the Roman economy is what led to this level of slavery. Because the reason slavery collapsed in Europe after the fall of the Roman Empire was because the economy just lost a lot of complexity.
- [116:51]
Ege Erdo
I'm not sure if I would say that slavery collapsed. I mean, I think this just depends on what you mean by slavery. I mean, you know, a lot of ways people who in feudal Europe were like they.
- [116:59]
Keshav Murugesh
But his point is that actually serfdom was not the descendant institution from Romans.
- [117:03]
Ege Erdo
No, I agree, it was not descendant, but like, in fact this is sort of the point I'm trying to make is that values that exist at a given time, like what the values we will have in 300 years or from the perspective of someone 1,000 years ago, what values people are going to have in a thousand years. Those questions are much more determined by the technological and economic and social environment that's going to be there in a thousand years. Which values are going to be functional? Which societies, which values end up being more competitive and being more influential so that other people adopt their values. And it depends much less on the individual actions taken by people a thousand years ago. So I would say that the abolitionist thing, it's not the cause of why slavery came to an end. And slavery comes to an end also because people's own people just have natural preferences that I think are suppressed in various ways during the agricultural era where it's more efficient to have settled societies in cities which are fairly authoritarian and don't allow for that much freedom. And you're in this Malthusian world where people have very low wages perhaps compared to what they enjoyed in the hunter gatherer era. So it's just a different economic period. And I think people were not, they didn't evolve to have the values that would be functional in that era. So what happened is that there has to be a lot of cultural assimilation where people had to adopt different values. And in the industrial revolution people become also very wealthy compared to what they used to be. And that I think leads to different aspects of people values being expressed. Like people just put a huge amount of value on equality. This has always been the case. But I think when it is sufficiently functional for that to be suppressed, they are capable of suppressing it.
- [119:01]
Keshav Murugesh
I mean, if that's a story, then this puts all the more reason, this makes alignment all the more important or the value alignment all the more important because then you're like, oh, if the AIs become wealthy enough, they actually will make a concerted effort to make sure the future looks more like the utility function we put into them, Which I think you have been underemphasizing.
- [119:18]
Ege Erdo
No, I'm not underemphasizing that. I, I think what I would say is there are certain things that are path dependent in history, such that if someone had done something different, something had gone differently a thousand years ago, then today in some respects would look different. I think, for example, which languages are spoken across which boundaries or which religions people have or those kinds or fashion maybe to some extent, though not entirely those things are more path dependent. But then there are things that are not as path dependent. So for example, if some empire, if the Mongols had been more successful and they somehow, I don't know how realistic it is, but they became very authoritarian and had slavery everywhere, would that have actually led to slavery being a much more enduring institution a thousand years later? That seems not true to me. The forces that led to the end of slavery seemed like they were not contingent forces. They seem like deeper forces than that. And if you're saying, well, if we aligned AIs today to some bad set of values, then that could affect the future in some ways which are more fragile. That seems plausible, but I'm not sure how much of the things you care about the future and the ways in which you expect the future to get worse, you actually have a lot of leverage on at the present moment.
- [120:41]
Keshav Murugesh
I mean, another example here might be factory farming where you could say like, oh, us having, it's not like us having better values over time led to suffering going down. In fact, our suffering might have gone up because the, a lot of people say that the incentives that led to factory farming emerging are just like.
- [120:56]
Ege Erdo
And probably when factory farming comes to an end, it will be because the incentives start going that way.
- [121:02]
Keshav Murugesh
But suppose I care about making sure the digital equivalent of factory farming doesn't happen where it may be more efficient, maybe in the all else being equal, it's just more economically efficient to have suffering minds doing labor for you than non suffering minds because of the, because of the intermediary benefits of suffering or something like that, Right. What would you say to somebody like me where I'm like, I really want that not to happen. I don't want the like home filled with suffering workers or whatever. Is it just like, well, give up because this is the way economic history is.
- [121:41]
Ege Erdo
No, I don't think you should give up. It's more like it's hard to anticipate the consequences of your actions in the very distant future. So I would just recommend that you should just discount the future. Not for a moral reason, not because the future is worth less or something, but because it's just very hard to anticipate the effects of your actions in the near term. I think there are things you can do that seem like they would be beneficial. Like, for example, you could try to align your present AI systems to value the things that you're talking about. They should value happiness and they should dislike suffering or something. You might want to support political solutions that would basically, you might want to build up the capacity so that in the future, if you notice something like this happening, then we might have some ability to intervene. Maybe you would think about the prospect of, well, eventually we're going to maybe colonize other stars and civilization might become very large and communication delays might be very long between different places. And in that case, competitive pressures between different local cultures might become much stronger because it's harder to centrally coordinate.
- [122:51]
Keshav Murugesh
That's right.
- [122:51]
Ege Erdo
And so in that world, you might expect competition to take over in a stronger way. And if you think the result of that is going to be a lot of suffering, maybe you would try to stop that. Again, I think at this point it's very far from obvious that trying to say limit competition is actually a good idea. I would probably think it's a bad idea. But maybe in the future we will receive some information and we'll be like, oh, we were wrong, actually, we should stop this. And then maybe you want to have the capacity so that we can make that decision.
- [123:21]
Keshav Murugesh
Right.
- [123:21]
Ege Erdo
But that's a very hard. Like, that's a nebulous thing. How do you build that up? Well, I don't know. I mean, you would need to. That's the kind of thing I would be trying to do.
- [123:28]
Tameh Besaroglu
Yeah. I think the overall takeaway I take from the way that I think about it, and I guess we think about it, is be more humble in what you think you can achieve and just focus on the nearer term, not because it's more morally important than the longer term, but just because it's much easier to have a predictably positive impact on that.
- [123:50]
Keshav Murugesh
One thing I've noticed over the last few weeks of thinking about these bigger picture topics and interviewing Daniel and Scott and then you two, is how often I've changed my mind about everything from the smallest questions about when AI will arrive. It's funny that that's the small question in the grand scheme of things. To whether there will be an intelligence explosion or whether there'll be an R and D explosion, to whether there'll be explosive growth or how to think about that. And if you're in a position where you are just like incredibly epistemically uncertain about what's going to happen, I think it's important to just directly acknowledge this is instead of just, instead of becoming super certain about your next conclusion. Just being like, well, let me just, at least from my perspective, I'm just like, let me just take a step back. I'm not sure what's going on here. And I think a lot more people should be from that perspective. Unless you've had the same opinion about AI for many years, in which case I have other questions for you about why that's the case. And in other situations, I mean generally how we as a society deal with topics on which we are this uncertain is just to have freedom, decentralization, both decentralized knowledge and centralized decision making take the reins and not to do super high volatility centralized moves like, hey, let's nationalize so we can make sure that we can make sure the software only singularity is aligned, or not to make moves that are just incredibly contingent on one worldview that are brittle under other considerations. And that's become a much more salient part of my worldview. I think just classical liberalism is the way we deal with being this epistemically uncertain. And I think we should be more uncertain than we've ever been in history, as opposed to many other people who seem to be more certain than they are about other sort of more mundane topics.
- [125:44]
Tameh Besaroglu
Yeah, I think it's very hard to predict what happens because of this acceleration basically means that you find it much harder to predict what the world might be in 10 years time. I think these questions are also just very difficult and we don't have very strong empirical evidence. And then there's a lot of this kind of disagreement that exists.
- [126:11]
Ege Erdo
I would say that it's important to, in a lot of cases, in a lot of situations, it's much more important to maintain flexibility and ability to adapt to new circumstances, new information than it is to get a specific plan that's going to be correct and that's very detailed and has a lot of specific policy recommendations and things that you should do. So that's actually also the thing that I would recommend if I want to make the transition to AI in this period of explosive growth go better. I would just prefer it if we in general had higher quality institutions. But I am much less bullish on someone sitting down today and working out, okay, what will this intelligence explosion or explosive growth be? What should we do? Like, I think plans that you work out today are not going to be that useful when the events are actually occurring, because you're going to learn so much stuff that you're going to update on so many questions that these plans are just going to become obsolete.
- [127:12]
Tameh Besaroglu
That's right, because one thing you could do is you could look at, say, the history of war planning and how successful war planning has been for actually anticipating what actually happens when the war actually happens.
- [127:22]
Ege Erdo
So for one example, I think I might have mentioned this in, like, off the record at some point. But before the Second World War happened, people were obviously, people saw that there were all these new technologies like tanks and airplanes and so on, which were now like they existed in World War I, but in a much more primitive setting. So they were wondering, what is going to be the impact of these technologies now that we have them in much greater scale. And the British government had estimates of how many casualties there will be from aerial bombardment in the first few weeks of the Second World War. And they expected hundreds of thousands of casualties basically in like two weeks, three weeks after the war begins. So the idea was that air bombing is basically this unstoppable force. All the major urban centers are going to get bombed. Tons of people will die. So basically we can't have a war because if there's a war, then it will be a disaster because we will have this aerial bombardment. But later it turned out that that was totally wrong. In fact, in all of Britain, there were fewer casualties from air bombing in the entire sort of six years of the Second World War than the British government expected in the first few weeks of the war. They had less casualties in six years than they expected in three weeks. So why did they get it wrong? Well, I mean, there are lots of boring, practical reasons. For example, it turned out to be really infeasible to bomb, especially early on to bomb cities in daytime, because your aircraft would just get shot down. But then if you tried to bomb at nighttime, then your bombing was really imprecise and only a very small fraction of it actually hit. And then people also underestimated the extent to which people on the ground could, like, firefighters and so on, could just sort of go around the city and put out fires from bombs that were falling on structures. They overestimated the amount of economic damage that it would do. They underestimated how economically costly it would be. Basically, you're sending These aircraft and then they're getting shot down. Well, an aircraft is very expensive. So in the end what turned out is when the Allies started bombing Germany, they were like for each dollar of capital they were destroying in Germany, they were spending like $4 to $5 on the aircraft and fuel and training of the pilots and so on that they were sending in missions. And the casualty rate was very high, which later got covered up by the government because they didn't want people to worry about. So that is a kind of situation where all the planning that you would have done in advance, predicated on this assumption of air bombing is going to be this nuclear weapons light. Basically. It's extremely destructive. There's going to be some aspect to.
- [129:57]
Keshav Murugesh
Which, I mean, it was though, right, 84,000 people died in one night of firebombing in Tokyo. Like Germany, like large fractions of their.
- [130:07]
Ege Erdo
But that was over the period of.
- [130:09]
Keshav Murugesh
Six years there were like single firebombing attacks. I mean, it was a case that during the end of World War II, when they were looking for the place to launch the atomic bombs.
- [130:20]
Ege Erdo
That's right.
- [130:21]
Keshav Murugesh
They just had to go through like a dozen cities because they're like, it just wouldn't be worth nuking them because they're already destroyed by the firebombing.
- [130:28]
Ege Erdo
That's right. But if you look at the level of destruction that was expected within the space of a few weeks and then this level of destruction took many years. So there was like a two order of magnitude mismatch or something like that, which is pretty huge.
- [130:43]
Keshav Murugesh
Yeah.
- [130:43]
Ege Erdo
So that affected the way people think about it.
- [130:45]
Keshav Murugesh
Right.
- [130:46]
Tameh Besaroglu
An important underlying theme of much of what we have discussed is how powerful just reasoning about things is to making progress about what specific plans you want to make to prepare and make this transition to advanced AI go well. And our view is, well, it's actually quite hard and you need to make contact with the actual world in order to inform most of your beliefs about what actually happens. And so it's somewhat futile to think to do a lot of war gaming and figure out how AI might go and what we can do today to make that go a lot better. Because a lot of the policies you might come up with might just look fairly silly. And I think there's in the thinking about how AI actually has this impact. Again, people think, oh, you know, just AI reasoning about doing science and doing R and D just has this drastic impact on the overall economy or technology. And our view as well, actually, again, making contact with the real world and getting a lot of data from experiments and from deployment and so on. It's just very important. So I think there is this underlying kind of latent variable which explains some of this disagreement both on the policy prescriptions and about the extent to which we should be humble versus ambitious about what we ought to do today, as well as for thinking about the mechanism through which AI has this impact and this underlying latent thing. It's like, what is the power of reason? How much can we reason about what might happen? How much can reasoning in general figure things out about the world and about technology? So that is a kind of core underlying disagreement here.
- [132:28]
Keshav Murugesh
Yeah. I do want to ask you say in your announcement, we want to accelerate this broad automation of labor as fast as possible. As you know, many people think it's a bad idea to accelerate this, the broad automation of labor and AGI and everything that's involved there. Why do you think this is good?
- [132:49]
Ege Erdo
So the argument for why it's good is that we're going to have this enormous increase in economic growth, which is going to mean enormous amounts of wealth and incredible new products that you can't even imagine in healthcare or whatever. And the quality of life of the typical person is probably going to go up a lot early on. Probably also their wages are going to go up because the AI systems are going to be automating things that are complementary to their work, or it's going to be automating part of their work and then you'll be doing the rest and then you'll be getting paid much more on that. And in the long term, eventually we do expect wages to fall just because of arbitrage with the AIs. But by that point, we think humans will own enormous amounts of capital. And there will also be ways in which even the people who don't own capital, we think, are just going to be much better off than they are today. I think it's just hard to express in words the amount of wealth and increased variety of products that we would get in this world. It would be probably more than a difference between 1800 and today. So if you imagine that difference, that's such a huge difference. And then imagine two times, three times.
- [133:57]
Keshav Murugesh
Whatever the standard argument against this is, why does the speed to get there matter so much? Because especially if the trade off against the speed is the probability that this transition is achieved successfully in a way that benefits humans, I mean, it's unclear.
- [134:13]
Tameh Besaroglu
That this trades off against the probability of it being achieved successfully or something.
- [134:18]
Keshav Murugesh
There might be an alignment tax.
- [134:20]
Tameh Besaroglu
I mean, maybe you can also just do the calculation of how much a year's worth of delay costs for current people. This is this enormous amount of utility that people are able to enjoy and that gets brought forward by year or pushed back by year if you delay things by year. And how much is this worth? Well, you can look at simple models of how concave people's utility functions are and do some calculations and maybe that's worth on the order of tens of trillions of dollars per year in consumption. That is roughly the amount consumers might be willing to defer in order to bring forward the date of automation one year.
- [135:04]
Keshav Murugesh
In absolute terms, it's high. In relative terms relative to if you did think it was going to nudge the probability one way or another of building systems that are aligned and so forth, then it's just so small compared to all of the future.
- [135:18]
Ege Erdo
I agree. So there are a couple of things here. First of all, I think the way you think about this matter. So first of all, we don't actually think that it's clear whether speeding things up versus slowing things down actually makes a doomy outcome more or less likely. I think that's just a question that doesn't seem obvious to us. Partly because of our views on the software R and D side. We don't really believe that if you just pause and then you DO research for 20 years at a fixed level of compute scale that you're actually going to make that much progress on relevant questions on alignment or something. I think imagine you were trying to make progress in alignment in 2016 with the compute budgets of 2016. Well, you would have gotten nowhere basically you would have discovered none of the things that people have today discovered and that turned out to be useful. And I think if you pause today, then we will be in a very similar position in 10 years. Right. Like we have not made a bunch of discoveries. So the scaling is just really important to make progress and alignment in our view. And then there's a separate question of how long term should you be in a various different senses. So there's a moral sense or like how much should you actually care about people who are alive today as opposed to people who are not yet born? That's just a moral question. And there's also a practical question of as we discuss how certain can you be about the impact your present actions are actually going to have on the future?
- [136:43]
Keshav Murugesh
Okay, maybe you think it really doesn't matter whether you slow things down right now or you speed things up right now. But is there some story about why speeding Them up from the alignment perspective actually help. It's good to have that extra progress right now rather than later on. Or is it just that, well, if it doesn't make a difference either way, then it's better to just get that extra year of people not dying and having cancer cures and so forth.
- [137:05]
Ege Erdo
I think I would say the second. But like, but it's just important to understand the value of that. Even in purely economic terms, imagine that you would be. Each year of delay might cause maybe 100 million people, maybe more, maybe 150, 200 million people who are alive today to end up dying. So even in purely economic terms, the value of statistical life is pretty enormous, especially in Western countries. So it's like sometimes people use numbers as high as $10 million for a single life. So imagine you do like $10 million times 100 million people. That's like a huge number, right? So I think that is just so enormous that unless you're just so. I think for you to think that speeding things up is a bad idea, you have to first be just have this long termist view where you look at the long run future. You think your actions today have high enough leverage that you can predictably affect the direction of long run future.
- [138:10]
Keshav Murugesh
In this case it's kind of different because you're not saying I'm going to affect what some emperor a thousand years from now does, like somebody in the year zero would have to do to be a long termist. In this case you just think there's this incredibly important inflection point that's coming up and you just need to have influence over that crucial period of explosive growth, of intelligence explosion or something. So I think it is a much more practicable prospect than.
- [138:36]
Ege Erdo
So I agree in relative terms. So in relative terms I agree the present moment is a moment of higher leverage and you can expect to have more influence. I just think in absolute terms the amount of influence you can have is still quite low. So it might be orders of magnitude greater than it would have been 2,000 years ago and still be quite low.
- [138:54]
Tameh Besaroglu
And again, I think there's this difference in opinion about how broad and diffuse this transformation ends up being versus how concentrated within a specific labs where the very idiosyncratic decisions made by that lab will end up having very large impact. If you think those developments will be very concentrated, then you think the leverage is especially great. And so then you might be especially excited about having the ability to influence how that transition goes. But our view is very much that this transition happens very Diffusely, by way of many, many organizations and companies doing things. And for those actions to be determined a bunch by economic forces, rather than idiosyncratic preferences on the part of labs or these kind of decisions that have these kind of founder effects that last for very long.
- [139:48]
Keshav Murugesh
Okay, let's go through some of the objections to explosive growth, which is most people are actually more conservative, not more aggressive about the forecasts you have. So obviously one of the people who has articulated their disagreements with your view is Tyler Cowen. He made an interesting point when we did the podcast together and he said most of sub Saharan Africa still does not have reliable clean water. The intelligence required for that is not scarce. We cannot so readily do it. We are more in that position than we might like to think along other variables.
- [140:22]
Tameh Besaroglu
I mean, we agree with this. I think intelligence isn't the bottleneck that's holding back technological progress or economic growth. It's like many other things. And so I think that this is very much consistent with our view that scaling up your overall economy, accumulating capital, accumulating human capital, having all these factors, scales.
- [140:45]
Ege Erdo
In fact, this is even consistent with what I was saying earlier that I was pointing out this good management and good policies and those just contribute to TFP and they can be bottlenecks. And.
- [140:56]
Keshav Murugesh
But right now we could just plug and play our better management into sub Saharan Africa.
- [141:02]
Ege Erdo
It's hard, I don't think.
- [141:04]
Keshav Murugesh
Okay, so that's what maybe I should have said. One could theoretically imagine plugging and playing with. I can imagine many things, but we cannot so readily do it because it's hard to articulate why and it wouldn't be so easy to do in just capital or labor. Why not think that the rest of the world will be in this position with regards to the advances that AI will make possible?
- [141:32]
Tameh Besaroglu
I mean, if the AI advances are like the kind of geniuses in a data center, then I agree that that might be bottlenecked by the rest of the economy not scaling up and being able to accumulate the relevant capital to make those changes feasible. So I kind of agree with this picture and I think this is like in objection to the geniuses in a data center type view. And I buy basically this and also.
- [141:58]
Ege Erdo
The fact that it's also plausible you're going to have the technology and. But then some people are not going to want to deploy it, or some people are going to have norms and laws and cultural things that are going to make it so that AI is not able to be widely deployed in their economy or not as Widely deployed as otherwise it might be. And that is going to make those countries or societies just slower. Some countries will be growing faster, just like Britain and the Netherlands were sort of the leaders in the industrial revolution. They were the first countries to start experiencing rapid growth. And then other countries, even in Europe, sort of had to come from behind. Well, again, I just think we expect the same thing to be true for AI. And I mean, the reason that happened was exactly because of these kinds of reasons, where those countries had a culture or governance systems or whatever, which were just worse and bottlenecked the deployment and scaling of the new technologies and ideas. It seems very plausible.
- [142:52]
Keshav Murugesh
But you're saying as long as there's one jurisdiction. Yeah, but then again, you also previously emphasized the need and the need to integrate with the rest of the global economy and the human economy. So doesn't that contradict.
- [143:06]
Tameh Besaroglu
Doesn't often require cultural homogeneity. We trade with countries. The US Trades with China quite a lot, actually. And there's like a bunch of disagreements.
- [143:15]
Keshav Murugesh
What if the US is like, I don't like, the UAE is doing an explosive growth with AI, we're just going to embargo them?
- [143:22]
Tameh Besaroglu
That seems plausible.
- [143:23]
Keshav Murugesh
And then would that not prevent explosive growth?
- [143:26]
Tameh Besaroglu
I mean, I think that would be plausible at the point at which it's revealing a lot about the capabilities and the power of AI. And you should also think that that creates both an incentive to embargo, but also an incentive to adopt the very similar styles of governing that enable AI to be able to produce a lot of value.
- [143:48]
Keshav Murugesh
What do you make of this? I think people interpret explosive growth from an arms race perspective, and that's often why they think in terms of public private partnerships for the labs themselves. But just this idea that you have the geniuses in the data center, they like, you can have them come up with the mosquito drone swarms, and then those drone swarms will. If China gets to the swarms earlier. I mean, even within your perspective where it's like, not, is this a result of your whole economy being advanced enough that you can produce mosquito drone swarms, you being six months ahead means that you could decisively win. Does it? I don't know. Maybe you being like, a year ahead in explosive growth means you could decisively win a war against China or China could win a war against you. So would that lead to an arms race? Slight dynamic?
- [144:33]
Ege Erdo
I mean, I think it would to some extent, but I'm not sure if I would expect that, like, a year of lead to be enough to take a risk. Because if you go to war with China, I mean, for example, if the US went to war with, if you replace China today with China from 1990, or if you replace Russia Today with Russia from 1970 or 1980, it's possible that their ICBM and whatever technology is already enough. It's already enough to have very strong deterrence. So maybe even that lead, a technological lead, is not sufficient so that you would feel comfortable going to war. So that seems possible.
- [145:13]
Keshav Murugesh
Yeah. And actually this relates to a point that Guern was making, which is he was like, okay, this is going to be a much more unstable period than the Industrial Revolution, even though Industrial Revolution saw many countries gain rapid increases in their capabilities. Because this is just like within this span, if you have a century's worth of progress compressed within a decade, one country gets to ballistic missiles first, then the other country gets to railroads first and so forth. But if you have this more integrated perspective about what it takes to get to ballistic missiles and to railroads, then you might think, no, basically this isn't some orthogonal vector. It just like you're just like churning on the tech tree further and further.
- [146:02]
Ege Erdo
Yeah, I mean, for what it's worth, I do think it's possible if you have it just happen in a few countries which are relatively large and have enough land or something, those countries could just, they would be starting from a lower base compared to the rest of the world, so they would need to catch up to some extent. So if they are just going to sort of grow internally and they're not going to depend on the external supply chains, but that doesn't seem like something that's impossible to me, that some countries could do it, it would just be more difficult. But in this setting, if some countries have a significant policy advantage over the rest of the world and they start growing first, and then they won't necessarily have a way to get other countries to adopt their norms and culture. So in that case, it might be more efficient for them to do the growth locally. Right. So that's why I was saying the growth differentials will probably be determined by regulatory jurisdiction boundaries more than anything else. I'm not saying, say the US by itself, if it had AI, but it couldn't get the rest of the world to adopt AI, I think that would still be sufficient for explosive growth.
- [147:04]
Keshav Murugesh
How worried should we be about the fact that China today just has, because it industrialized relatively recently, just has more industrial capacity and know how and all the other things of learning by doing and so forth? If we buy your model of how Technology progresses with or without AI. How like are we just underestimating China? Because we have this perspective that like what fraction of your GDP you're spending on research is what matters, when in fact it's the kind of thing where I've got all the factories in my backyard and I know how they work and I can go buy a component whenever I want.
- [147:41]
Tameh Besaroglu
I don't think people are necessarily underestimating China. I mean, it depends on who you're looking at. But it seems like the discussion of China is this very big discussion in these AI circles. And so people are very much appreciating the power and the potential threat that China poses. But I think the key thing is not just like the scale in terms of pure number of people or number of firms or something, but the scale of the overall economy which is just measured in how much is being produced in terms of dollars. The US is ahead, but doesn't the.
- [148:17]
Keshav Murugesh
We're not expecting all this explosive growth to come from financial services. We're expecting it to start from a base of industrial technology and industrial capacity.
- [148:25]
Ege Erdo
No, financial services can be important if you want to scale big projects very quickly.
- [148:30]
Tameh Besaroglu
Financial services are very important for raising funding and getting investments in data centers.
- [148:36]
Keshav Murugesh
If I understood you correctly, it just seems like man, you know how to do all the like, you know how to build the robot factories and so forth, that like know how, which in your view is still crucial to technology growth and just general economic growth is lacking and you might have more advanced financial services. But like it seems like the more you take your view seriously, the more it seems like the having the Shenzhen.
- [148:59]
Tameh Besaroglu
Locally matters a lot relative to what starting point. I think people already appreciate that China is very important. And then I agree that there are some domains where China is leading, but then there are very many domains in which the US is leading or the US and its allies where countries that are producing relevant inputs for AI that the US has access to but China doesn't. So I think the US is just ahead on many dimensions and there's some that China is ahead or at least very close. So I don't think this should cause you to update very strongly in favor of China being a much bigger deal, at least depending on where you stand.
- [149:40]
Ege Erdo
And I think people already think China is a big deal. This is the big underlying thing here. If people were just very dismissive of China, then maybe this would be a reason to update.
- [149:47]
Keshav Murugesh
But I get your argument that thinking about the economy wide acceleration is more important than focusing on the IQ of the smartest AI. But at the same time, do you believe in the idea of superhuman intelligence? Is that a coherent concept in the way that you don't necessarily stop at human level go playing, you just go way beyond it. In ELO Score, where we get to systems that are like that with respect to the broader range of human abilities. And maybe that doesn't mean they become God because there's other ASIS in the world. But you know what I mean. Will there be systems with such superhuman capabilities?
- [150:27]
Tameh Besaroglu
Yeah, I mean, I do expect that. I think there's a question of how useful is this concept for thinking about this transition to a world with much more advanced AI? And I don't find this a particularly meaningful or helpful concept. I think people introduce some of these notions that on the surface seem useful, but then actually when you delve into them, it's very vague and kind of unclear what you're supposed to make of this. And you have this notion of AGI which distinguishes from narrow AI in the sense that it's much more general and maybe can do everything that a human can do on average. I mean, AI systems have these very jagged profiles of capabilities. So you have to somehow take some notion of average capabilities. And what exactly does that mean? It just feels really unclear. And then you have this notion of asi, which is AGI in the sense that it's very general, but then it's also better at humans on every task. And is this a meaningful concept? I guess it's coherent. I think this is not a super useful concept because I prefer just thinking about what actually happens in the world. And you could have a drastic acceleration without having an AI system that can do everything better than humans can do. I guess you could have no acceleration when you have an ASI that is better than humans at everything, but it's just very expensive or very slow or something. So I don't find that particularly meaningful or useful. I just prefer thinking about the overall effects on the world and what AI systems are capable of producing those types of effects.
- [152:07]
Keshav Murugesh
Yeah, I mean, one intuition pump here is compare John von Neumann versus a human flick from the standard distribution. If you added a million John von Neumanns to the world, what would the impact on growth be as compared to just adding a million people from normal distribution?
- [152:25]
Ege Erdo
Well, I agree it would be much greater.
- [152:27]
Keshav Murugesh
Right? But then because of Moravak's paradox type arguments that you made earlier, that evolution has not necessarily optimized us for that long along the kind of spectrum on which John von Neumann is distinguished from the average human and given the fact that already within this deviation, you have this much greater economic impact, why not focus on optimizing on this thing that evolution does not optimize that hard on further?
- [152:52]
Ege Erdo
I don't think we shouldn't focus on that. But what I would say is, for example, if you're thinking about the capabilities of go playing AIs, then the concept of a superhuman go AI, yeah, you can say that is a meaningful concept. But if you're developing the AI, it's not a very useful concept. If you just look at the scaling curves, it's just like it just goes up and there is some human level somewhere. But the human level is not privileged in any sense. So the question is, is it a useful thing to be thinking about? The answer is probably not. Depends on what you care about. So I'm not saying we shouldn't focus on trying to make the system smarter than humans are. I think that's a good thing to focus on.
- [153:32]
Keshav Murugesh
Yeah, I guess I'm trying to understand whether we will stand in relation to the AIs of 2100 that humans stand in relationship to other primates. Is that the right mental model we should have? Or is it going to be a much greater familiarity with their cognitive horizons?
- [153:48]
Tameh Besaroglu
I mean, I think AI systems will be very diverse. And so it's not super meaningful to ask something about, you know, this very diverse range of systems and where we stand in relation to them. I mean, but can we'll be able.
- [154:00]
Keshav Murugesh
To, like, cognitively access the kinds of considerations they can take on board? Like, humans are diverse, but no chimp is going to be able to understand this argument in the way that another human might be able to. Right. So I'm just like, if I'm trying to think about my place or a human's place in the world of the future, I think it is a relevant concept of. Is it just that the economy has grown a lot and there's much more labor, or are there beings who are in this crucial way super intelligent?
- [154:28]
Tameh Besaroglu
I mean, there will be many things that we just will fail to understand. And to some extent, there are many things today that people don't understand about how the world works and how certain things are made. And then how important is it for us to have access or in principle be able to access those considerations? And I think it's not clear to me that that's particularly important that any individual human should be able to access all the relevant considerations that produce some outcome that just seems like overkill. Why do you need that to happen? I think it would be nice in some sense. But I think if you want to have a very sophisticated world where you have very advanced technology, those things will just not be accessible to you. So you have this trade off between accessibility and maybe how advanced the world is. And from my point of view, I'd much rather live in a world which has very advanced technology, has a lot of products that I'm able to enjoy and a lot of inventions that I can improve my life with. If that means that I just don't understand them, I think this is a very simple trade that I'm very willing to make.
- [155:45]
Keshav Murugesh
Okay, so let's get back to objections to explosive growth. We discussed a couple already. Here's another which is more a question than an objection. Where is all this extra output going? Who is consuming it? If the economy is 100x bigger in a matter of a decade or something, to what end?
- [156:05]
Ege Erdo
So first of all, I think even if you view that along what you might call the intensive margin, in the sense that you just have more of the products you have today, I think there is just a lot of. There will be a lot of appetite for that. Maybe not quite 100x that might start hitting some diminishing returns.
- [156:23]
Tameh Besaroglu
Current GDP per capita on average in the world is 10k a year or something thing. Right. And there are people who enjoy millions of dollars. And so there's a gap between what people enjoy and like don't seem to be super diminished in terms of marginal utility. And so there's a big room, there's a lot of room on just purely the intensive margin of just consuming the things we consume today. But more. And then there's this maybe much more important dimension along which we will expand, which is productivity. Yeah, Extensive margin of what is the scope of things that you're consuming. And if you look at something like the Industrial revolution, that seemed to have been the main dimension along which we kind of expanded to consume more. There's just on any kind of sector that you care about, transportation, medicine, you know, entertainment and food, there's just this massive expansion in terms of variety of things that we're able to consumed that is enabled by new technology or new trade routes or new methods of producing things. And so that is I think really the key thing that we will see come along with this kind of expansion in consumption.
- [157:36]
Keshav Murugesh
Another point that Tyler makes is that there will be some mixture of Bamoul Kos disease where you're bottlenecked by the slowest growing thing which grows in proportion the fastest. Productivity things Basically diminish their own share in outputs. Yeah, that's right.
- [157:58]
Tameh Besaroglu
Yeah, I mean, we totally agree with that. I would say that that's just like a kind of qualitative consideration. It doesn't itself. It isn't itself sufficient to make a prediction about what growth rates are permitted given these BOMO effects versus not. It's just like a qualitative consideration. And then you might need to make additional assumptions to be able to make a quantitative prediction. So I think it's a little bit.
- [158:25]
Ege Erdo
So the convincing version of this argument would be if you did the same thing that we were doing earlier with the software only singularity argument, where we were pointing to essentially the same rejection, where there are multiple things that can bottleneck progress. So I would be much more convinced if someone pointed to an explicit thing, they would be like here, healthcare is this very important thing and why should we expect AI to make that better? That doesn't seem like that would get better because of AI. So maybe healthcare just becomes a big part of the economy and then that bottleneck. So if there was some specific sector.
- [158:58]
Keshav Murugesh
Maybe the argument is that if there's even one.
- [159:01]
Ege Erdo
No, if there's one though, if that's a small part of the economy, then you could just still get a lot of growth. You just automate everything else. And that is going to produce a lot of growth.
- [159:08]
Tameh Besaroglu
So it has to quantitatively work out. And so you actually have to be quantitatively specific about what this objection is supposed to be.
- [159:16]
Ege Erdo
Right. So first of all, you have to be specific about, okay, what are these tasks? What are the current share in economic output? The second thing is you have to be specific about how bad do you think the complementarities are. So in numerical terms, economists use the concept of elasticity of substitution to quantify this. So that gives you a numerical estimate of if you just have much more output on some dimensions but not that much on other dimensions, how much does that increase economic output overall? And then there's a third question. You can also imagine you automate a bunch of the economy. Well, a lot of humans were working on those jobs. So now, well, they don't need to do that anymore because those got automated. So they could work on the jobs that haven't yet been automated. So for example, as I gave the example earlier, you might imagine a world in which remote work tasks get automated first and then sensory motor skills lag behind. So you might have a world in which software engineers become physical workers. Instead. Of course, in that world, the wages of physical workers will be much higher than their wages are today. So that reallocation also produces a lot of extra growth even in the. If bottlenecks are maximally powerful, even if it's literally, you just look at all the tasks in the economy and literally take the worst one for productivity growth, you would still get a lot of increase in output because of this reallocation.
- [160:36]
Tameh Besaroglu
So I think one point that I think is useful to make our experience talking to economists about this is that they will bring up these kind of more qualitative considerations. Whereas the arguments that we make are like, make specific quantitative predictions about growth rates. So, for example, you might ask, how fast will the economy double? And then we can think about an H100 does about. There are some estimates of how much computation the human brain does per second, and it's about 1e15 flop or so it's a bit unclear. And then it turns out that an H100 roughly does on that order of computation. And so you can ask the question of how long does it take for an H100 to pay itself back?
- [161:21]
Ege Erdo
If you run the software of the human brain.
- [161:23]
Tameh Besaroglu
If you run the software of the human brain, you can then deploy that in the economy and earn, say, human wages on the order of 50 to 100k a year or whatever in the US and so then it pays itself back because it costs on the order of 30k per H100. And so you get a doubling time of maybe on the order of a year. Right. And so this is like a very quantitatively specific prediction about. And then there's the response, well, you have bommel effects. And then you're like, okay, well what does this mean? Does this predict it doubles every two years or every five years? You need just more assumptions in order to make this a coherent objection. And so I think a thing that's a little bit confusing is just that there are these qualitative objections that I agree with, like bottlenecks are indeed important, which is part of the reason I'm more skeptical of this software singularity story. But I think this is not sufficient for blocking explosive growth.
- [162:24]
Keshav Murugesh
The other objection that I've heard often, and it might have a similar response from you, is this idea that a lot of the economy is comprised of O ring type activities. And this refers to, I think then the Challenger, the Challenger space shuttle explosion. There was just like one component. I forgot what the exact problem with the O ring was. But because of that being faulty, the whole thing collapsed.
- [162:49]
Tameh Besaroglu
I mean, I think it's quite funny, actually, because the O ring model is Taking the product of many inputs and then the overall output is the product of very many things. That's right. But actually this is pretty optimistic from the point of view of having fewer.
- [163:07]
Ege Erdo
Bottlenecks because I think we pointed this out before where again talking about software on the singularity, I said like if it's the product of compute for experiments with research.
- [163:14]
Keshav Murugesh
But if one of those products because of human.
- [163:17]
Tameh Besaroglu
But you have constant marginal product there, right?
- [163:20]
Ege Erdo
Yeah. But if one of those products doesn't scale, that doesn't limit. Like yeah, it means you're less efficient at scaling than you otherwise would be. But you can still get a lot unbounded.
- [163:31]
Tameh Besaroglu
You can just have unbounded scaling in the O ring world. So actually I disagree with Tyler that he's not conservative enough that he should take his bottlenecks view more seriously than he actually is. And yet I disagree with him about the conclusion. And I think that we're going to get explosive growth once we have AI that can flexibly substitute.
- [163:50]
Keshav Murugesh
I'm not sure I understand. There will be entirely new organizations that AIs come up with. We've written a blog post about one such with the AI firms. And you might be a productive worker or a productive contributor in this existing. The organizations that exist today in the AI world, many humans might just be like zero or even minus.
- [164:12]
Ege Erdo
I agree.
- [164:14]
Keshav Murugesh
Why won't that put that in the multiplication?
- [164:19]
Tameh Besaroglu
But why would humans be in the loop there?
- [164:22]
Ege Erdo
You're both saying that humans would have. Humans would be like negatively contributing to output. But then you're also saying that we should put them into the.
- [164:32]
Keshav Murugesh
It seems like, okay, fair, fair, fair. The main objection often is regulation. And I think we've addressed it implicitly in different points, but might as well just explicitly address why won't regulation stop this?
- [164:43]
Ege Erdo
Yeah. So for what it's worth, we do have a paper where we go over all the arguments for against explosive growth and regulation, I think is the one that seems like strongest as against. Because the reason it seems strong is because even though we have made arguments before about international competition and variation of policies among jurisdictions and these strong incentives to adopt this technology both for economic and national security reasons. So I think those are pretty compelling when taken together. But even still, the world does have a surprising ability to coordinate on just not pursuing certain technologies and human cloning. That's right. So I think it's hard to be extremely confident that this is not going to happen. I think it's less likely that we're going to do this for AI than it is for human cloning, because I think human cloning touches on some other taboos and so on. Also less valuable and probably less important also for national security in an immediate sense. But at the same time, as I said, it's just hard to rule this out. So I wouldn't say if someone said, well, I think there is a 10% or 15%, whatever, 20% chance that there will be some kind of global coordination of regulation and that's going to just be very effective. Maybe it will be enforced through sanctions on countries that defect, and then that is going to. Maybe it doesn't prevent AI from being deployed, but maybe it just slow things down enough that you never quite get exposed for growth. I don't think that's an unreasonable view. If it's like 10% chance, could be.
- [166:20]
Keshav Murugesh
I don't know if there's any. I don't know. Do you encounter any other objections? What should I be hassling you about?
- [166:26]
Ege Erdo
Yeah, I mean, some things that we've heard from economists, again, there was this argument that people sometimes respond to our argument about explosive growth, which is an argument about growth levels. So we're saying we're going to see 30% growth per year instead of 3%. They respond to that with an objection about levels. So they say, well, how much more efficient, how much more valuable can you make, like hairdressing or taking flights or whatever, or going to a restaurant? And that is just fundamentally the wrong kind of objection. We're talking about the rate of change and you're objecting to it by making an argument about the absolute level of productivity. And as I said before, this is not an argument that economists themselves would endorse if it was made about a slower rate of growth continuing for a longer time. So it seems more like special pleading.
- [167:21]
Keshav Murugesh
I mean, why not just the deployment thing where the same argument you made about AI, where you do learn a lot just by deploying to the world and seeing what people find useful. ChatGPT was an example of this. Why won't a similar thing happen with AI products and services where you're just. One of the components is you put it out to the marketplace and people play with it and you find out what they need and it clings into the existing supply chain and so forth. Doesn't that take time?
- [167:49]
Tameh Besaroglu
Well, I mean, it takes time, but it is often quite fast. In fact, ChatGPT grew extremely fast.
- [167:56]
Keshav Murugesh
Right, but those are just purely digital service.
- [167:58]
Ege Erdo
Well, I think the important thing would be like, yeah, one reason to be optimistic is if you think the AIs will literally be drop in remote workers or drop in workers in some cases. If you have robotics, then companies are already kind of experienced at onboarding humans. Onboarding humans doesn't take a very long time. Maybe it takes six months even in a particularly difficult job for a new worker to start being productive. Well, that's not that long. So I don't think that would rule out companies being able to onboard AI workers, assuming that they don't need to make a ton of new complementary innovations discoveries to take advantage. I think one way in which current AI systems are being inhibited and the reason we're seeing the growth maybe be slower than you might otherwise expect, is because companies in the economy are not used to working with this new technology. They have to rearrange the way they work in order to take advantage of it. But if AI systems were literally able to substitute for human workers, then well, the complementary innovations might not be as necessary.
- [169:01]
Keshav Murugesh
Actually, this is a good excuse to maybe go to the final topic, which is AI firms. So this is this blog post we wrote together about what it would be like to have a firm that is fully automated. And the crucial point we were making was that people tend to overemphasize and think of AI from the perspective of how smart individual copies will be. And, and if you actually want to understand the ways in which they are superhuman, you want to focus on their collective advantages, which because of biology we are just precluded from, which are the fact that they can be copied with all their tacit knowledge. You can copy a Jeff Dean or Elias or whatever the relevant person is in a different domain. You can even copy Elon Musk and he can be the guy who's every single engineer in the SpaceX rig. And if that's not an efficient way.
- [169:49]
Tameh Besaroglu
To the equivalent of them.
- [169:50]
Keshav Murugesh
Yeah, and it's not best to have Elon Musk or anything. You just copy the relevant like team or whatever that. And we have this problem with human firms where there can be very effective teams or groups, but over time their culture dilutes or the people leave or die or get old. And this is one of the many problems that can be solved with these digital firms where you actually firms right now have two of the three relevant criteria for evolution. They have selection and they have variation, but they don't have high fidelity replication. And you could imagine a much more fast paced and intense sequence of evolution for firms. Once you have this final piece, click in and that relates to the onboarding thing where right now they just aren't Smart enough to be onboarded as full workers. But once they are, I just imagine for my own, the kinds of things I try to hire for, it would just be such an unlock. It doesn't even matter. The salaries are totally secondary. The fact that I can, this is the skill I need or the set of skills I need and I can have a worker and just like I can have 1,000 workers in parallel. If there's something that has a high elasticity of demand. I think is probably, along with the transformative AI, the most underrated, tangible thing that you need to understand about what the future AI society will look like.
- [171:22]
Ege Erdo
Right. I mean, I think there's the first point about this very macroeconomic picture where you just expect a ton of scaling of all the relevant inputs. And I think that is like the first order thing. But then you might have more like micro questions about, okay, how does this world actually look like? How is it different from a world in which we just have a lot more people and a lot more capital and a lot more. Because it should be different. And then I think these considerations become important. I think another important thing is just that AIs can be aligned. You get to control the preferences of your AI systems in a way that you don't really get to control the preference of your workers. Like your workers again, just select. You don't really have any other option but for your AIs, you can fine tune them. You can build AI systems which have the kind of preferences that you want. And you can imagine that dramatically changing basic problems that determine the structure of human firms. For example, the principal agent problem might go away. This is a problem where you as a worker have incentives that are either different from those of your manager or those of the entire firm or those of the shareholders of the firm.
- [172:29]
Keshav Murugesh
I actually think the incentives is a smaller piece of the puzzle. I think it's more about bandwidth and information sharing where it's. Often with a large organization it's very hard to have a single coherent vision. And the most successful forums we see today is where for an unusual amount of time a founder is able to keep their vision instilled in the organization. Like SpaceX or Tesla are examples of this. People talk about Nvidia this way, but just imagine a future version where there's this hyper inference scaled Mega Jensen who you're spending $100 billion a year on inference on, and copies of him are constantly, you know, like writing every single press release and reviewing every pull request and answering every customer service request and so forth and monitoring the whole organization making sure it's like proceeding along a coherent vision and getting merged back into the hypergensen. Hypergensen, megagensen, whatever.
- [173:31]
Ege Erdo
Yeah, I agree, that's a bigger deal. At the same time, I would point out that part of the reason why it's important to have a coherent vision and culture and so on in human companies might be that there's incentive problems exist. Otherwise I wouldn't rule that out. But I agree that the like, aside from the overall macroeconomic thing, I think the fact that they can be replicated is probably the biggest deal. That also enables additional sources of economies of scale where if you have twice the number of GPUs, you can run not only twice the number of copies of your old model, but then you can train a model that's even better. So you double your training compute and your inference compute. And that means you not only double, like you don't get just twice the number of workers you would have had otherwise, you get more than that because they are also smarter because you spend more training computers. So then that is additional source of economies of scale. And then there's this benefit that you can for humans. Every human has to learn things from scratch. Basically they are born and then they have a certain amount of lifetime learning that they have to do. So in human learning there is a ton of duplication. While for an AI system it could just learn once, it could just have one huge training run which are tons of data and then that front could be deployed everywhere. So that's like another massive advantage that the AIs have over humans.
- [174:43]
Keshav Murugesh
Yeah, maybe we'll close up with this one debate we've often had offline, which is will central planning work with these economies of scale?
- [174:53]
Ege Erdo
So I would say that again the question of will it work, will it be optimal? Right. My guess is probably not optimal, but I think it's hard. I don't think anyone has thought this question through in a lot of.
- [175:10]
Tameh Besaroglu
So it's worth thinking about just why one might expect central planning to be slightly better in this world. So one consideration is just communication bandwidth being potentially much, much greater than it is today. And then in the current world, the information gathering and the information processing are co located. Humans observe and also process what they observe. In an AI world, you can disaggregate that.
- [175:43]
Keshav Murugesh
That's actually a really interesting point.
- [175:44]
Tameh Besaroglu
Yeah, so you can have the sensors and not do much processing, but just collect and then process centrally. And that processing centrally might make sense for a bunch of reasons. And you might get economies of scale from having more GPUs that produce better models and also be able to think more deeply about what it's seeing.
- [176:06]
Keshav Murugesh
It's worth noting that certain things already work like this, for example, Tesla fsd, it will benefit from the data collected at the periphery from millions of miles of driving and then the improvements which are made as a result of this centrally directed.
- [176:19]
Tameh Besaroglu
It's coming from this HQ being like we're going to push an update.
- [176:24]
Keshav Murugesh
That's right.
- [176:24]
Tameh Besaroglu
And so you do get some of this more centralized platform and it can.
- [176:28]
Keshav Murugesh
Be a much more intelligent form than just whatever gradient averaging that they. I mean sure, I'm sure it's more sophisticated than Tesla, but it can be a much more deliberate intelligent update.
- [176:37]
Tameh Besaroglu
So that's one reason to expect. And the other reason I guess is just having current leaders or CEOs don't have bigger brains than the workers do. Maybe a little bit.
- [176:50]
Keshav Murugesh
I don't know if you want to open that.
- [176:52]
Tameh Besaroglu
But not by or orders of magnitude. So you could have just orders of magnitude more scaling of the size of the models that are doing the planning than the people or the agents or workers doing the actions.
- [177:04]
Ege Erdo
And I think a third reason is the thing about the incentive thing where you wouldn't face this problem. That part of the reason you have a market is that it gives people the right kind of incentives. But you might not need that as much if you're using AI. So I think there's an argument that if you just list the traditional arguments people have made against why does central biling not work, then you might expect them to become weaker. Now I think that is still there's a danger when you're doing that kind of analysis to fall into the same kind of partial equilibrium analysis where you're only considering some factors and then you're not considering other things. For example, you consider get more complex.
- [177:45]
Tameh Besaroglu
You just have a much bigger economy. And so on the one hand your ability to kind of collect information and process it improves, but also the need for doing that also increases as things become more complex.
- [177:59]
Keshav Murugesh
And I mean one way to illustrate that is like imagine if Apple, the organization today, with all its compute and whatever was tasked with managing the economy of Uruk. Right. I think it actually could centrally plan the economy. The economy of Uruc might work even better as a result. But like Apple as it exists today, cannot manage the economy. The world economy as it exists today.
- [178:18]
Ege Erdo
That's right. I think that's a good.
- [178:20]
Keshav Murugesh
Yeah, yeah, okay. Actually this will be the final question. Look, one of the Things that makes AI so fascinating is that there's no domain of human knowledge that is irrelevant to studying it, because what we're really trying to.
- [178:34]
Tameh Besaroglu
I don't know about that.
- [178:37]
Keshav Murugesh
There's no serious domain of human knowledge that's better, that is not relevant to studying it, because you're just fundamentally trying to figure out what a future society will look like. And so obviously computer science is relevant, but also economics, as we've been discussing history and how to understand history and many other things we've been discussing. Right. Especially if you have longer timelines and there is enough time for somebody to pursue a meaningful career here. What would you recommend to somebody? Because both of you are quite young. I mean, you especially ige, but like both of you. So it's like this is not, you would think this is the kind of thing which requires crystallized intelligence or whatever. Especially given what we said earlier about. Look, as we get more knowledge, we're going to have to factoring what we're learning into building a better model of what's going to happen to the world. And if somebody is interested in this kind of career that you both have, what advice do you have for them?
- [179:28]
Ege Erdo
Yeah, that's a hard question. I mean, I'm not sure. I think there is an extent to which it's difficult to deliberately pursue the implicit strategy that we would have pursued. It probably works better if it's spontaneous and more driven by curiosity and interest than you make a deliberate choice. Okay, I'm just going to learn about a bunch of things so that I can contribute to the discourse on AI. I would think that strategy is probably less effective, at least I haven't seen anyone who deliberately used that strategy and then was successful. It seems like, yeah, I guess not.
- [180:06]
Keshav Murugesh
That I've contributed to discourse directly, but maybe facilitated other people contributing. I guess it wasn't a deliberate strategy on my end, but it was a deliberate strategy to do the podcast, which inadvertently gave me the opportunity to learn about multiple fields.
- [180:20]
Tameh Besaroglu
Yeah. So given if you're already interested and curious and reading a bunch of things and studying a bunch of things and thinking about these topics on the margin, there are a bunch of things you can do to make you more productive at making some contributions to this. And I think just speaking to people and writing your thoughts down and finding especially useful people to chat with and collaborate with, I think that's very useful. So just seek out people that have similar views and you're able to have very high bandwidth conversations with and seemingly, you know, and kind of make progress on these, on these topics. And I think that's just pretty useful.
- [181:00]
Keshav Murugesh
But like how, how exactly? Like they DM you? Like, how do they get it?
- [181:05]
Ege Erdo
Yeah, sure.
- [181:05]
Tameh Besaroglu
And like, I don't know, set up signal chats with your friends or whatever.
- [181:09]
Keshav Murugesh
Yeah, yeah, I've done a lot actually. It's like crazy how much awful I've gotten out of that.
- [181:13]
Ege Erdo
But yeah, I mean, I think like one of the. In fact one advice I would give to people in general, even if they are not thinking about AI specifically. But I think it's also helpful for that. It's just people should just be much more aggressive about reaching out.
- [181:26]
Keshav Murugesh
That's right.
- [181:26]
Ege Erdo
Yeah. A lot of the communication that maybe people have an impression that if you reach out to someone who looks really important and they're not going to respond to you, but if what you send to them is just interesting and high quality, then it's very likely that they will respond. There's a lot more edge there that you can get, which is just being more aggressive and less ashamed or something of looking dumb. That's the main advice I would give. Because if you want to be productive, then again there are these complementarities and so on. You need to be part of some community or some organization.
- [182:03]
Keshav Murugesh
And it goes back to the thing about reasoning alone not being that helpful. It's just like other people have thought a long time and have randomly stumbled upon useful ideas that you can take advantage of.
- [182:12]
Ege Erdo
That's right. So you should just try to place yourself in a situation where you can become part of something larger which is working on the front. That's just a more effective way of contributing. And to do that you have to, well, let people know.
- [182:25]
Keshav Murugesh
That's right.
- [182:26]
Ege Erdo
That's right.
- [182:26]
Keshav Murugesh
And I think just coming to the Bay Area is especially for AI in particular.
- [182:30]
Ege Erdo
Yeah, coming to Bay Area is nice. Just post like just writing things and like posting them in a way people can see them just aggressively reaching out to people with interesting comments provided your.
- [182:40]
Tameh Besaroglu
Like thoughts are interesting and so on.
- [182:42]
Keshav Murugesh
I mean they should. They probably are. Like in many cases I think it's like my thoughts weren't. My thoughts still might not be interesting, but people will tolerate my cold emails and are like, you know, will still like do collaborate with me and so forth. The other thing I've noticed. Tell me if this is actually the wrong pattern or the wrong. Yeah. With people like you with Carl or something is that as compared to a general person who's intellectually curious or reading widely, you tend to focus much more on key pieces of literature than say I'm gonna go read the classics or just generally read like it's like I'm gonna just like put like a ton more credence in something like the Roemer paper. And a normal person might not even read the. A normal person who's like intellectually curious would just like not be reading key pieces of literature.
- [183:31]
Ege Erdo
Yeah, I think you have to be very mindful of the fact that you have a very limited amount of time. You're not an AI model, so you have to kind of aggressively prioritize what you're going to spend your time reading.
- [183:43]
Tameh Besaroglu
Even AI models don't prioritize that heavily. They read Reddit mostly or a large part of their corpuses.
- [183:49]
Keshav Murugesh
Yeah, key pieces of empirical literature at least, at least among you guys. It might not be the most productive thing in general, but I think that's useful.
- [183:55]
Tameh Besaroglu
I also just think it's useful to read Twitter. I think we were having this conversation about people often say that they should like they're spending too much time reading Twitter and they wish they spend more time reading Arxiv, but actually the amount of information per unit time you get reading Twitter is often just much higher and it's just much more productive for them to read Twitter. I think there are key pieces of literature that are kind of important and I think it's useful to figure out what people who have spent a lot of time thinking about this find important in their worldview. So in AI this might be key papers. I don't know, the Andy Jones paper about scaling laws for inference was a big thing. And in economics like this Roemer paper or the paper on explaining long run population from Kramer or from David Rudman and so on. I think just if people who you think are having really good, who think very well about this, suggest a certain paper and they highly recommend it, then I think you should take that seriously and actually read those papers.
- [185:10]
Keshav Murugesh
And for me it's been especially helpful to instead of just skimming a bunch of things, just really stop on if there's a key piece of literature or in order to for example, understand the transformer. There's obviously chirpathic lectures, but one that was really useful is the Anthropics original transformer circuit paper. And just spending a day on that paper instead of skimming it and making a bunch of space repetition cards and so forth was much more useful than just generally reading widely about AI.
- [185:42]
Ege Erdo
Yeah, I think it's just much more important here if you want to prioritize things correctly again to be part of a community, or to be getting input from a community, or get from people who have thought a lot and have a lot of experience about what is important and what is not. This is true even in academic fields. So if you want to do math research but you're not part of a graduate program, you're not at a university where there are tons of people who do math research all day for many years, then you're not even going to know what are the open problems that I should be working on. What is reasonable to attack, what is not reasonable to attack, what papers in this field are important, contain important techniques. You're just going to have no idea. So it's very important to be plugged into that feed of information somehow.
- [186:27]
Keshav Murugesh
But how did you know all this shit before being plugged in? Because you weren't talking to anybody in Ankara.
- [186:30]
Ege Erdo
I mean, you don't need to talk. I mean, the Internet is a pretty useful thing in this respect. And you don't need to necessarily talk to people. You can get a lot of benefit from reading. You just need to identify, okay, who are the people who seem most interesting. And you can also get a lot of benefit of maybe you found one person, and then often that person will know some other people who are interesting. And then you can start tracing the social network. So for example, maybe, I don't know. One example I can give, which I think is actually accurate is maybe you know about Daniel Ellsberg. So you look for a podcast where he appears on and you notice that he's appeared on 80,000 Hours podcast, which he has. And then you notice there are some other guests on 80,000 Hours podcast. So maybe there's Bryan Caplan, who has also appeared on the podcast. And then maybe Robin Hansen has also appeared on the podcast. And then maybe there are some people those other people know. And then just tracing that kind of social network and figuring out who to listen to that, I think that can be really cool.
- [187:26]
Tameh Besaroglu
And I think you're doing a very big service to making that possible. Where I think your selection is often very good.
- [187:34]
Keshav Murugesh
I'm actually curious if you're offline, what I got wrong. Actually, I think I know the answer.
- [187:38]
Tameh Besaroglu
To that and I think that makes it a bunch easier to track who are the people doing the most interesting thinking on various topics.
- [187:47]
Keshav Murugesh
That's right. Cool. I think that's a good place to end with you praising me. No, I'm kidding. Again, I highly recommend People follow Epoch. There's a great weekly newsletter, Gradient Updates, which, I mean, like people plug newsletters but this is like. I can't believe this is a thing that comes out on a weekly basis. And it's like. It's like. Anyways. And you now have a new podcast, which I will not plug as a competitor, but you can check it out as well.
- [188:18]
Tameh Besaroglu
Thanks for lending your studio to. Yeah, that's very generous.
- [188:24]
Keshav Murugesh
Anyways, cool. Thanks, guys.
- [188:27]
Tameh Besaroglu
All right, thanks.