DeepMind's Pushmeet Kohli on AI's Scientific Revolution

0:00So I went to a biology conference and after I gave my talk, a biologist approached me and he said, push me, I have been working on this protein for the last 10 years. And I had collected so much lab data to characterize this protein, to figure out its structure, but somehow this has eluded, like this has evaded all kind of investigation and we still didn't know this structure. But we had all this data. If we knew the structure we could sort of validate it very quickly. I ran Alpha Fold 2. It gave me the structure. It perfectly fit the answer. I've been working on this for 10 years. Wow. What do I do next?

1:04What happens when AI stops just answering questions and starts asking them? In this episode, Pushmet Cully discusses DeepMind's alpha evolve, a breakthrough evolutionary AI system that discovers entirely new algorithms. Pushmet reveals how coupling language models with evaluators creates something unprecedented, AI that can tackle decades -old math problems and generate human -interpretable code that outperforms expert design solutions. PushMeat shares stunning examples of AI uncovering hidden mathematical truths and explains why we're witnessing the emergence of a new scientific method. One where AI doesn't just accelerate discovery, but transforms which problems we can even attempt to solve.

1:44Enjoy the show. PushMeat, thank you so much for joining us today. We've all been eagerly waiting for the moment that AI is capable of making novel scientific discoveries. Do you think that alpha evolve is that watershed moment. Yeah, so it's certainly a key sort of milestone. What we have shown is that you have AI sort of model, a large language model when coupled with a harness is able to discover new algorithms and not only that, it's basically able to view new mathematical results which have been studied for many, many years. You use the words when coupled with a harness. Can you tell us more about that harness?

2:28Yeah, so if you now go back to AI models, like the history of AI for science is very long. We have a number of different models that have tried to do a scientific discovery, like one of the key models in this category is AlphaFold, which is the prototypical example of what can be achieved by AI in science. We released Alpha412 at the end of 2021 and it won the Nobel Prize last year. So the impact of AI in science is very, very distraught. Now the question is whether LLMs and foundational models, how can they impact science? In around two years back, we had an agent called Fonserch, in which we took an LLM and we coupled it with an evaluator.

3:28And the evaluator allowed the LLM to figure out when it was making new conjectures on sort of coming up with new ideas to solve problems, whether they were hallucinations or whether they were brilliant insights. So essentially in this particular case hallucinations were great because those, some of those hallucinations were in fact brilliant new insights that nobody had thought about. So which, this is where the harness comes in that you have this evaluated evaluation function and essentially a search protocol associated with the LLM that together is able to come up with completely new discoveries that are really impactful.

4:13You mentioned fund search. Could you say a word on the difference between the results you all accomplished with fund search versus alpha evolve? Yeah, so fund search was our first instantiation of taking a large agnodal and trying to see if it can discover new algorithms. The models at that time were weaker, right? And the, and the type of search that we were trying to do, we were, we had not sort of explored things much further. So what we asked the LLM to do was essentially try to complete a small function and see if it can do that much better. And surprisingly, it was able to discover completely new algorithms that mathematician had been trying to study for a long time.

5:06But the limitation was that the mathematician or the researcher had to give a template on in which the algorithm should be found. With alpha evolve, we have removed that restriction. Alpha evolve is not just searching for a few lines, it's basically looking at whole algorithms themselves, right? Very, very large pieces of code and optimizing them over a long period of time. And secondly, fund search, original model used a lot of function evaluations to discover to make these new discoveries. alpha evolve can work with many fewer function calls and it can basically by looking at fewer proposals that can discover new algorithms much more quickly.

5:59Can you tell us about the role that the evolving Gemini models play in the capabilities of the alpha evolve? And I think I saw in your blog posts, you have both Gemini Flash and Pro involved in the harness. What is each responsible for? Yeah, so I think, see, we have been evaluating, or as Gemina improves with various sort of generation, it is becoming much, much better at its understanding of code. Now, if you have a proposal generator, which can understand code much more effectively than it gives it generates proposals which are not only syntactically correct. They are also semantically trying to solve the task and then you are sampling what are the different ways in which the task can be solved.

6:50So as the baseline model Gemini's abilities to perform coding improve are sample effectiveness in searching for the right solution on these very hard mats and computational problem becomes much better. So if you want to search in a large space, there are two sort of elements. One is the speed of how you can generate these proposals. And then the speed at which you can evaluate those proposals. So first, how quickly can you say, can we give me a new kind of candidate algorithm? And then secondly, how quickly can you evaluate whether the algorithm is any good or not? And both things are really important.

7:40and the fact that you have these variants of Gemini flash, which can do that very efficiently and very quickly, this is really important. I know AlphaVolve is more of a broad domain model than some of its predecessors. How broad is it? What's in scope? What's out of scope? Yeah, AlphaVolve essentially allows you to search not only in terms of the size, of what you can search over, right? You can now discover whole new algorithms, but it also is extremely general in its ability of thinking about algorithms in various different languages. So not only can it sort of search in C++, but it can also do it in Python.

8:28It can also sort of do in very log, which is what the language is for defining describing chips, right, in chip design. So, the generality of alpha evolve is in its ability to search for these large algorithmic spaces, but also in different syntactic and semantic representations, right? It is not restricted to a particular language like Python, but it can sort of do that search across many different types of languages. And many different types of tasks. the only expectation it has is that you have a function evaluator that you can quickly evaluate whatever proposal there is and say how good it is.

9:15It seems like the rough cognitive architecture so to speak of generating a bunch of algorithm candidates evaluating them and then I guess evolutionarily deciding which ones to keep and then going forward from there. And this, it seems like it roughly mirrors the scientific method? Is that intentional? Yeah, so I think there is, there are also, like if you think about it, there is another sort of agent that we released earlier this year, which was called co -scientist. And in co -scientist, essentially what you had was Gemini playing the role of the whole scientific academic process. So Gemini playing the role of a hypothesis generator.

9:59Gemini playing the role of the critique. Gemini playing the role of sort of ranking, different sort of of reviewing those ideas and ranking those ideas and then editing those ideas. So it was Gemini playing all these roles in a multi -agent setup. And these were all sort of Gemini models prompted differently to play different roles and very sort of interestingly, this combined multi -agent system came up with behavior that went much beyond a single Gemini sort of models answer. So it was able to give much much better proposals and new ideas compared to a single model. What's the intuition behind why that works?

10:52Yeah, so I think it is something that is still being studied. But it is a fascinating sort of thing. The one thing that I actually sort of noticed is that especially with regards to a sort of co -scientist, you would run sort of co -scientist on a particular problem. And the very first answer that you might get might not be very different from the baseline in Gemini sort of model. And but what happens is even as you sort of increase the amount of computation over sort of this is you're not talking about just a few minutes or a few hours, but even days As the whole multi -agent system sort of looks at the solutions and then refines them and sort of tries to sort of rank them, it just becomes much much better.

11:51So why that might be happening? It might be that the proposal, there is deep inside. So there is some sort of intuitions that are buried in the tail of the distribution. And then somehow, Gemini's ability to evaluate which sort of proposal, which idea is better is much better than its capability to come up with a new idea. It's the same sort of thing in computer science. Sometimes we are able to find, if sometimes we know whether a particular solution is correct or not, but it's very difficult to come up with a solution. Right? So it's the same sort of thing appearing again in this multi agents setup that somehow the agents working together are able to extract many more impactful results.

12:47It seems like the architecture of kind of you know generators and verifiers. It seems like that paradigm is being echoed across the broad AI space whether it's you know very general models or very specific kind of like AI systems for very specific applications. Is that fair that that's sort of the consensus architecture right now and do think that'll be the thing that people continue to push and scale? Yeah, so I think there is going to be more work in agents, right? What we are seeing is basically the very start of research on agents. Whether in alpha evolve, you had a generator coupled with an evaluator.

13:32The generator was a neural network, a foundational model, an LLM, and the evaluator was even hand coded. But together with an evolutionary sort of search scheme, you were able to sort of get these much more effective results. In whole scientist, you didn't have just one agent. You had multiple agents working in a shared memory. Now, like, what is the optimal agent configuration? Like, this is still an open research problem. It's super interesting. Are the results that you're getting? Are they different from the ways that humans would derive them? And I'm kind of thinking of the alpha -go move 37 stuff.

14:17are the methods different or the results? How do they compare the ways humans would think about them? So let's go back to the original motivation for why we started working on the first iteration of using LLN's Valgris McDiscovery, which was fun search. So a few years back, like as you know, DeepMind has spent a lot of, has done a lot of work in using AI systems for searching over large spaces. We have done a lot of work in building agents which have been trained using reinforcement learning which can deal with many complex challenges from the game of go to playing star craft, which are quite complex challenges.

15:09We set ourselves a challenge that can we take the same kinds of models, like the alpha 0 family of models, which were extensions of what we had done in go and the development of alpha go, can we use the same types of models for discovering new algorithms. And we came up with a new sort of agent called alpha tensor, which was particularly sort of focused on finding solutions for the matrix multiplication problem. And we found that this agent was able to improve over the past known results which had stood for 50 years. But the key question sort of remained, can you do something better? And secondly, can you sort of come up with a solution that is more interpretable?

16:11At the same time, like when we were looking at practical problems in Google, like how do you schedule jobs in a data center? Now, there has been a lot of work on coming up with new algorithms and these heuristics have been designed by some of the best researchers and engineers at Google. Because they have a huge amount of impact in terms of computer utilization. If you use a typical reinforcement learning agent on this kind of problem, you might get better results, but it might come at the cost of interpretability. because now you have a neural network deciding which workloads go to which computers.

16:56And if something breaks, then you don't know how do you debug this thing? So what engineers would really prefer is instead of giving them a neural network, you gave them a piece of code that they can interpret and they can run. And this was essentially the motivation can we now use LLIMS instead of searching in the space of specific algorithms like we had done metric multiplication algorithms like we had done in alpha tensor or coming up with a neural network policy to directly solve the problem. Can we come up with an agent which can search in the space of programs and come up with a program that solves this hard problem?

17:40And the benefit of course will be into probability that you can see the sort of code, you can see what its properties are and so on. And that's what is what happened. We found sort of programs that not only were effective, but when the experts actually saw those programs, they could sort of recover insights. So for instance, one of the math problems that we had looked at for fund search was called the cap set problem. This is a problem that Ted and and style, one of the famous mathematicians, he is very interested in. And we collaborated with this mathematician, Jordan Allenberg at NYU, and when we looked at the program that Fonserch had produced, he found that they were certain symmetries that were in the problem that had not been recognized before.

18:34And somehow the program, like one search, the agent had discovered those and was utilizing those to get a better solution. Can you say where the, you mentioned working with Tarynstown and other famous mathematicians. Is math considered the gold standard for, you know, testing and benchmarking if these models are generating novel scientific results? Yeah, so Math certainly sort of has some properties which are very interesting, right? The fact that it's very precise, like you can, you know whether the property that you have looking for, whether you have found it or not, right? You know the matrix multiplication, it's a formatic multiplication and how many multiplications you require, like for a 4x4 matrix, what was known was that you can do it with 49 sort of multiplications, that's by stressor, and we showed that you can do it by 48.

19:36So that's a very precise result, right? There is no sort of arguing about that. So it gives you a very crisp way of evaluating how well you have done. And there is no sort of like RLHF needed in terms of human feedback whether this was an ice result or whether this was a nice output or not. And you don't need to rely on an LMSS scores, you just know that you're better. Yeah. Okay, so then when you go from the beautiful pristine environment that is net to the real world. It seems like you all have found a real world applications and data centers in the very log world. Could you say a little bit about which applications you expect alpha evolved to be most impactful for?

20:29Yeah. So wherever you can find a good function evaluator, wherever you can find an evaluator, where you can say, I really trust this evaluation scheme. If you give me a program, I can tell you very concretely how good it is. If your problem satisfies that setup, then you can use alpha evolve. Because alpha, unlike a human sort of programmer who can try 10 things or 100 things or 1000 things, alpha of all does not, it does not sort of, it can go on and on and on and on and on. It can come up with very counterintuitive strategies to find to solve that problem. Some things that you might not have ever imagined.

21:20Can you have humans be the function evaluators? Or does that not work? Humans can be the function evaluators. It's a question of sort of scale. Like how many can you sort of evaluate? and whether you can evaluate the property of the program effectively. So at scale and with the right level of accuracy. How do you do that? Do you build that into the application itself so that there's a human in the loop evaluating as it goes? Do you do that offline separately before the application is produced? I guess how do you do that or how do you imagine people doing that? Yeah, so I mean like so we haven't used a human in the loop for alpha evolve right most of our evaluators were programmatic evaluators right but imagine a hypothetical scenario where alpha evolve was told that you have to solve this math problem and come up with a new algorithm to solve this problem and suppose it came up with many different types of problems many different kinds of solutions which are all equivalent in performance.

22:30Okay, but then which one is the best? It's the best is the one which is not only sort of very effective on the problem, but is the most elegant according to a mathematician or the most simple to understand. Right, and that's a very subjective human thing. Like simplicity or interpretability, like we don't have a sort of crisp definition of it. it depends on, it is grounded in the human observer. At what point do you need a pair kind of what's happening in the digital world to any kind of physical world stuff? I think in your blog post, you mentioned that you could see also evolved being useful for, for example, for material science.

23:15Do you need to be able to connect to a real world laboratory to kind of get any of that feedback or do you think all of this can kind of happen in the algorithmic domain? Yeah, that's a very good question. And I think this goes back to how much do you trust the evaluator? If you, if sort of, if your evaluation was based on a computational sort of method, and the computational method was perfect and you completely trusted it, then you don't have to. Then you think, well, I believe the computational model, the computational model says that the solution that AlphaEvol came up with is satisfies these properties.

23:56Job is done, right? But if you don't believe that the computational model is the perfect characterization of reality, then you want to make sure that you sort of validate that result in the real world, right? And you see whether that assessment of the evaluator was indeed correct. as AlphaEvolve becomes more and more successful, as Gemini becomes more and more powerful. What do you think happens to these domains, and how will the human scientists and engineers working in them adapt? So for example, if you take chip design, as an example, you mentioned these models are getting very good at creating new, generating very log creating new chip designs.

24:42Does that mean the role of a chip designer goes away, changes, how do you think this changes the world? Yeah, so I think that's that's that's again sort of a very interesting Question I give you the example of what happened With alpha fold so we started working on this problem of protein structure prediction So if for those of you who don't know like proteins are the building blocks of life. They are the Lego blocks of of life and for many many decades scientists have been trying to figure out what is the shape of proteins. Because if we understand the shape of proteins, we understand how they function and we can use that to sort of develop new drugs to treating sort of the most sort of challenging diseases on the planet.

25:33We can let metal, wetter, sort of enzymes and so on. Now in 2021, as I sort of mentioned, we release a football too. Before that, you used to take or even single protein sometimes 1 to 5 years to find the structure of a single protein and it might take a million dollars. And there were some proteins which were so notoriously hard that people had been trying to study them for almost one or two decades, it had not found the solution. and which is why only 37 % roughly 37 % of the human proteins their structure was known. So after we sort of released Alpha4 2, I went to a biology conference and because Alpha4 2, with Alpha4 2 we could find the structure of all proteins, not just human proteins, all proteins on the planet and we made the structures available to everyone on the planet.

26:28So, I went to a biology conference and after I gave my talk, a biologist approached me and he said, push me, I have been working on this protein for the last 10 years and I had collected so much lab data to characterize this protein, to figure out its structure, but somehow Now this has eluded, like this has evaded all kind of investigation and we still didn't know the structure. But we had all this data. If we knew the structure, we could sort of validate it very quickly. I ran alpha 4 .2. It gave me the structure. It perfectly fit the answer. I've been working on this for 10 years. What do we next?

27:19Next. So what has happened after I was a full two? What happened is basically suddenly it did three things. It first advanced structural biology. What was not possible earlier? It would take a synchrotron and six months and a million dollars is now done in a second, right? So it really advanced what was possible. Secondly, it accelerated it and thirdly it democratized it. Like that particular scientist working in Latin America or South Asia or Africa on some neglected tropical disease had no chance to sort of figure out the structure of their protein. They did not have the funds or have access to instruments that could find them the structure.

28:09Now they have access to those things to like the any sort of parasite that they're working on. So what do they do? They are now working in this new model where structures of proteins are not hard to get, they are everywhere. And so they are working on the next set of things. Like how do you now use that knowledge to treat diseases and design better drugs? And I think the same thing will happen with the with alpha wall when you have these agents which can go beyond Human abilities in solving these problems. Then the question becomes which problems to be solved What are the important characteristics of a chip that we need to improve on Right, like we want to make it much more efficient much more sort of So that it requires less cooling, it requires sort of less expensive construction mechanism.

29:06It's more fall -tolerant. Many other things you can make the problem more and more sophisticated because now you have more sophisticated systems to sort of optimize them. Well, I have you something I've always wondered the alpha fold results are phenomenal and this the story you share with us is really impactful. Do you think that it's causing inflection points in the kind of availability of new drugs? Or have there been other bottlenecks now that are just, we're faster at one part, but unfortunately everything else is just hard, so we're still slow overall? No, it has been a thing that I think there's one has to understand that drug discovery is a long process.

29:54Now, what are the what are the root blocks for drug discovery? First, you have to understand a target. You have to understand here's a protein in the body that I need to bind because this protein is somehow involved in the disease. So, if I can somehow bind something to this protein and change its function, it will have an effect that can sort of treat the disease. Like, first you have to come up with that conductor. Then you have to say, okay, now I have a target protein. How do I develop a drug? How do I develop a small molecule or another protein that binds to it? So for that, you needed to understand the structure of the protein, which other proteins that it interacted with?

30:36How did it interact with this molecule? This would take a significant amount of time, sometimes two years. Now that process is dramatically sort of accelerated. Now you can do it in sort of a few weeks or a month, a few months that took you multiple years sometimes. But that's not the end of the story. After that, you need to now clinically validate it. So you have to go through phase one trials, phase two trials, phase three trials, you have to think about toxicity, all these other sort of things. So what Alpha4 did was take one blocker away, made the overall timeline faster, but there are other sort of blockers which are new generation of AI for biology models are hoping to accelerate and much make much faster.

31:23So we have taken a big step, but we need to take a few more big steps. What do you think will be most lucrative for this family of models? I think the question is, what is, I mean, the answer to your question is basically what domains do you think are important for society? Because AI is going to accelerate everything. It's going to accelerate healthcare. it's going to accelerate sort of the ability for us to develop more smart systems from healthcare to sort of material science. Like if you think about our the history of civilization, we even describe our civilization in the sense of there was first we were sort of cave dwellers and then we went into the stone age and then we sort of when to the iron age and then the bronze age and now depending on who you talk to you're either in the silicon age or in the plastic age whether you're a stick or feeling a bit sort of sad.

32:31But if you take a step back and you think about what has humanity achieved, what we have achieved compared to any other species is the ability to transform energy, to leverage energy, right? We have been able to leverage energy and do big things with that power. Now, if you can come up with say a new room temperature superconductor that completely transforms your ability to handle energy, right? What changes will it bring about in society? They're hard to predict. If you can, if you can deal with energy in that way, If we can unlock fusion, and energy becomes so cheap, every like if you think about geopolitics, if you think about economy, a lot of it is about energy.

33:22And suddenly if energy sort of goes down to zero, what will be the impact on economics of the whole thing? Similarly, like if you think about coding and if you have the reagents which can code, what does that mean? If everyone can sort of code, intelligence sort of is completely ubiquitous. Everyone has access to all these different things. So they will be dramatic changes and everything will be impacted. So from materials to energy to sort of coding to healthcare. Really cool. Do you think we're going to have a fast takeoff moment for scientific discoveries? Do you think we're at the ramp of one?