Richard Socher on The MAD Podcast: When AI Improves Itself

Summary

Richard Socher talks with host Matt Turck about why scientific progress is slowing and how AI can bring fragmented knowledge back together. Drawing on his book The Eureka Machine, out September 22, he argues that AI will do for biology what calculus did for physics.

He explains why next-token prediction works as a model of the world, why anything that can be simulated and verified will be solved by AI, and the four pillars of the Eureka Machine: language models, scientific measurements, simulation, and robotic labs, with agent swarms on top. They also cover drug discovery, cancer, his AI Economist research, and worries about jobs and who benefits from progress.

Socher also describes Recursive, the company he co-founded to build recursive self-improving superintelligence, which starts with AI for AI research. He closes with his ten spaces of intelligence and why he thinks we are far from the upper bounds.

Key highlights

• Why progress is slowing. Fields have split into thousands of subfields, so it's nearly impossible to be a generalist, and academia punishes ideas that are too far out. His own early neural-net papers were rejected.
• AI as calculus for biology. Socher says AI will do for biology what calculus did for physics: help weave many well-understood pieces, like neurons or the microbiome, back into complex systems we can reason about.
• Next-token prediction learns the world. Predicting the next word forces a model to absorb geography, and trained on proteins, it picks up which parts end up close together in 3D space.
• "Anything you can simulate, AI will solve." Games, math, and programming fall first because they can be simulated and verified, and Socher expects humans to improve by competing against AI, as in chess.
• Four pillars and a swarm. The Eureka Machine combines LLM knowledge, scientific measurements, simulation, and robotic labs, with agent swarms and scientists working on top.
• Hallucination can be a feature. For discovery, like generating novel proteins, going outside the training distribution helps. For factual questions, he says models should be grounded in search results.
• No crazy hard takeoff. Physics and long clinical trials will still take time, but he says better predictive models let bio companies put several drugs into late-stage trials within 6 to 18 months.
• Recursive starts with AI for AI. The company will focus on automating AI research first, with physical sciences later, and Socher says it plans to ship real products this year.

Watch the interview

https://www.youtube.com/watch?v=kyyLku5F3bo

Transcript

Lightly edited for readability: timestamps follow the source transcript, speaker labels are not included, and obvious transcription errors and name spellings are corrected.

Cold open and introduction

[0:00] Anything you can simulate, AI will solve. But before you know it, you're in this recursive self-improvement loop. And we believe that that will be a great unlock. Boy, are we far away from the true upper bounds of any of the spaces of intelligence, and there's still so much further that AI can go.

[0:16] Hi, I'm Matt Turck. Welcome back to the MAD Podcast. My guest today is Richard Socher, one of the most cited researchers in AI and now at the center of the term everyone in AI is suddenly talking about, recursive self-improvement, AI that makes better AI. Richard just raised $650 million for Recursive, a company built to do exactly that. And his new book, The Eureka Machine, is a fascinating blueprint for how AI and RSI are about to revolutionize science. Please enjoy my conversation with the always excellent Richard Socher.

[0:48] Hey, Richard, welcome back. Great to be back. Thanks for having me.

[0:54] All right, so lots to catch up on. We're going to talk about recursive intelligence. We're going to talk about Recursive the company, but first and foremost, and most importantly perhaps, we're going to talk about your new book, entitled The Eureka Machine, which I read with great interest and would strongly recommend, coming out in a couple of weeks, I believe.

Why Scientific Progress Is Slowing

[1:19] The book opens with a premise that I think a lot of people find surprising and shocking, which is this claim that scientific progress has slowed down, which feels counterintuitive given the number of researchers we have around the world and the sheer amount of money that goes into the space. So why is that?

[1:41] Yeah, it's a somewhat surprising fact, and you may argue clearly not, like we're making so much progress on so many different things. But when you think about how much progress we've made on antibiotics, bacterial infections went from a death sentence and the plague to a nuisance. We now have antibiotics for almost all the different bacteria and we truly solved that. And we have clearly not solved viruses or cancer the same way we solved bacterial infections. When you think about foundational, novel things like E=mc² and general relativity, we clearly have not made progress on many theories in physics either, when it comes to such foundational things that then literally led to nuclear energy and fusion and fission and other kinds of research that could be conceptually done. And so a lot of the fields have gone from "we understood some foundational pieces" to "we can now do a lot of engineering," but they've also split up into thousands of different subfields. It is almost impossible nowadays to be the sort of general genius that can dabble in all of these different fields, because each field takes years and years and years to get really deep into. And so what we found is that as there are more and more subfields and niches, it's actually hard to have enough people in each of these subfields. And [unclear] and others have talked about that and predicted that that will be a big part of why we're slowing down.

The Labyrinth of Human Knowledge

[3:10] Yeah, you have a great expression. You talk about how we evolve from a body of knowledge to a labyrinth of knowledge. 34,000 journals that might as well have no trespassing signs.

[3:21] That's right. Yeah. It's so hard to even understand all the lingo. And I've gone through this myself, first when I started studying linguistics and then computer science, but now that I'm trying to study, and have studied over the last few years, sort of on the side, biology, it's like, man, every time you have a conversation with biologists, 10 sentences in, they're just telling you so many abbreviations and terms that you're not familiar with that most people after a while just space out. And so it's really hard to describe and explain very complex, deep fields to someone who hasn't been in them.

[4:00] You also mentioned that there is some level of a social, human element to this, where in academia you're not necessarily encouraged to take risks, and that goes to your own experience as well.

[4:10] Yeah, 100%. The way careers work is you want to be kind of novel, but if you're too novel, if you're too far out there, then your papers will get rejected. And that certainly happened to me a lot in the early days, like 2010, of neural networks for natural language processing, where the majority of my first couple of papers got rejected from NLP conferences and then got accepted in some small subniches and subgroups. I still remember the first deep learning workshop at NIPS back in the day, which was basically like 30, 40 people, all now super famous folks, but it was just a couple of us renegades who thought that this would clearly be the right way of going about it. And we came from different directions. Feature engineering seemed like not the right path to doing things. Feature learning was a big part of what got us started. Some were neuroscience-inspired, and it was again that combination of different fields that was at that intersection where it's interesting but also often hard to publish well in the beginning.

[5:17] Okay, great. So to play it back, it's fundamentally that idea that there is too much knowledge everywhere, even people within the field are not encouraged to go super far and come up with crazy ideas, and the challenge is to bring everything back together. So the Renaissance men, so to speak, had only a small body of knowledge and therefore were able to come up with cross-disciplinary insights, but this is no longer possible.

[5:35] That's exactly right. A good example again in biology is that you study not all of biology anymore. You study either biology at the cell level, or the tissue, sort of medical level, or the biochemistry level, or the protein level. And if you ask a PhD in biology about a deep question in one of the other layers, they often don't know either.

Can AI Put Science Back Together?

[6:02] Great. All right. So the premise of the whole book is that AI is about to usher, or is in the process of ushering, a whole new paradigm of scientific progress. So just walk us through the high-level idea.

[6:16] I love the idea. It's that science has gotten really good at understanding smaller and smaller pieces, but to bring them back up and bring them together, we actually have to use AI. AI is kind of what calculus did for physics. AI will do for biology in the sense that it'll help us weave back together lots of very complex pieces that build these complex systems that then have certain properties. So concretely, for instance, your microbiome or the brain, they have so many different pieces, and we actually understand many of the individual neurons. Okay, this neuron has these synapses and this is how it fires, and this is the chemistry, which often people still ignore in a lot of neural models. But then as we put more and more of them together, we stop understanding why the whole brain can now have a thought or do certain things. And I think AI is the perfect language, the perfect way of thinking about some of these complex systems, and has the ability to also exhibit sometimes patterns that are almost hard for us to then understand. But of course it's always easier to understand and study a neural network than it is to study the original biological system. And so I think when you put all these together, and I'll talk about the details of the Eureka Machine and its four pillars and so on, and you look at the data, you look at different fields, everything from physics, chemistry, biology, neuroscience, medicine, economics, astrophysics and so on, you look at all of these levels and you see so many small improvements in all of them that can help you extrapolate that AI will usher in this new age.

How LLMs Learn Biology and Proteins

[7:58] There's still a good number of people out there that think of generative AI as a next-token predictor, as a chatbot, as something for language. Here the claim is very different. It's predicting protein structures. Why is the technology that's good at being the best chatbot in the world also good at scientific discovery?

[8:23] I think it's quite counterintuitive. So I'm glad you asked, because if you ask a biologist if we'll have a model that can predict something as complex as aging or different cancers and so on, they'll say no, this is decades out. And in a similar fashion, 10 or 20 years ago, natural language processing researchers would have told you that it is impossible to build one neural network that could answer any and all kinds of questions. In fact, you can find this online on OpenReview, my paper where I described prompt engineering, one neural network that you can just prompt with any kind of question, called DecaNLP. That paper was wildly rejected by basically all the reviewers and the area chair and so on as just completely useless, too crowded, made no sense. Not even humans have one system to answer all these different kinds of questions. And so it's something that's so obvious now, that people can't even invent prompt engineering because it's so obvious. It was such a non-obvious thing to the field. And so we've seen that time and time again throughout different fields. But now, once you showed that we didn't have to actually truly understand and have perfect rules for every single aspect of translation or question answering, we will see similar things, and have seen already, in, for instance, the language of proteins. You have just sequences of amino acids. No human has been evolutionarily trained and has learned to speak the language of proteins. But just like the natural language of English and so on, AI doesn't really care if it's English or a sequence of amino acids. And so you can now generate completely new kinds of proteins, like you can generate new kinds of sentences that have never been in that combination in the training data. And I think when you now apply similar ideas to chemistry and even lower levels such as molecules, you'll see that that idea, that you have very complex systems that with a lot of data can then make very useful predictions, that explore the overall space better than any human could have manually, helps you say, okay, this next-token prediction is such a beautiful yet simple idea that incorporates knowledge about almost any domain.

Next-Token Prediction as a World Model

[10:56] Because there is a concept of a world model, and I realize that term has a precise meaning that I may not be using precisely here, but a concept of a world model that's built into what enables the system to predict the next token. I think in the book you have a lovely example about driving south from Dresden. Maybe unpack that.

[11:23] Yeah. So next-token prediction, why is it so powerful? Imagine you could try to have a model where you understand the world's geography and where every city is. And here we're in New York, so maybe I'll use a New York example. Driving south to New Jersey, for example. Yeah, to Princeton or somewhere, or north to Boston. And so you basically, just by trying to predict the next token, will stumble upon sentences where someone somewhere wrote, "Oh, I was in New York and I was driving north to..." And now it might be Yale, but maybe more likely it's Boston, right? And the more likely it is, probably the bigger the city is. And so you now incorporate, by trying to predict the next word in that one sentence, "I was in New York driving north to," that next token being Boston, you predicted something about geography and locations of cities. And so by virtue of doing that billions and billions of times, in fact trillions of times nowadays, basically as much data as you can, you actually incorporate knowledge about geography in that. And we're seeing very similar things. A lot of people are familiar with protein folding. When proteins fold, some proteins are actually closer to each other in 3D space than they would be in a sequence of these folded proteins. And it turns out that you can analyze the neural network that was trained to do next-token prediction, and you see that indeed the ones that are closer in physical space after being folded, the neural network, just by doing next-token prediction, also has those correlations. And so we basically know that large neural nets trained on very large specific domains will incorporate the knowledge of that domain just by trying to predict the next token, quote unquote, in a sequence. And token, for the non-technical folks, can mean an English word, can mean a sub-phrase of an English word, just a sequence of characters that together make up a word, but a token can also mean a protein, it can also mean a piece of a few pixels in an image or in a video or in a piece of sound. So everything can be kind of tokenized, basically discretized into a set of vocabulary tokens that are then able to be predicted.

[13:36] By the way, if I may, you know, you're talking to an extraordinarily accomplished AI researcher when it comes to your sense of East Coast geography: what's south of New York is Princeton and what's north of New York is Yale. [laughter] Is there a part of the analogy that breaks down, if you have this implied idea that across physics and bio and what have you, this language does truly translate as an analogy?

[14:01] So there are clearly some places where it feels weird, like it feels oversimplified, and you could clearly say all models are wrong but some are useful. And that's also true. We know proteins aren't just sequences, they have a 3D structure, for example. But it's surprising how far this analogy can be pushed. Even in chemistry, you have various loops within molecules, but somehow there's a standard way of describing molecules as a sequence, as a string, and using that is very helpful for a variety of different aspects of chemistry too. So yes, it's oversimplified, yes, all models are wrong, but this particular set of models is quite useful for all the different sciences. And then, of course, you can go into arguing, but how truly novel can ideas be from some of these models? And the truth is that in almost all of science we stand on the shoulders of giants, and just exploring the recombination, the clever combination of all the different ideas that are already out there, will lead to incredible progress. We have a lot of the foundational pieces figured out in small pieces, but to weave them back together will lead to incredible progress, especially in biology. And then you can also explore how well these models can create novel ideas. And there we also have real examples of researchers having had an AI ideate and then a few months later people publishing a paper with essentially the same idea. Jeff Clune, one of our co-founders at Recursive, has tweeted about several such things happening, where he had novel evolutionary ideas or evolutionary algorithms create ideas that later have been published by people also. So clearly the novelty threshold has been met in several cases. Now, will it derive a completely novel way of looking at the universe? Maybe be able to disprove or prove any of the many string theories that are out there and things like that? There's still some research that has to be done. But I'd argue that, with existing technology and giving it more data and more compute, we can already cure a lot of diseases. We can already cure many of the aspects of aging. We can already build better fusion reactors and systems. We can already create new materials. You can already help with economic questions of how much to tax or subsidize certain populations if you have a certain objective or reward that you're looking for to achieve inside your economy. All of these things are already within our grasp.

Can AI Generate Truly Original Ideas?

[16:44] I was going to visit that idea later, but since we're on it and it's such a fascinating concept and so central to the whole discussion, let's double-click on it. So indeed, as we talked about, recursive self-improvement, RSI, one of the big questions has been creativity and the ability to think out of the box. And you also hear people talk about Move 37, which was the move in Go nobody expected and that enabled the DeepMind model to beat Lee Sedol. What is the mechanism by which this can or cannot happen? Do we understand that, or is it precisely the fact that we do not understand it the reason why we don't think AI can be as creative as it could be?

[17:31] I think we can basically predict where AI will certainly have superhuman capabilities...

Simulations, Verifiers and Superhuman AI

[17:40] ...and those are all scenarios and all domains where we can either have a simulation and/or a verification tool. Any kind of domain that we can simulate will basically result in a world where the AI can essentially infinitely many times experiment inside that simulation. And assuming the simulation doesn't take years to run every time you want something useful from it, you then know that you can solve the problems in that domain. And so all the games, I was never that surprised that AI would eventually, fairly soon, be better than us in games, especially the games that are perfectly visible. They don't have hidden variables, like you don't know other people's cards. In chess and Go you see everything and it's all on the board. Now, there were too many combinations on the board to just do brute-force winning of those games. But if you have enough training data, the AI can learn the intuitions also behind it by playing many, many games. And in this case in particular, it's even easier because the AI can play against itself. And that idea to play against yourself is also a big part of open-endedness and recursive self-improvement that we've applied, and I can talk about rainbow teaming and security and safety research for example with LLMs also. But yeah, just to go back: simulations, anything you can simulate, AI will solve. Now, what else can be simulated that's interesting and can be verified that's useful? Math. Math is going to change massively within the next few years. Terence Tao and the most famous, frontier mathematicians are already fully aware of it. The whole field will change, just like the field of AI has changed, and a lot of skills that used to be useful, where you manually feature engineer and then you manually architecture engineer and you manually do these things, are not that useful anymore. That will be true for a lot of mathematics also. And so I think that's a great situation if you cared about solving as many things, proving as many theorems as possible in math. It's not a great thing if you just love the pursuit of math for intellectual sake and for fun. And we'll actually see that play out in many different ways. Chess is now more popular, despite being dominated by AI, if you wanted to. I think most sports, intellectual and physical, in the future will probably benefit. I don't know if you saw the robot running, really funny, in the Chinese Olympics recently. My hunch is humans will try to see if they can emulate that funky run to then run faster too, because AI in simulations, in the robotic simulations, tried many different ways of running and found this weird new way that somehow humans, despite having run all our existence, haven't thought of yet. Right? And so just like the best chess players and best Go players are better now because they can compete against an almost perfect AI, I think even that will be true for athletes in the future. So I think it will push the field forward. And what's the most powerful thing we can simulate and verify and build verifiers for? It's programming. "Software is eating the world," famously said, I think, Marc Andreessen. And so AI is eating software. So now you can basically, one simple example is you can show the picture of a website and you say, make it, program such that it looks exactly like that, right? And then you can create infinitely many examples like that and then have a perfectly capable AI building front ends for websites. And the truth is you can create all kinds of verifiers like that, and so all of programming will change, and that will be a major impact on the entire digital economy, which is the knowledge economy and so on. And then the question is, where does it stop? Well, what things are hard to simulate right now? And that's where it becomes interesting to look at the natural sciences, where we cannot yet perfectly simulate a complex cell, let alone tissues or organs and full humans, and we do need to collect a lot more data in various robotic forms. And those are essentially the four pillars that I talk about in the Eureka Machine. One, start with human knowledge and LLMs. Then the second pillar is all the measurements we can take, and more and more of those that we already have, and we should incorporate that into the model. The third thing is simulation, and the fourth pillar is essentially robotic process automation to collect even more data and verify whether the inventions really made sense. And on top of those four you have an agent swarm and a community of scientists.

The Path to Recursive Self-Improvement

[22:15] So we'll definitely unpack that part in a minute. Just going back to... there's the question of intuition and creativity, which is one of the unexplored frontiers, and then related to what you just said there's also the big question of generalization. So starting with coding and math, that seem to be increasingly conquered domains by AI, you seem to be saying that then there's the life sciences that could be next. What is your sense for how far we can go in making AI truly general, and what is the process to get there? Is that brute-force RL for this domain and then that domain and that domain, or is there a more sweeping generalization effort that could be produced?

[23:13] So at Recursive we are fairly sure that we have to start with AI for AI and then make it really, really good at doing research on creating better AI, so that it has the equivalent of 50,000 PhDs in terms of knowledge and its own capabilities, and only then go after the physical natural sciences like physics, chemistry and biology, especially biology, I think, will be most interesting. I do believe that in the next two or three years, while we're focused on recursive self-improvement also, we will have more and more data collection. Tahoe Therapeutics is a great example. Parallel Bio is another one. I think I mentioned both of them in the book, that basically help create much, much more training data. And then in the case of biology, for instance, you can do these perturbation studies, like you take a cell, you try to knock out one gene and you see what happens, when I knock out this one gene or I add this one molecule to it and I see what happens. So if you do many, many perturbation studies, eventually maybe the AI will learn the underlying patterns behind it, just like it learned the underlying patterns of, oh, I'm in New York and I'm driving north to, and then predicting Boston. It might be like, oh, I add this molecule to this kind of cell and then I get this output. And then it can start to eventually generalize. But we're just nowhere near having enough training data for biology, and so we need organoids, we need eventually all these perturbation studies to come together so that we can then try to build a virtual cell, and then in that virtual cell the AI can go and experiment many times.

Why AI Hallucinations Can Drive Discovery

[24:52] There is another counterintuitive idea in the book that I thought was fascinating, which is that when it comes to AI-based science, hallucination might be a feature rather than a bug. Can you explain?

[25:09] Yeah. So a lot of folks struggle with hallucinations in models, for a long time especially in the earlier versions of these models. The book, I started thinking and writing the first ideas down like three years ago, and had to change a lot of chapters, because things where I'd written "someone should do this" had since been done, and I talked to those folks and wrote about their startups and whatnot. But I do think hallucinations can also be very helpful for AI when you want it to explore novel kinds of proteins. Yes, every AI can memorize things, every computer can easily memorize things, right? But where it's interesting is, how well can you hallucinate? How reasonable, or just outside of the distribution in some interesting way, are your predictions, right? And we also know that we can, just like with humans, right, you give them a certain kind of molecule and their visual cortex goes off into a really different world. You can also increase the temperature, as it's called, a technical term, so that the large language model will generate tokens that are more and more different from things it has seen before. And so I think hallucinations are in some cases a feature and not a bug. Of course, when you ask a factual question online to a search engine or an LLM, then you want to have it be correct. And when we know that that's the kind of question you're asking, it's easy to prime the model and say, well, here are search results. That's what You.com does, of course: basically take the facts from a real search engine built for agents, plug them into the prompt, and then the AI will summarize that. And so I think initially people thought, oh, we need symbolic reasoning, blah blah, to do all this. We just needed more examples of "don't hallucinate now, take real facts from a search engine and then mostly summarize those," and then those hallucination problems were to a large degree resolved. And then if you want to write a poem for your wife, you don't want it to just sound like the other poems that were out there. You want to create a new one. You can also do that.

[27:13] And as a funny moment in the book, you mentioned that actually a lot of scientific discoveries were made by scientists in a semi-state of hallucination, through diseases or otherwise.

[27:20] That's right. Yeah. Heisenberg and other physicists, and there are all kinds of interesting stories about [unclear], and in some cases also actual mental states that were eventually quite unhealthy, psychosis and so on, that have in some cases pushed the field forward.

From Reading Biology to Writing It

[27:47] All right. So you alluded to some of this, but let's take some of the life sciences as examples just to unpack some of the thinking there. So starting with medicine, the deeper shift that you described is going from reading biology to writing it. And your own team did that, was one of the first teams to do it. So do you want to tell us what you guys did and what that means in terms of where science is going?

[28:11] Yeah, I think when I first studied biology, in high school, I never, to be honest, loved it back in high school, because it's just memorize these processes with all of these different pieces. You write them out, you get an A and then six months later you mostly forgot about that process. And so that wasn't that interesting to me. But what's changed in the last few years is that biology is becoming a programmable science. It's becoming an engineering science. And that's often the case, I think, in the transition of different sciences. Once you've understood most of the basic pieces, you now want to learn how to put them together in novel ways such that they are useful for you, and there's low-hanging fruit when a field transitions into becoming sort of an engineering science. And I think biology is in that state right now, where we know, okay, this protein does this, but if we change that protein a little bit, maybe it can do something else. And you can package, you know, sometimes you can connect different things, so one piece, for instance, attaches to a cell, but then you can have different loads, like connectors to it. So once it's attached to the cell, you can actually inject something into the cell, and now you can recombine these molecules. And so I think that engineering aspect of it is truly exciting. And the first aha moment for us was, I think, in 2018 when we trained the largest language models for proteins. It's called ProGen. Ali Madani is the first author of that paper. That was back in the day when I was the chief scientist at Salesforce still. And he's since started Profluent. They've now closed multi-billion-dollar contracts with Eli Lilly, at Profluent and his company, because they've created new kinds of proteins that are, for instance, even better than CRISPR-Cas9 at gene editing, and being even more specific and targeted for changing certain genes inside living people potentially, and creating new kinds of therapies from that. And so proteins being such an important piece of all the building blocks of life and disease and health, making them programmable will unlock very, very obviously many, many exciting use cases. And I think you're starting to see this in this recent trial that is making a lot of progress, where they basically created a different drug for every different patient in the trial. And this is a first for the FDA too. And more will happen there. It's actually unfortunate how hard it has become in the US and certainly in Europe to run clinical trials. And so a lot of folks are now moving to either China or Australia for their clinical trials. Interestingly enough, in China it's cheaper, it's faster, but you also have to worry a little bit whether your IP gets sucked into the ether and is gone. And in Australia, they had a clever move where they actually decentralized clinical trials and every hospital can run its own clinical trials. So all of a sudden, you get competition instead of having one centralized decider on which clinical trials to run and how to sort them and all of that. And so, yeah, anyway, there's not enough people in Australia, so it would be great to get that kind of system happening in the US too. But yeah, clinical trials will be more and more efficient over time. We'll collect more data and then the AI will be able to automate more and more of that.

Can AI Accelerate Drug Discovery?

[31:37] And when you think about the drug discovery and creation life cycle, from initial intuition to being available, that takes, what, 10 or 15 years. What are we talking about here in terms of accelerating discovery? Realistically, what portion of the process does it shave off?

[31:58] It's a good question and it sort of touches upon what some people call the hard takeoff, where some people think once we have RSI, and generally with AI, there will be this really hard takeoff and then everything will just happen very quickly. And as bullish and excited as I am about AI, I'm not a believer in this crazy hard takeoff. I think yes, things will accelerate, but there are certain things that will just require time because of physics and constraints in the real world, such as long-term trials, where you want to know whether people have some issue like three years after they stop taking the drug and things like that. And so there will be some delays. But the biggest difference is that the whole bio market, and a somewhat contrarian take that we have at AIX Ventures too, a lot of folks think bio is just a terrible space to invest in, because in the past a lot of drug companies spent eight, ten years, they finally, you know, have to go public because there are not enough late-stage bio investors. So they go public with one drug or maybe two drugs in late-stage trials, like stage three, and then the stage three trial fails and then the whole company is dead. Now what we're seeing is the difference: we now have companies that, instead of having one molecular drug after eight years in late-stage trials, actually within six to 18 months have multiple different drugs in late-stage-two trials already. And by the time they go public, it'll be with eight-plus different drugs that are also much more likely to succeed, because we have better predictive models.

Will AI Help Cure Cancer?

[33:35] So clearly we are living in a moment when AI has become quite controversial, whether that's the jobs question or the data center question. The number one thing the industry keeps saying as a way to justify why AI is a great thing is, AI is going to cure cancer. What is your sense of the reality of that claim and what it's going to take to get there?

[33:58] There's a lot to unpack there. Maybe at a very high level, I think if you care about the outputs of an industry or a company, then you love AI. If you get paid hourly, you probably hate AI. And so AI, in the positive instantiation of this future, is a huge force towards more entrepreneurial thinking. If you're an entrepreneur, generally you kind of love AI because it's making your things more efficient. You just get more done. You have an unlimited list of things to do if you're a startup founder or just running a company, and to have AI do many of those things for you just means you can do a lot more. But if you're being told you're training your replacement and all your data is being collected hourly, then at some point those hours will end and then the AI will just do the thing you just taught it how to do. And so it's understandable that if they have this very unentrepreneurial mindset of just getting paid by the hour and they don't own any equity in creating that IP, then they're understandably unhappy. And then you can go one level deeper and think about, well, what is the impact on jobs? And my theory here, after thinking about this for quite some time, is largely dependent on the elasticity of the demand of the product when its prices go down. So concretely, for instance, illustrators. Illustrators hate AI. The world needs a certain amount of illustrations. Because of AI, you cannot charge 200 bucks anymore for one illustration. So now any little blog post has illustrations. If your goal was just to see more illustrations in the world that are specific to a text, you love AI. If you got paid 200 bucks for one illustration and now it's worth two cents, maybe you hate it, right? And so the problem was that the demand for illustrations didn't go up 1,000x when the price went down by a thousandfold. It didn't grow because you just don't need that many illustrations in the world. Now in coding it was a very different world. Actually, as coding got cheaper and cheaper, you had this famous Jevons paradox that everyone's talking about now. I think I was the first, at least I didn't see it online for a while, it's an interesting fact from history: the thing got cheaper and cheaper, but we actually used more and more of it, and for coding that will definitely be the case. And so we're seeing actually more demand for programmers now, because they're so much more productive when they use AI, and anyone ultimately could have dozens of apps on their phone that are unique to that person, that are modified in some way and very special. And there are so many other ideas that people didn't explore because maybe the market wasn't that big, but now that you can just create an app really quickly, why not? And so I think that is another aspect of jobs. And so, going back to cancer: yes, I do believe AI will play a big role in curing multiple cancers. We're seeing trials now where AI is being used to make a specific cocktail of drugs, create specific RNA sequences, and so on for the types of cancer, and each cancer often is also not one homogeneous thing. It has different types of subcancers in it and so on, and you need to specialize treatments for each person and for the various different forms of the different cancers that you can have. And so all of that is much, much more feasible to be done with AI, and we're seeing it.

[37:35] Is that precisely the point, that cancer is just extremely complex and ultimately a systems problem that AI is uniquely equipped to solve?

[37:42] Exactly, exactly. So yeah, it will take some time, and obviously even if AI, let's say, had the perfect molecule and said, okay, for this type of cancer this is the molecule, I came up with it, you'd still have to run it through many clinical trials, and it will still take years to come out. So everything in biology just takes longer than it does in software.

AI Breakthroughs in Biology, Energy and Materials

[38:06] What else are you optimistic about in that field? Your predictions for the next decade: rare disease cures, organs designed for individual patients, pollution-eating synthetic cells, what else? You covered some of this in the book. What are you most excited about in terms of what may come first?

[38:21] Yeah, I'm excited about all of these things. I think we can design bacteria that eat microplastics, and once there are no more plastics, they just die. I think that would be extremely helpful for the oceans and so on. Obviously you have to be very, very careful that they don't somehow mutate into eating other things and so on. So when you mess with the environment at large scales, it's important... humans have done that many times and sometimes it worked out pretty well, many other cases maybe not so much. Forests are a good example. People deal with forests too much. They don't let small forest fires happen, and then they get even bigger because the small ones didn't clear out the underbrush and so on. Everything is a system. Everything is a complex system. And we need AI more and more to do some of that engineering better than we've done in the past. And so I'm excited at all the different levels. When you look at how to balance a plasma in tokamaks for nuclear fusion, that's already a machine learning control problem. I think we'll have a better handle on that. So at the lowest level of physics, clearly there are more and more materials, more efficient solar cells and solar panels that we can design with AI. There are companies I've invested in that do that. Better batteries, better materials, so we don't need only lithium. We can try to build batteries with more abundant molecules that are easier to get and mine, with less pollution. And especially again in biology, we're seeing a lot of things. I think it sounds like science fiction, and I understand the famous saying of, if you want to know why something doesn't work, ask the experts. I think that was true in natural language processing and neural nets, and I think it is currently also true for longevity and cancer and other kinds of research with neural nets applied to biology and medicine. I do think we'll make more progress than the most skeptical people think, but we also won't have a hard takeoff, again, because things do require careful experimentation in medicine. And I'm personally excited for all of these things. But I think if you're mostly interested in making humanity more productive and more efficient and creating more outputs and growing, then you're going to love AI.

Will Some Societies Reject AI?

[40:24] But I also, in some ways it becomes a philosophical question, and I think we already observe many subcivilizations or subgroups of people, and cultures, that have essentially off-ramped from progress. If you're living on some beautiful island in Greece, you don't really think about AI. You don't have to think about AI and you just enjoy life. You go fishing, and sometimes there's a storm and things are bad, but most of the time the weather is good, the fish are abundant and you just kind of live your life. And so I think there will be different groups of people who will want to off-ramp from civilization's progress, right? There are already people who prefer to live way deep in the countryside and never go into the big city and so on. And I think we'll have more of that. In some ways I personally love progress. I think scientific progress especially is what helped humans solve most of the hard problems that were in our biosphere. David Deutsch has a whole section in his book, The Beginning of Infinity, which I highly recommend people read too, where he talks about how there are all these different material problems and we came up with solutions thanks to science and better explanations and better research. And I'm personally all for that, but some people will not want to participate in that world anymore. And I think AI is such an accelerant that it makes that question even more pertinent for people.

[42:05] That's fascinating. Just to keep going down that path before we go back to our little tour of frontier science, how would that manifest? So we would end up with groups of people that would deliberately opt to just not participate in progress. I guess progress has been jagged throughout humanity in different regions obviously, but as it spreads and as the world keeps going more global, those people make a political decision to organize around a principle of non-participation in AI. Is that city states, that kind of stuff?

[42:39] Yeah. A sort of example that I'd love to visit actually is Bhutan. Bhutan decided we will not measure our gross domestic product based on money but based on happiness, and happiness mostly for people who want to keep it simple and have a simple life, not want to build startups and so on. I'm pretty sure those folks aren't quite as happy in Bhutan, but overall Bhutan is just very green and cares about the environment and cares about a specific subset of religions, and people are more often content in keeping things the way they are rather than trying to progress in various different ways.

[43:22] This is why this book and this conversation today, from my perspective, is so important, right? I think the AI industry has done a terrible PR job in general. So if you and others can clearly articulate why AI is good, that may hopefully unlock some of this debate.

[43:37] Yeah, it's really interesting because clearly people use the technology. If no one used ChatGPT or Claude Code, there wouldn't be a problem. People clearly like it. It's just that the people who get a lot of use out of it are not quite as vocal, and there are negative things. There's also some amount of moral panic about chatbot friends, in similar fashion to how novels used to be a really bad thing. There are all kinds of stories of older people saying, oh, these novels are ruining the youth, they're now living in these dream worlds and are distracting themselves from the real world. The Sorrows of Young Werther, a very famous book in Germany, actually led to some suicides, which is really sad, and now it's the book every German kid has to read in high school and it's a high form of literature in Germany. And then comic books were really bad and computer games are really bad, and there are various levels of that, and currently the chatbots are really bad. But there are also clearly a lot of people who get a ton of value out of these chatbots, and now all of a sudden you make access to medical advice cheaper, legal advice cheaper, and sometimes also emotional advice cheaper. But you don't hear many of the people, the many hundreds of millions of users of these technologies, talk about how much this helped them not commit suicide or not be very sad and dysfunctional and so on. So I do think you're right, in some ways it's not just AI, I feel like the future as a whole needs better marketing.

Building the AI Economist

[45:07] All right, going back to our tour, because I want to make sure we cover some of the fascinating parts of the book. So we talked about drug discovery. We talked about computational biology. Another fun example or domain that you mention is economics, with a fun stat where you said economists failed to predict 148 of the last 150 recessions. And so your team, while you were at Salesforce, built an AI Economist that basically operated on a simulated society and you came up with policy recommendations that were better than the state-of-the-art [unclear]. Walk us through that.

[45:54] Yeah, economics is a really interesting field that unfortunately doesn't have obvious benchmarks the way computer science and many other sciences have, where you just say, if you do better in this benchmark you clearly have the better ideas, the better algorithms, and we should all learn and study those. When we submitted these papers on two-level reinforcement learning systems to Nature and Science, they just desk rejected them. In one case, some random ethicist who had no idea about AI, it was just desk reject. "I'm not even going to read the full paper because AI for economics with reinforcement learning is just a weird thing." And so it was just gone. And so because of that, economics often becomes just a political field. And if you're in one economics department that has a certain political slant and direction they want to see the world move into, you just have to write papers that make sense for that political ideology. And so that unfortunately makes it very hard to do more objective research. And so we tried to create the simulations, a very simple simulation where you have a bunch of agents. This is from 2018. The agents were much, much simpler back then. They just had a certain utility function. They had certain hours in the day that they would be willing to work. They were sampled from certain priors that you may make assumptions about. Not everyone wants to work 14-hour days. But some people do. You basically make all these assumptions. And then you let these agents collect resources, build houses, they can block other agents from those resources to try to build monopolies and become even wealthier. And then you had a sort of meta agent that looked at all of these other agents and basically chose how to tax and subsidize different groups of agents. And in that fairly simple simulation, you could essentially give it an overall reward. In our case, we said, let's maybe start with equality times productivity. You want the economy to grow, but you also don't want one agent to have access to everything and everyone else is really poor. And so obviously you don't want just equality and you don't want just productivity. So you have a combination of these two, multiplicatively. And now if you agree that that's a good reward, you could have politicians say, well, I'm going to do this and that to help, for instance, the middle class. But if we had a much larger scaled-up simulation, you could then run their one proposal through billions and billions of years of simulations of taxation and subsidization to say, well, will that proposal really result in that outcome, given the goal that you have? Or maybe, probably, if you simulate billions and billions of years of different tax years, maybe there are better ways. And what we found is that the agents will try to avoid taxes by dumping a bunch of stuff before, or making a bunch of gains just after the tax year, and so on. And the funny thing is, that paper... the baselines that the field uses, one very famous formula is called the Saez formula in economics, and it's beautiful math and it shows that provably it's the optimal taxation scheme, but it's the optimal taxation scheme in a one-step economy where you make one economic decision and then no other decision again. And so we showed that this very complex RL system basically recovers that, and does come up with the same solution. But now you can actually deal with the fact that economics is a temporal sequence of many different decisions, and you can learn and adapt, and there are counter-adaptations from the agents to certain taxes and subsidy schemes. They're trying to play things, and then you can still simulate it. And so my hope is eventually that that paper will have kind of a GPT-3 moment, where someone actually scales it up, builds a really realistic simulation, and then we could have AI give us feedback. Obviously we don't want to let the AI make those decisions without any human oversight, but at least have some economic policy suggestions on how to most objectively try to achieve the goals we want to set. And of course humans then have to really formalize what the goal of our society is. And in many ways these are very deep questions that philosophy and political philosophy have asked many times. Socialism, capitalism, maybe social market economies where there's some regulation in healthcare but maybe not in other areas, and you want competition. You can actually define once what your real goals are. So I think hopefully over the years this kind of system will help us run economics much better and make it a much more objective science.

[50:35] Do you think that's realistic, that we could model all of the economy with all its nuances? There is an emerging space around simulation of worlds, and a couple of exciting companies in the space, but at the same time the economy is a lot of rational decisions but a lot of irrational stuff. It's very human. There are fears, there's greed. Can all of this be modeled by AI?

[51:02] All models are wrong. Some are useful. I think we can make those models more and more useful, and they'll be less and less wrong. We've seen surprising results where you can prompt an LLM and say, you are now a 43-year-old from this region, blah blah blah, give them all kinds of prompts on what they're supposed to act like. And then after having trained on tens of trillions of tokens on the internet, they can say similar things to what people might say from that setting. And so I do think these models will get better and better. The fidelity of the simulations will get higher, and once they cross a certain threshold then the recommendations from such a simulation with an AI could become more useful. I don't think this is very feasible in the United States for a very, very long time. There's just so much identity politics and special interest groups, and how super PACs get funded, that it's very, very unlikely to be used. My hunch is Singapore and China will probably be more likely to try to use those ideas, say, hey, we all agree, or we at least make it very clear that this is our objective function, and then we're going to really try our best to set the various taxes and subsidies and so on in a way that really achieves that objective function.

The Scientific Data Bottleneck

[52:24] Okay. Great. All right. So we talked about, again, drug discovery, computational biology, and the economics aspect. You talk about astronomy. You talk about neuroscience. So I would again strongly encourage people to read the book and hear all the stories and all the nuances. Let's talk about the Eureka Machine itself. Like you alluded to, four stages, and as we get into that, there is also the question of the quality of the data that is fed into all those machines, because if you train AI on a lot of AI data, you inherit all the biases and the assumptions and all the stuff that is just wrong that is spread out through all of human history.

[53:05] Yes. I think AI often is only as good as the people, the data, the systems, the infrastructure, the rewards that we give it. And we have to be very careful about how we design and filter out all of those things. I think we have more and more control over it, but it is still surprising how poorly engineered some of the environments are, and some of the sandboxes, that frontier labs use for AI.

The Four Pillars of the Eureka Machine

[53:41] So let's get into the machine itself. You've got four core pillars. Walk us through the first one.

[53:49] So, yeah, the four pillars I briefly alluded to earlier. The first one is just large language models, essentially to try to ingest the world's knowledge into the Eureka Machine. And I think the interesting bit here, actually, is that in some ways there's this weird cycle that happened, that I don't talk about in the book as much, but it sort of lived through this now, which is: the few large closed labs, Anthropic and OpenAI, took almost everything they could from the open internet and trained a model, but then the Chinese open-source companies basically siphoned a lot of that knowledge out of those closed-source models by distilling it. But then they open-sourced the model back into the open domain. So now the knowledge is back in the open internet where it started. And so I think it's very clear that that first pillar, of having access to all the world's information, being able to reason through all these different concepts and the crazy large combinatorial space of existing knowledge, is the first pillar.

Teaching AI the Rules of Reality

[55:01] So pillar two is a model of reality itself. So what does that mean? If you think about how limited human perception and the current set of human knowledge is, and how we could actually expand that, you have to look at scientific measurements, right? We cannot observe gravitational waves. We cannot observe gamma rays. But we can build tools and scientific machines that measure these things for us. And so that is a clear second pillar that is different from human knowledge, that in some cases hasn't been fully described in human language and in some cases might be very complicated to describe in human language. We can already say, oh, a neural network predicted this word because of these five million parameters, but it's like, okay, well, you just list them all out, but you don't gain an intuition because the system is so complex. Similar to how no one can really say why did you move this muscle fiber in your pinky when you try to move the steering wheel. No one has access to that in their brain. And even if they did, it would just be, because of this very complex system. And so that is basically the ability of an AI to take in all of these measurements and start actually digesting them and taking real knowledge, extracting it from scientific measurements.

[56:26] And pillar one sounds like it already exists. Does pillar two exist? How do you teach a machine the rules of the universe?

[56:39] So you'd have to really collaborate with a lot of different sciences to put together this kind of foundational model of physics, chemistry, biology and larger and larger systems, and then have many universities and labs work together to bring all of that into one model. So I think we've had the projection of humanity's knowledge onto the internet, but there are just lots of things that don't make sense to put up on the internet, and so those things are still hidden. And many companies are now working on, quote unquote, foundational models. Some of them have now rebranded them as world models, with similar technology underneath, where they basically try to ingest as much information about one domain, and then they build a first example of a virtual cell that is particularly good at estimating particular gene variants or something but not lots of other aspects of a virtual cell. A virtual cell is a good example of a goalpost where many different teams would have to come together and bring all of that data into one unified model. In its full glory, that doesn't exist yet.

Simulations and Virtual Cells

[57:50] Okay. And then pillar three, again we're describing the four pillars of the Eureka Machine, which is this superpowered AI scientific discovery machine. So pillar three is simulation. That goes a little bit to what we were discussing about economics. So would you create different simulations for different domains, or one simulation for everything?

[58:10] In a perfect world, we'd create one crazy simulation for everything, but there are obviously different levels of abstraction. And sometimes for most aspects you actually get away with not having to simulate all the quantum details of a very complex subatomic particle. You can just say, all right, these are the molecules, and then you know how in chemistry those molecules will work together based on valence shells, blah blah. And then in biology you sometimes can just abstract from, oh, I don't even care about that molecule, I'll just say this is overall this protein, and that protein connects to a cell at that level. So as you try to build it all together, you have to have computational efficiencies and abstractions that humanity has been good at building, and computer science is particularly good as a field at building. No one has to program in zeros and ones anymore. They can now program in English, and a lot of the abstractions can be ignored. I think similarly in these physical simulations we can ignore more and more levels down. But sometimes there is sort of quantum biology and there are maybe some effects that we didn't realize and oversimplified, and those might come out from one large simulation in which the AI can then try to experiment. And that level three, or pillar three, exists in bits and pieces, in many small bits and pieces. The simplest example is a simulation of Go or chess, right? That's like, okay, we have it, it's easy. An interesting new one that many people are working towards now is a virtual cell. If we had that... a virtual cell is so complex, a real human cell is so complex. We're very far away from that. But I can see how, with enough people coming together with enough funding, we can eventually get to a fairly useful model of a virtual cell.

[59:58] Okay, great. And then pillar four is the real world.

[1:00:04] That's right. At some point, especially in biology but in all other fields, you have to engineer a system. You have to really put it together to see if you missed anything in your simulation, any confounding variables and so on. You have to really run experiments in the real world. And obviously in the smaller case of physics, chemistry and biology you can do that in a lab. At some point you have to build real machines and really get out there, build satellites and whatnot, and take measurements of the universe at all kinds of scales. And so I think it makes sense for us to put more and more resources behind that, as AI has gotten really, really good at the first three pillars.

Self-Driving Robotic Laboratories

[1:00:43] So there's some sequence to it. And is the future a concept of self-driving robotic labs, and if so, how far away are we?

[1:00:49] You know, I love that there are first efforts in this, like Periodic Labs is a great example of that. I love that we're starting to think about this. Personally, from an investing perspective, I feel like it's a little bit early, but in two to three years I think it'll be right on time. We'll have figured out a lot of the software. We'll have gotten really good at the LLMs, the scientific data, and connecting that also potentially to LLMs, building even more high-fidelity simulations, and then we can ask the AI to come up with really good expensive experiments that can take sometimes hours or days or weeks to really run through. We'll have better organoids, or tiny cell systems, where you can, based on human-derived stem cells, Parallel Bio, for instance, for human lymph nodes, immune cells, and then you can experiment with those cells more quickly, and that is done with robotics already. So few real examples of that exist. I think there's a Chemputer too, in chemistry, that can put together a small set of molecules. There are first examples of this with Parallel Bio, with organoids and doing clinical trials with that. By the way, that company alone has already gotten FDA approval to skip certain animal trials. So you save many, many millions over the next few years, and lives of animals that are just bred to then be tested upon and then dissected and evaluated. If you love animals, you can also love AI, because AI is now already, not just will eventually, but through this one company, Parallel Bio, already saving animal lives that are just bred for being tested upon. And so I think there's tons of really amazing work that is very targeted to build these out, and more and more generality is kind of what is required to then allow the AI and the agent swarms to sit on all of these four pillars more efficiently.

Agent Swarms and Open-Ended Discovery

[1:02:51] Yeah. And to finish the tour, there is this agent swarm. So what do the agents do? Do they decide which experiment to run, or is a human still deciding what the machine runs? And do the agents measure what's coming out, or is it a human measuring it? How does that work?

[1:03:14] The agents will ideally work on as much of the scientific process as possible, similar to how scientific communities do it. And in many cases the evolution of science and culture, and even biology, has aspects of open-endedness, which is very inspiring for us at Recursive also, and is actually basically exploring interestingly different ideas highly in parallel that can then be recombined. And so these open-ended processes have led in biology to everything from our fingers, eyes and brains. In technology there are lots of examples, and Jeff Clune, one of our co-founders at Recursive, talks about this a lot: how you can't get a microwave if you just say, make this pot faster in heating up my food, right? You just add all kinds of pressure and so on, but you had to work on radar technology and realize, like, a chocolate bar in your pocket was melting as you worked on radar, to then eventually get to a microwave to warm up your food faster. And so there are these different paths and recombinations of different research ideas that can come together, and we can model that better and better with agent swarms.

The Compute Bottleneck

[1:04:30] It sounds like an incredibly compute-hungry and data-hungry machine, given the complexity of what it is that we're trying to model, especially as we think about cross-domain pollination. Do we have enough compute? Do we have enough data? I mean, Ilya said we've reached peak data. Does the machine need to create its own data? What are the constraints?

[1:04:52] Indeed. Compute is the biggest constraint. I think in the future more and more humanity, and already we see this inside different companies, will have to decide what problem is worth solving, how much compute do we give to solving that problem, and then there will be new kinds of scaling laws where we give enough compute to really solve different kinds of problems. And yes, I think the majority of the public internet has been digested by a lot of these labs, but there's always new data, there's always new things that happen in the news. That's why at You.com we work with a lot of neolabs and other labs also to just give them constantly new search results when they ask about something that just happened last week and wasn't yet part of any training data set.

Inside Recursive

[1:05:46] All right, thanks for that. So a lot of those ideas are embedded in your new startup called Recursive. Tell us about the company.

[1:05:54] Yeah. So Recursive started with the goal of building recursive self-improving superintelligence to automate knowledge discovery and scientific discovery. And the eight co-founders came together, and we all in one form or another came to the same realization but actually from very different directions. Tim Rocktäschel and Jeff Clune, for instance, came very much from this open-endedness research direction of evolutionary algorithms and so on. I came very much from this idea of, well, we automated feature engineering to have word vectors, we have neural nets, then we automated architecture engineering by just having one unified architecture. What's the next level of automation? It's the actual ideation, implementation and validation of general ideas in all of AI research. And that's clearly and obviously the next level to unlock a new set of capabilities. And when you think about the automation of science and then you apply the automation of AI research to AI itself, before you know it, you're in this recursive self-improvement loop. And we believe that that will be a great unlock to then apply that kind of intelligence to all kinds of other scientific discoveries.

[1:07:04] And you guys raised a massive round of $650 million, and interestingly, to the compute discussion that we were just having, I read that you committed $410 million, basically most of what you've raised, to a single compute deal with Amazon. So that goes to show the fundamental importance of compute.

[1:07:15] Yeah, we raised, in the end, like $670-ish million, and that will probably be one of the smallest compute deals that will happen in our future.

What Recursive Will Build First

[1:07:35] And so what can we expect from the company? What is it that you guys are going to release first? By when?

[1:07:40] There will be a couple of interesting things coming up. I can guarantee you they will happen this year. And I struggle with this sometimes, the neolab category. I don't love it, because I think a lot of them will not succeed. We are a real company, not an academic lab. We are building real products, we're talking to real customers, and we're very excited about taking this technology and making it useful for real companies. I can't share the details yet of what we're going to release, but I think it'll be exciting, and we already know from some first conversations that it is exciting.

[1:08:09] But is it going to be horizontal or focus on a specific vertical along the lines of what we discussed?

[1:08:16] There's a sequence to it, and some things will be general, but then obviously at some point it'll be more and more specific. We did publish a blog post that gives you a little bit of a glimpse of things we're thinking about, that are essentially milestones towards full recursive self-improvement, that show that, for instance, when a lot of people use AI or do some auto research on small models, our system, the first instantiation of this Eureka Machine in a very narrow domain, can already outperform months and sometimes years of human endeavor on particular problems. We also showed that it can build new CUDA kernels, which is very useful for faster inference, which is very useful for all the large hyperscalers and people who provide tokens and run models. And we're very excited to keep pushing those, and we've heard very positive feedback from folks who are using these kernels now, and, at NVIDIA, folks that created these benchmarks, the SOL-ExecBench as a particular example there. So yeah, those are all just simple examples of artifacts that this Eureka Machine can produce when it comes to AI research, on the path to full RSI.

[1:09:31] And as an aside, I cannot resist asking the question: why are most neolabs not real companies? I mean, they're just ideas of, we want to explore this particular idea.

[1:09:45] And that particular idea is one of the many useful artifacts our Eureka Machine could also produce. But it's not really a product. If you just want to try to think about how humans interact with AI in the future, that's not quite anything very concrete. And so you have to be very careful about what the founding teams are and so on, that have not just done amazing research but also shipped real products.

How Do We Define Intelligence?

[1:10:15] To end, I want to talk about intelligence and superintelligence and where all of this is leading. So the book ends by asking a huge question: how far can intelligence go? And you propose your own definition. So maybe talk to this.

[1:10:27] Yeah, that one will take us more than the time we have left. I feel like it's almost like a new book. I had to wrap it up at that point, the book. And so one, I'm surprised no one has really defined intelligence in all of its complexity really well, neither in terms of the very foundational building blocks, which I currently think are prediction, action and goals, and a combination of those three. Those are sort of the three principal components. Just like energy has one unit, we don't yet know what the unit of intelligence is. Something I'm thinking about a lot right now. We don't yet then have a proper definition, sort of physics-inspired. In physics we have kinetic and potential energy, but then it also makes sense to study chemical energy and mechanical energy and electrical energy and different forms, and some are still pure science fields and others are very much engineering fields. And so I think a similar thing has happened in AI, where we have visual intelligence, language intelligence, physical intelligence, and robotics takes... and I define these 10 different spaces of intelligence.

How Far Can Intelligence Go?

[1:11:37] And each space basically has many different dimensions, and I'll just give you this one example on visual intelligence, right? Humans can only see in a specific part of the electromagnetic frequency spectrum, but you can go much beyond humans. When you think about the bounds of visual intelligence, how far could an AI or any kind of intelligent life form or entity in the universe push visual intelligence? And then you get into very interesting, often physics-inspired thoughts and loops. For example, you can see everything from gamma rays to gravitational waves, so very different, the whole spectrum of electromagnetic frequencies. You can try to have not just two eyes, but you can have millions and billions of different sensors all throughout. But then how far could they go? Oh, well, at some point you have communication bounds, like the speed of light, and each sensor has sort of a speed-of-light cone around what it can see, and you quickly get into these thoughts around bounds. And what you then realize is that, boy, are we far away from the true upper bounds of any of the spaces of intelligence. And there's still so much further that AI can go in research. And so when people think, oh, this set of algorithms, or the field of AI, is sort of a bubble, it's going to burst: I mean, maybe, like energy, the unit cost of intelligence may fluctuate depending on a bunch of factors, but we can still go so much further as a field and as a civilization in pushing that field forward.

[1:13:19] All right, Richard, this has been another fascinating conversation, and I could keep you for another couple of hours, but I know you actually have a couple of companies to run. So thank you for spending time with us. The book, again, is called The Eureka Machine. It comes out on September 22nd. That's right. And where else can people follow your work? On X (@RichardSocher), and recursive.com. That's right, recursive.com. Wonderful. Thank you so much. We appreciate it. Thanks for having me, and wonderful questions. Great chatting with you always.

[1:13:57] Hi, it's Matt Turck again. Thanks for listening to this episode of the MAD Podcast. If you enjoyed it, we'd be very grateful if you would consider subscribing if you haven't already, or leaving a positive review or comment on whichever platform you're watching or listening to this episode from. This really helps us build a podcast and get great guests. Thanks, and see you on the next episode.