[0:00:01] Get your quote today at Progressive.com. Progressive Casualty Insurance Company affiliates and other insurers not available in all states. When it's time to scale your business, it's time for Shopify.
[0:00:31] Get everything you need to grow the way you want. Like all the way. Stack more sales with the best converting checkout on the planet. Track your cha-chings from every channel right in one spot. And turn real-time reporting into big-time opportunities. Take your business to a whole new level. Switch to Shopify. Start your free trial today. So I suppose we might start by looking out at the audience and looking at the variety of people.
[0:01:04] And thinking that when they come to this room, they're not perhaps all thinking of AI in the same way. What would you say to that? Yeah, that's a very interesting way to start the conversation because we produced this report, I have a copy here, a couple of years ago. And at that time in particular, if you...
[0:01:28] Even in two years, we've come a remarkably long way in terms of people understanding what AI is. And I think if I asked everybody in the audience, AI means different things to them. Back in two years ago, when I was asked how do I need to understand about AI, lots of people would say, how do I learn about CNN and understand how it might work? Now, you do not get asked that question.
[0:01:55] So we have this notion, including in science, of although the technology moves on and we have different capabilities, people's personal views of AI has dramatically evolved, maybe even faster than the AI has in some sense in the last couple of years.
[0:02:25] Going to the report, if I might do that, we set out, the report was published to the month two years ago. And that meant about over the 18 months before that, we decided to do a piece of work on science and AI. If I may just frame that, you were chair of the working group. I was, yes.
[0:02:47] Can I just ask you what the purpose of the report was? What were you setting out to do? So the purpose of the report came about from our science policy work, deciding that we wanted to do a piece of work to understand how an area of disruptive technology was affecting scientists. And one thing I said was, do not come back and say AI.
[0:03:09] But of course, what happened when you go out and talk to people, the most disruptive thing that either then or they could see coming was to think about how AI would affect particularly the scientific process and methodology, as well as really what a scientist does. And so we decided that that would be a really interesting thing to look at, because although we hear about capabilities coming through, we hear...
[0:03:35] Less about how disruptive this is. And I think what you would find if you're in the report is that that's very much the type of thing that we were focusing on and also people are concerned about today. And how did you do the work? You had a working group, but then you solicited a far wider...
[0:03:55] Yes, so it was talking to fellows, so some in the room. There were over 100 fellows had input through attending either workshops. We had workshops on things like access to data. At the time, in the UK, it was...
[0:04:15] Hosting the first safety, AI safety meeting. So we had workshops on safety, large language models. And it was about trying to get a sense from these interactions to understand some of the issues to frame the piece of work. And it's a big volume with lots of findings. But do you want to attempt to highlight the key topics that you thought were important?
[0:04:44] The key things that I think are also very relevant today, thinking about access to compute, some of that certainly in the UK has advanced, and the infrastructure that's needed to be able to work with any type of AI methodology. Access to data, I think that's one that's very much on people's mind. And access for your use is also an international issue.
[0:05:12] And reproducibility, this of course is not a new concept, but the challenges you have in being able to reproduce work and be responsible in the use of AI tools. And finally, how this is affecting skills and what people do was another aspect we were pulling out.
[0:05:35] Often requires many different people to take on different roles. That means you're moving from maybe siloed areas where an individual might work in their own office if they were computational. But now you're going to be working to do team science. Team science is very natural for certain areas of science. In other areas, it's a brand new thing. Life is all about making decisions. Sometimes you make questionable ones.
[0:06:02] Like buying a new TV that's too big for the wall because the electronics store employee was a little too good at their job. Then sometimes you make great ones, like protecting your home and auto with Progressive. You get customized coverage options so you can be sure you have the right coverage. Get your quote today at Progressive.com. Progressive Casualty Insurance Company affiliates and other insurers not available in all states. When it's time to scale your business, it's time for Shopify.
[0:06:30] Get everything you need to grow the way you want. Like all the way. Stack more sales with the best converting checkout on the planet. Track your cha-chings from every channel right in one spot. And turn real-time reporting into big-time opportunities. Take your business to a whole new level. Switch to Shopify. Start your free trial today. The holidays are coming, but the Lincoln Wishlist sales event is already here.
[0:07:01] You might just cross the perfect gift off your own list. You know, the kind that doesn't need wrapping. Consider this a little reminder to be good to yourself this time of year. Visit your local Lincoln dealer or go to lincoln.com to learn more.
[0:07:25] And then it brings in things like should you partner or should you compete? And scientists love to be competitive. So how do we manage some of these big cultural changes? So these are all the things that bubbled to the top when we were thinking about the report. There's an awful lot in there, and it's a beautiful framing for much of what we'll talk about later in the discussion. At the beginning of the report, you asked the question whether AI is assistant, peer, or tutor.
[0:07:52] And that, again, is something we will come to. But you came up with recommendations as well. And those were at a slightly different angle to the topics that you identified as most important to think about. Do you want to talk about those at all? Well, I think that some of the recommendations were what I've said, like access to data, being able to come up with standards for data.
[0:08:21] A lot of this is discipline specific. You can't have umbrella rules. And also, I think it's also reflecting that areas that are new to using AI methods should be learning from those that are more advanced. The ones that are more advanced, we're going to hear about one later, where you set things like data challenges, you've invested in having high quality data. Those were ready to take on machine learning as being a tool that...
[0:08:51] You could discover new findings. As somebody who works deeply in this area already, you work on ultrasound imaging and you work on what you can get from multimodal data sets. So you're well into it. Do you find that people are keen to learn from your experiences or do you find that there are disciplinary boundaries to people's willingness to learn?
[0:09:17] Well, I think in my own area, because I work in healthcare AI, as you said, I've done a lot of work in multimodal and ultrasound. What's been really interesting in terms of collaboration, if this is what you're referring to, is that when, about a decade ago, it was very much computation. You were working, asking for data, and you were developing computational methods. It then went to co-design.
[0:09:40] And co-creation of solutions with the end user. In this case, it would be a sonographer who's doing the scanning. And now we've moved on to...
[0:09:51] As you deploy systems, you bring in experimental psychology, really to understand, do people trust what you've used? How do they, things like, do they become reliant on the tools? And again, these are things that I think in certain areas are emerging, and we're going to hear a lot more about that in the future too. The report is a production of a UK-based organisation.
[0:10:19] But you are also Foreign Secretary of the Royal Society, so you travel a lot and have these discussions around the world. Do you find that, it's a huge question, but do you find that attitudes to AI are broadly the same around the world? No, they're not. I think it's been very interesting to have my role at this time when everybody wants to talk about science and AI. Why? Because it's affecting people in different ways. I think there are some...
[0:10:48] Common things that we all would like to be able to understand how science evolves. They are things like, I've mentioned them, reproducibility, the effects of AI on publishing.
[0:11:02] So that's peer review, writing papers. How will the publishing of science evolve? And the Royal Society had a meeting. It wasn't dedicated to the AI aspects, but we had a very interesting conference last year on the future of publishing. And then there are things like, not everything translates to different parts of the world. In Europe, very much, and if we count ourselves as being associated with that, it's very much responsible AI.
[0:11:31] So they are in science, that is underpinning everything. But if you go to, for example, in India or...
[0:11:43] Or China, they are much more, they're very excited by the potential more and you hear a lot less about the concerns. But they're very also keen to talk about some of the ethical issues, the responsible issues. I was at the AI summit in India and there things like the fact.
[0:12:04] The large language models are built on Western world data is a big concern because it doesn't translate well into their own languages. So there are many things that you can start to have conversations about when you're on the, to do with internationally. I mean, whether this evening the conversation will get as far. Life is all about making decisions. Sometimes you make questionable ones.
[0:12:30] Like buying a new TV that's too big for the wall because the electronics store employee was a little too good at their job. Then sometimes you make great ones, like protecting your home and auto with Progressive. You get customized coverage options so you can be sure you have the right coverage. Get your quote today at Progressive.com. Progressive Casualty Insurance Company affiliates and other insurers not available in all states. When it's time to scale your business, it's time for Shopify.
[0:12:59] Get everything you need to grow the way you want. Like all the way. Stack more sales with the best converting checkout on the planet. Track your cha-chings from every channel right in one spot. And turn real-time reporting into big-time opportunities. Take your business to a whole new level. Switch to Shopify. Start your free trial today. The holidays are coming, but the Lincoln Wishlist sales event is already here.
[0:13:30] And with exceptional offers on select Lincoln vehicles, you might just cross the perfect gift off your own list. You know, the kind that doesn't need wrapping. Consider this a little reminder to be good to yourself this time of year. Visit your local Lincoln dealer or go to lincoln.com to learn more. Far as international agreements and regulation, I don't know. But it seems to me of fundamental importance that you are having these conversations at every level.
[0:14:04] To understand each other's perspectives, just setting the groundwork for any sort of future agreements. I think that we actually find that as national academies, that is often a good place to have some of these conversations because we're also talking about the science that should cross geographical boundaries. And there are things like, the ones that the real barrier are things like access to data and being able to...
[0:14:33] Share information. But I think that's the conversations for scientists in areas that affect global science are the ones that we're having a lot of conversations about at the current time. Thank you very much. And then just a couple more things and then we bring up our other panelists. One is that this was, of course, published in 2024, so a lot has happened. And work continues at the Royal Society. So the next iteration is another report or how does it go on?
[0:15:01] I don't know if policy is in the room. I don't believe the plan at the moment is to have another report, not of this size, but certainly looking at how machine learning has evolved. For example, we do not talk about agentic AI, which is, we mention it, but two years ago, it was not so prolific. And also, there are things like,
[0:15:31] Thinking about reasoning and uncertainty, which are things that machine learning models within the context of science. We need to be able to understand uncertainty. That might also be something that we look at. But I think it's watch this space for the future pieces of work. And then this is, I'm sorry for this question because it's a bit difficult, I think. I mean, has it had the impact you hoped it would have? Yes, I do.
[0:16:00] In the sense, the purpose of a report is to start the next conversations and to provide a point of engagement, to have a variety of different ways that you can interact, both nationally and internationally. So from that perspective, it is a report that's had big impact. Well, thank you. It must have been an enormous amount of work. And certainly, listening to you now has set the scene beautifully for this evening's discussion. You've covered almost every topic that was raised in the questions raised by the audience.
[0:16:29] So I could just leave it to you now, but I'm going to invite others. So could I invite Paul Nurse and Demis Osiris to join me on stage, please? Thank you both very much indeed. I don't think great introductions are necessary. I mean, you were introduced in the starting comments. But Demis, if we could start with you. So you are CEO and co-founder of Google DeepMind. You're also CEO and co-founder of Isomorphic Labs.
[0:17:09] And you have used this phrase that what we're witnessing is science at digital speed. So perhaps just it would be nice to again set the scene to bring us up to speed very quickly with what we've seen so far, what AI has done to science so far.
[0:17:28] Yeah, that phrase I use, science at digital speed, I sort of coined it after seeing AlphaFold in 2020, 2021. We spent about a year folding all 200 million proteins sort of known to science. And that's when it struck me that we were using the sorts of techniques and methods that were used in technology areas and engineering.
[0:17:54] But now this was a scientific subject that we were applying it to. And I kind of mean it in three different ways. Two originally, and I think one that's coming in now. One is the speed of the solution itself. So AlphaFold, just using that as an example, but obviously I'm meaning this more generally, not only was it very accurate, accurate enough for biologists for it to be useful for experimentalists.
[0:18:19] Obviously not perfect, but accurate enough. It was also able to fold an average protein in a few seconds. So it was very fast. The second way I meant science at digital speed was the dissemination.
[0:18:30] So if you invent a new method, let's say even like something really famous like CRISPR or a sort of experimental method, it takes a while for that to percolate through even the biggest discoveries into the wet labs, into the scientific process and then become a kind of standard tool maybe a PhD student would use. Maybe it's a 10 plus year kind of process. Are you ever shopping online and suddenly a discount that helps you save pops up? It makes your day.
[0:19:00] Sometimes even you're weak. Want that feeling month after month? Well, new customers who save by switching their home and auto insurance to Progressive save hundreds. Cue your happy dance. Get your quote today at Progressive.com. Progressive Casualty Insurance Company affiliates and other insurers not available in all states. When it's time to scale your business, it's time for Shopify. Get everything you need to grow the way you want. Like, all the way.
[0:19:32] Stack more sales with the best converting checkout on the planet. Track your cha-chings from every channel, right in one spot. And turn real-time reporting into big-time opportunities. Take your business to a whole new level. Switch to Shopify. Start your free trial today. The holidays are coming, but the Lincoln Wishlist sales event is already here. And with exceptional offers on select Lincoln vehicles, you might just cross the perfect gift off your own list.
[0:20:03] You know, the kind that doesn't need wrapping. Consider this a little reminder to be good to yourself this time of year. Visit your local Lincoln dealer or go to lincoln.com to learn more. But with AlphaFold, once we folded these structures, we were able to obviously work in collaboration with the European Bioinformatics Institute, part of EMBL, amazing collaboration we did, and just put it on a database very quickly in a couple of months. And then it's just as simple as a keyword search.
[0:20:38] And there it is. And now over 3 million researchers around the world have used AlphaFold and the structures from 190 countries. So it doesn't matter where you are, of course. You don't even have to have a wet lab. You can access these structures and then build on top of it. So that was pretty fascinating. That's more like a piece of technology, like an application, something you would be more used to in big tech. But here, again, appearing in science.
[0:21:01] And then the final way, which I think is becoming more apparent now, is obviously AI itself accelerating scientific discovery itself.
[0:21:11] And the pace of that. And that's what I think we're in the beginnings of feeling now. I was quoting this phrase last week at the big Google annual event of we're in the foothills of the singularity, which caused a bit of a stir. But I really mean that. I can feel it. If we look back in 10 years, I think we'll start feeling this period now, around now, next year, this agentic era that Alison was mentioning, which is really actually coming to the fore now and actually really working for the first time. We predicted it for a while, but it's really working now. That's going to be the beginnings of it.
[0:21:40] And when we look back, and I think it's going to be monumental, that change. You know, I think it's the whole reason I worked on AI my whole life is it's my expression of what I can contribute to science was a kind of meta contribution of building the ultimate tool to help scientific discovery and science and medicine. And I think AI and AGI is that. And I think we're starting to see the beginnings of that now and more and more people are able to see what we saw, you know, 20 years ago.
[0:22:08] That's an awful lot of what you just said. The choice of the word singularity, foothills of the singularity, is an interesting one. It is. I mean, because the reason I use that phrase is it's not just, it's the wider milieu around the technology. So we call the technology AGI, Artificial General Intelligence. In fact, it was coined by my co-founder, Shane Legg.
[0:22:29] But obviously, that's just the technology. What's going to happen to society, including science and economics and everything, I think is going to affect everything, clearly. And I think more and more people agree with that. So it refers to that era that we're about to start to live through now. I want others to jump in, but may I just... When you talk about...
[0:22:52] AI actually contributed to the discovery process. For instance, there were two papers published in Nature last week, one of them from Google DeepMind, where AI has conducted experiments and made hypotheses. Does the pace that you mentioned of scientific discoveries filtering out into...
[0:23:11] The inter-general use and general acceptance is one of the strengths, in a way, of science. It takes time to assess things. You need time to find out whether something is real and works. Do you feel that...
[0:23:26] People have that time anymore, or are things just moving so fast that it becomes a little difficult? I'd like us to have more time so that we could apply as a field, this I'm talking about AI now, the scientific method more rigorously, be better if we understood the systems we're building, they're not just black boxes, but understood the deep details of how they work. People are working on that, but it's lagging behind the pace of the actual progress of the performance.
[0:23:50] And obviously now it's become a commercial technology with chatbots and LLMs.
[0:23:56] That's just added fuel to the fire in terms of the pace and progress of the engineering ahead of the science. So I'd love to see the science catch up with the engineering. I'm both a scientist and an engineer is how I'd consider myself. But that's just the nature of the market forces that are currently pushing that. And it's amazing for producing progress. But especially when we apply it to scientific areas, it would be better if we could be more considered and perhaps more thoughtful about how it gets deployed. But having said that.
[0:24:26] You know, we don't have a lot of time before I think AGI arrives. I think we're only a few years out now. And society's had, I think, had a lot of time to sort of advance warning of this in a sense. I think it's time for all of us to take this more seriously, not just those people building the technologies. And I mean economists, social scientists, but also scientists of the society. AlphaFold was finished in 2020.
[0:24:52] So it's actually like ancient era of AI, if you think about it. It's 10 years since AlphaGo. I just came back from Korea to meet the president and other things, celebrating the 10-year anniversary of the match, which really was the, if you look back on it now, was sort of the marking of the beginning of the modern era of AI. And so we've had time. So it's not like a surprise. Those are people who have been paying attention. We've had time to, I don't know if we wisely spent that five, six years as a society, but there's been a sort of warning shot.
[0:25:20] Progressive's accident response can contact emergency services if you're in a major accident. The only thing worse than someone you love being in an accident is not being able to help them. So Progressive is here to help when you can't. So get peace of mind knowing your loved ones are protected with Progressive's accident response. Learn more at Progressive.com. Progressive Casualty Insurance Company and Affiliates. Limited availability in California. Driver is responsible for emergency and towing charges not covered by the policy. Coverage subject to policy terms and conditions.
[0:25:50] When it's time to scale your business, it's time for Shopify. Get everything you need to grow the way you want, like all the way. Stack more sales with the best converting checkout on the planet. Track your cha-chings from every channel right in one spot and turn real-time reporting into big-time opportunities. Take your business to a whole new level. Switch to Shopify. Start your free trial today. The holidays are coming.
[0:26:22] But the Lincoln wishlist sales event is already here. And with exceptional offers on select Lincoln vehicles, you might just cross the perfect gift off your own list. You know, the kind that doesn't need wrapping. Consider this a little reminder to be good to yourself this time of year. Visit your local Lincoln dealer or go to Lincoln.com to learn more.
[0:26:51] It was an anomaly or it's an outlier. And it was. I mean, it's a long time ahead of its time, given that nothing else of that significance happened in the following five years. But these general systems are getting really good now. And we should take that very seriously. Thank you. Alison, you wanted to jump in.
[0:27:08] Alison, please. All right. What I was going to say is one of the challenges of trying to, everyone wanting to be first is one of our problems. We do not reward people for slowing down, particularly in academia. You will not get a paper published, particularly if you think about if you're not first. And I think that for...
[0:27:30] We do need to think about this as well, about reward systems that will reward people who do go in and explore and understand these methodologies and that that is important for going forward. And we will learn from that as well. That was the point I wanted to make. Thank you. Nice point. Thank you. You two, you're both fellows of the Royal Society, you're both Nobel laureates. Seems utterly right that you're sitting here together, but you also have a long history together. You've known each other a long time.
[0:27:55] We do. We even sort of dress the same. I was thinking as we came up, you know, he was a chess master, and what we got here is the black piece and the white piece. But anyway, that's...
[0:28:09] Yes, maybe Demis can remind me exactly, but I think we first spoke when you were probably just still an undergraduate at Cambridge or just after. I think we talked about a PhD, but you wanted to set up a gaming company. Yeah, which he disapproved of, I think. Well, I thought I can't handle that. I can't handle that. So this is nearly 30 years or maybe over 30 years. Yes, and Paul's basically, I consider him to be my mentor in biology.
[0:28:39] We've talked about things like virtual sales. I'm sure we'll come to that later for 30 years now, haven't we? We have. We haven't got very far. We will do soon. We will do soon. Yeah, that's true. But it was clear Demis was very interesting at the time. You can spot him, but not live in the lab. For sure, yeah. Okay, so let's continue with this discussion of what's been happening and what is going to happen.
[0:29:09] What sort of problem is amenable, is tractable to the approach that you are spearheading? And I'd like everyone's input. And for instance, on Isomorphic Labs, which is a very ambitious project to reinvent drug discovery, you're an advisor, Paul. So that would be good to have your thoughts too.
[0:29:31] Well, maybe I can just explain that in the general case. So this really would sort of dawn on me after we did AlphaGo. So the program that beat the world champion at the game of Go, much more complex than chess. And hopefully you've all heard about that. So I won't explain that part. But the sort of...
[0:29:50] Problems that the methods we've developed are useful for, I can describe in a very general way. Basically, you have a huge combinatorial space. The number of moves in go is more than there are atoms in the universe. So you can't solve it in a brute force way. Same with the number of conformations of a protein. It's estimated 10 to the 300 or something enormous.
[0:30:11] And so that's the first part. I like those problem spaces because they can't be done with some kind of brute force method. Second thing is you need a clear objective function. Are you trying to win a game? Are you trying to find the best move? Are you trying to minimize the free energy in a system? So that's the second thing. So in order to hill climb towards that solution. And then the third thing is you need some source of data.
[0:30:34] And that can be real data or it could be an accurate simulator. Ideally, you have both. And there's really interesting interactions between using the data you have to create a simulator, to create synthetic data, additional synthetic data. And actually, that's what we had to do with AlphaFold. The 150,000 proteins in the PDB were not enough on their own. We had to create an early version of AlphaFold, fold a million proteins, and then select two, 300,000 extra ones that we thought were, at least AlphaFold thought it was confident, was accurate.
[0:31:03] And you have to be very careful with that, obviously, with synthetic data that actually the distribution matches. But if you have those three things, or you could couch the problem in that way, then I think the current methods we have, where you use a deep learning model to learn a model of the domain, and that guides a search process to be able to make that intractable combinatorial space tractable. And that's basically what AlphaGo and AlphaFold did.
[0:31:31] To find the kind of needle in the haystack that would be impossible to find otherwise. And drug discovery and isomorphic, I feel, is a continuation of that kind of problem. There is a compound out there in the laws of physics, the laws of chemistry, 10 to the 50. Life is all about making decisions.
[0:31:52] Sometimes you make questionable ones, like buying a new TV that's too big for the wall because the electronics store employee was a little too good at their job. Then sometimes you make great ones, like protecting your home and auto with Progressive. You get customized coverage options so you can be sure you have the right coverage. Get your quote today at Progressive.com. Progressive Casualty Insurance Company affiliates and other insurers not available in all states. The holidays are coming.
[0:32:22] But the Lincoln wishlist sales event is already here. And with exceptional offers on select Lincoln vehicles, you might just cross the perfect gift off your own list. You know, the kind that doesn't need wrapping. Consider this a little reminder to be good to yourself this time of year. Visit your local Lincoln dealer or go to Lincoln.com to learn more. Kate holds the key to every brand asset your business has ever made.
[0:32:53] She's the only person who knows the difference between teal, turquoise, and the correct company blue. But she's going on vacation. Luckily, before the company falls into design disarray, you realize you can put Kate into Canva. Save every color, font, and logo to a brand kit everyone can access. And Kate can vacation happily in the knowledge that no one will ever use Comic Sans. What will you Canva today?
[0:33:19] Or however you want to say how many compounds are there of the sort of drug compound size.
[0:33:25] And one or more of them may fit the properties that you're looking for in terms of the disease profile and so on. Then it's a question of can you find it, can you synthesize it, and so on. So isomorphic, you can think of us developing half a dozen more alpha-fold systems that, of course, alpha-fold is one component, the protein structure, but it's only one small component of the whole drug discovery process. And you can think of, we're looking at biochemistry and chemistry things to predict toxicity and ADME properties and so on.
[0:33:51] I wanted to sort of go back a bit. I'm a wet scientist, so to speak, and I'm in no sense an AI specialist.
[0:34:00] Two colleagues here. So what has it done for me, we could ask. And what it's done for my lab, we start with actually rather simple things. Machine learning. We shouldn't just sort of disregard this type of thing. I mean, we now can analyze images in minutes, which used to take us days. It saves an awful lot of time to actually be able to do it. And we can do a media survey in a minute or two. It's pretty boring what you get back with a large language model.
[0:34:29] I can be blunt with you, but it gets you started. So that sort of actually helps as well.
[0:34:36] Where now, and this is what Demis is referring to, I think, is how can we now implant beyond those sort of rather simple things, which I don't want to denigrate. They are very important. And by the way, I should say, AlphaFold has been amazing. We do experiments. We can try and imagine what mechanism there might be. We can put it through AlphaFold. We can see what might touch something else. It may or may not.
[0:35:01] Do that, but it immediately creates ideas and hypotheses. All of that is great. So what about the next steps? So the next steps I think people are thinking about is how can we apply a sort of connected loop of ways of working in a laboratory where AI can help at every stage as we go around. So it's not just simply a particular...
[0:35:27] Now, various places in the world are thinking about this and thinking about doing it. And I mean, there's difficulties in...
[0:35:39] Part of it is the type of data you have to analyze. AlphaFold worked with extremely useful data. I mean, immense amounts of data collected in the same way in a public database, by the way. I mean, for those who wonder why do we do public funded science, AlphaFold would have been totally impossible without that. And we... That's why we open sourced it afterwards. Absolutely. And by the way, they did it very decently.
[0:36:07] But it was a consequence of that access to public data curated in exactly the same sort of way. Much of the biological data, where I think this is going to be useful, isn't actually collected in that way. So we need to get, and Dennis may have views about it, ways of analysing it, which is of a different type. What we tend to have is, if you like, rather than lots of similar data,
[0:36:35] We have sort of wells in different places and different lengths. And somehow we've got to connect all this sort of thing together. And at this moment, I think that's a bit tricky. And we need to get the methods there. But I think the idea is how we can connect all the stages of data, hypothesis, hypothesis testing, generating an idea, back producing more data and going around that cycle and how AI can actually help us with, in my view, the ultimate.
[0:37:04] Thank you. We'll come on to building a virtual cell in a minute. But Alice, did you want to chip in on this point? Well, I guess in my space, it's very, very similar. If you work, if you pick up an ultrasound probe and you want to scan, we can't, you can't.
[0:37:28] Do that with a robot. You can't actually, you require the human expertise. So now people are very much thinking about having solved these individual little tasks, which is what we start with, is how can we now put them together and integrate every step. So pick up, you move the probe, you apply certain pressure. And how do we, some of it you can understand what to do. Some of it humans just do. And it's that.
[0:37:57] It's that fine-grained, what we call fine-grained information that is very difficult to be able to model at the moment. So the ultimate would be able to get a robot to be able to perform all these tasks and integrate them together. And that's what people are working. I think five years ago we would say that's a dream. Now people think that that's possible. Things are moving so fast because now you're talking about AI and humans working together beautifully, integrating workflows.
[0:38:27] Helping each other. But already people are worrying about what will happen next. Young scientists coming up through the, starting out on their career paths, I think are increasingly asking, where will it be when I'm trying to set up my own lab? What will the status be? And what will the role of the creative scientist be? And this is a whole other topic, but I'd like to explore it.
[0:38:53] The first thing to realize is doing research in a wetland is enormously boring. I mean, it really is boring. We spend most of our time transferring little volumes of liquid from one tube to another. And there is no way to excite anybody, let alone their intellect.
[0:39:11] So we could work on how we can get robots to do much of that. And in the Crick Institute, where I work, we have very good technical cores, which are not yet doing this in a way that is integrated in what we would like. But we have the resources and the way of thinking that would allow it. The reason I said it's boring is I think most of my colleagues in the lab wouldn't mind if they didn't have to do all of that all the time.
[0:39:39] However, what we do need to keep and to actually develop is creative thinking. So I think the focus here should be how can we turn these tools into assist the human being to be more creative in thought. Now, large language models are not creative in thought.
[0:40:03] They might sort of push you in certain directions. But working, I think, with AI processes where you, as your human mind is involved in different parts of this, may actually provide a way of operating in this sort of way of thinking about experiments, which could be very productive. Because what switches us on, it isn't transferring little volumes of liquid from one tube to another. It is a creative idea.
[0:40:29] Most of which are wrong, of course, and which you can then prove are wrong and sort of move on. So that's what excites me. Working with the machine, if I can put it this way, to generate creative thought. But yes, although it's true that a lot of time when you're doing pipetting and just watching the meniscus, that's a good time to be thinking about your experiments.
[0:40:54] Back in the 70s, I mean, when I could, you know, I didn't even do that. I was just looking at things. I had a lot of time to think. There is a problem, though. When we use complex technology to spend all your time getting the damn technology to work, I mean, and you stop thinking about the process, the biological process you're trying to study. If the technology is simple, it has to be said you do have a lot of time for thinking. But if the robot does it, maybe we have enough time to think.
[0:41:24] What do you think, Demis? I agree with what Paul said. And it's one of the reasons you couldn't attract me into the wet lab back in the day, Paul. I kind of understood that. He had no patience. Yes. So I'd rather stay in the digital world. But look, I think if I totally agree with you, I think that what this technology should be enable, if used correctly, is give us more time back for the creative thinking, the kind of taste that we talk about with great scientists have, setting the hypothesis or the...
[0:41:53] Choosing the right problem, even. Asking the right question. I always think asking the right question is much harder than solve. It's like, what is actually the right question? How can you state it in a really concrete way? The controls around it. All of that is the current AI systems have no way of doing that today. I really want to support that. It's asking the question, the agency of the right question. And actually, whether you're even asking the right question, asking endless. Yeah.
[0:42:20] questions, which sometimes just come to you from over here. And I don't think the ways we do it can deliver that. No, not at the moment. I mean, one day, I think AI systems, I don't think it's impossible for them to do that, but certainly not in the next five, 10 years, probably the next bit we can see. I think it will enable this. I think the other thing is it should enable much faster iteration.
[0:42:44] If you think about it, right? So through your idea space, maybe we can, again, a little bit more like how engineering is today. You quickly prototype something and then you move, you know, you test it and then you move on. There's going to be an interesting new way of bringing that into more and more scientific disciplines, I would say, that type of approach.
[0:43:04] I think it will also free up PhD students and postdocs to do higher level work, let's say, instead of, let's say, you're doing something to do with imaging, like I was doing fMRI and neuroscience at UCL. A lot of that analysis of the images will be done by the AI systems. You don't have to painstakingly write the support vector machines and other things we used to do back 20 years ago. And then instead, we can use that time to maybe think about the experimental side of what would be a good next question or hypothesis.
[0:43:33] I think in terms of, I think the amount that a single student, for example, will be able to do is going to be incredible. That's what I'd recommend in general, outside of science even for the youth of today is.
[0:43:45] I think there's the kind of cat is out of the bag in the sense like AI is not going back into the box. But what you can do, like I did in my generation in the 90s, is growing up with home computers sort of in the first generation and then the internet, is lean into that and get incredibly au fait with the technologies and including understanding how they work. So STEM subjects, all of that, I still recommend to the students of today because you'll be able to use, even if you might not be coding as much, you'll be able to use the coding tools.
[0:44:15] Way better, if you understand it. And I think the output of what you might be able to do as a PhD student in the future, in a few years' time, is like what a whole lab would have needed before. So for the enterprising students, the really smart ones, the go-getting ones, the very motivated ones, that could be hugely empowering. And of course, it might be available all around the world. Because these tools are basically distributed everywhere.
[0:44:40] Like phones and apps. So you can maybe access some of the most advanced technology in the world from whichever country you're in. So that could be pretty exciting from encouraging the most talented, most motivated students from anywhere around the world could maybe do cutting edge science, which would not be possible today. You'd have to go to one of the, relocate and find your way to one of the top centers around the world.
[0:45:02] So I think there's lots of really interesting ways things are going to change. And in terms of using the pipetting time for thinking, I think we can replace that with something else, go for a long walk in the country. But I do think this is an important point. In general, one thing I want to do with our Gemini chatbots or LLM side of the house that I also run at Google is one of my dreams is that...
[0:45:26] When those become really capable assistants, which we're maybe a couple of years away from, I would love it that we can then use technology less.
[0:45:35] Not more, less. Because I think today we're just bombarded by these unintelligent systems that are trying to hack our attention, whether it's social media or whatever, just the torrent of it. But we have to dip into that torrent because we need the information, the knowledge, and whatever it is that you're looking for that's useful out of that noisy stream. But if you had an AI system that you could effectively work on your behalf, you could sort of delegate it to that and just say, look, I'm going to do some deep thinking now between 10 and 6 p.m.
[0:46:05] Just don't interrupt me. Summarize what's happening in the world, these things I'm interested in at the end of the day. And you could really rely on it to do that. So I think that could actually sort of protect our mind space and attention space. That's what I dream of doing is getting through this period and then having technology that's actually smart and personalized and knows what you need. And you can sort of delegate that aspect, the information gathering aspect to it. So you've got more time to think if you're a scientist or do the things that you're interested in as a general citizen.
[0:46:33] I was just trying to imagine my graduate student wanted to leave. I'm just off now to do my deep thinking. Yes, exactly. There was a lot in what Demis said. Did you want to come in on the engineering question as a professor of biomedical engineering? I was going to say, I think he's said it well. I think in my own group, what I'm already seeing is postdocs, computational postdocs, who now it's just amazing how they can program.
[0:47:01] And get assistance with their software. And that is what they say. They have more time to think. They have more time to be creative and come up with.
[0:47:12] With much more advanced solutions where you would normally, just a few years ago, you would have probably had a few master's students as well. We need to be careful here because we do treat that all as part of training. So we mustn't forget that we do need to think about the education as well. But I do think that these tools are very accessible. So we are just going to see, as you say, more done by a PhD student.
[0:47:37] And the notion of a team, although I mentioned it earlier, the notion of who will be in a team, you still need interdisciplinarity coming in to understand how to build the appropriate models. But maybe they're small teams and not the large teams. So there were many directions we could go from what you said. One thing to pick up on.
[0:47:58] It's perhaps just the deluge of information that's coming out. Now, you mentioned that these systems should, in the right hands, be making people more creative. But there is, I'd be interested in your opinions of whether, in general, the way that AI is rolling out at the moment in science is leading to lots more creative science or a great mix of people doing creative things and people doing things because you can and because grant.
[0:48:23] Grant awarding bodies and journals are wanting to see new ways of working. And also then coming back to the point you made earlier about the publication system, a whole different question of how we can communicate science effectively if there's just so much more going on and resources are still stretched. So there's a lot there. Sorry, who wants to? Paul. Say something about...
[0:48:52] I think the technology has got dangerously seductive in some ways. If I look at some of our high-profile journals now, they will invent some way of looking at a problem and accumulate a lot of data. And it's clever, it's expensive, so only a few people can actually do it, but it doesn't actually come to very firm conclusions about anything of interest. In fact, I sense they're not even interested.
[0:49:21] In coming to conclusions of great interest because they're so full of themselves of having developed this technology to do something. It's actually something that's beginning to worry me. There's a journal called Cell which used to be a great journal. I can't bear to read it anymore because it's just full of this stuff and they just have no conclusion because they are not thinking about what it all means. And I think we've got to get thinking.
[0:49:49] And asking the questions back into this, because I think the seduction of the technologies is a problem. And it won't be solved by the computational, because it's actually a human problem that they need to think the objective of all of this is to understand and not just collect data. Yeah, I agree with that.
[0:50:11] There's some areas of science, which I wasn't going to name them, but then you named cells, so maybe I shouldn't. Where I think it is exercising, let's just get as much data as possible without really thought about how it tells us anything about functionality and other things. I think the early days of connectomics maybe was like that. When I was looking at that as a postdoc at MIT, I was like, well, what is this actually going to tell us functionally about the brain?
[0:50:34] But it's obviously a very seductive data gathering approach. And I think there's many areas like that where we need to think about the scientific question. I see this in different ways. We already were being deluged by this idea of big data, right? That was the buzzword, if you remember, in the early 2000s. It was big data. Now the buzzword is obviously AI. But I used to pitch what we were doing at DeepMind is big data is the problem. AI is the solution.
[0:51:03] That was one way I used to try and describe.
[0:51:06] What we were doing to early venture capitalists. They were still a bit confused about what we were trying to do, obviously, like solve intelligence. What does this mean? But I think big data is the problem. And it's increasingly, if you view it as the deluge of information we're creating as a society, obviously in science, we have these big machines, large Hadron colliders, all these things. And it's just producing reams.
[0:51:33] of data where maybe there's some insight in it, maybe not, or some structure or some pattern, but maybe not enough scientists are actually thinking about it. So first of all, we need more scientists to do that, perhaps even before they generate that data. But certainly, it's sort of the complexities beyond the level any human mind, even the smartest minds we have, can probably comprehend. And they're usually in areas which are not amenable to
[0:52:01] Making into kind of elegant mathematical equations, right? So that's why I used to say that AI is the perfect description language for biology, just like maths was for physics. Because biology, I think, you know, a lot of people try to, I don't think we're going to get Newton's three laws of motion for a cell. It's just too emergent. It's too dynamic. It doesn't mean there aren't rules and there aren't things to discover and understand about it. But I think it's going to be more like a simulation.
[0:52:25] Which is why we're thinking about virtual cells. And I think if you think about wider than that economics, such an emergent dynamic system, obviously, with humans as the atomic unit, which are already unbelievably complicated. How are we going to actually understand those types of systems?
[0:52:43] How does one do science on those types of systems? And I have all sorts of ideas about that, but it involves using AI to try and make sense of data and find the structure and maybe build a simulator on top of that like we've done with weather systems. Is this definitely possible? We've shown it in many directions, obviously with AlphaFold, but also more recently with predicting hurricanes and which direction they're going. The Met Office is using that. You know, that's super complicated. You used to have to have supercomputers doing fluid dynamics calculations for two weeks. Now we can.
[0:53:13] And so I think we're just scratching the surface of what really is going to be possible. And just to answer your first question about that, I think...
[0:53:35] There's going to take time for science to adapt to how to best use these tools. I think in a lot of cases, the tools are not being used in very useful ways or it's pretty mundane ways. Still useful, again, not to cost any doubt on that. It's very useful for dealing with paperwork that we have to do every day or just gathering, interpreting images. But that's really step one. So we should do that right now. That's clearly useful. But it's in itself a creative process to understand
[0:54:05] How should one apply these tools to the question that you have in mind? And I think that holds for business as well as science. Now I get, you know, CEOs for the last couple of years, they've, you know, especially a few years ago, they went from not hearing about AI to what is that, to like, oh, we need something on AI. And then they're like looking around and they're running an insurance company or whatever it is. And it's like, oh, we've just got to apply AI to something. And I often, you say to them, look, you don't, when I looked into the problem, it's like, no, you just need to use stats for that. It's fine.
[0:54:34] You don't need AI to do whatever it was you're doing. And so, in fact, if you try to use one of the latest systems, it would actually get in the way of that. It's just not the right...
[0:54:44] Problem for the proposed method. And I think that in itself is a taste question issue. Not only can you come up with the right question, can you come up with the right question and use the tools in the right way in service of that question? And that's, again, on the part of the scientist. You've got to understand your domain really well and also understand machine learning well.
[0:55:09] Students who understand that and then shape that correctly for that new world. I think that's part of this going to be this new taste that is required. We've mentioned the virtual cell a few times, and I think we should step up the ambition level and talk about the simulation of the virtual cell. A dream that you share. Yeah. Well, I'll start. I mean, this is complicated.
[0:55:36] People do have a go at it. They'll take a simplified cell, bacterial cell, reduce it to 400 or 500 genes, so the problem is better than 4,000. And then they'll do some... They'll probably try and identify all the reactions, build 400 differential equations, put it all together and predict something. Now, my comment there is if you...
[0:56:04] Have 400 different differential equations and you can't predict whatever you like. You're not very good at it.
[0:56:13] Differential equations. So I actually think it's a nonsense to start with. I don't want to say you shouldn't do it because we should let people try it because it certainly could be wrong. But I think it's exceedingly unlikely to work, especially given that all of these kinetics have been determined almost certainly in vitro and almost certainly won't work like that in in vivo. Now, I give that example simply to say we are nowhere near dealing with this problem.
[0:56:42] Let's think what we've got. We have disorder around us. Within a cell we have order. And the physicists worry about that a little bit because of the second law of thermodynamics. But as soon as it's sort of explained, it's isolated. And you put energy and they lose interest because they can see it's not contradicting the second law of thermodynamics. But it isn't just creating order. It's creating order with purpose.
[0:57:08] Order with purpose. Purpose for the entity, the cell, to survive and also to reproduce in a way upon which natural selection can work. This is ferociously difficult. And I struggle with it because I can't quite see how to do it. The nearest I've got, which I don't think it's a good idea, but it's the best I could, is to see whether we can divide up the cell into domains.
[0:57:35] A number of black boxes where we don't really care what's going on in the box. We just need to know what are the critical inputs and the critical outputs and see whether that simplification might produce something. Now, my lab's done an experiment which I just wanted to explain because I think it could be telling us something. We measured the rate of protein synthesis in the cell. Now, that's simple.
[0:58:03] Let's not worry about how we do it. If we have one culture and we measure it, and it's the same culture here, the mean will be absolutely identical. If we look in the population, the variability is incredibly high. So there will be a lot of cells that will be 50% of the normal level and 50% more. Now, us biologists think everything is tightly regulated. Actually, it isn't.
[0:58:31] It's very floppy or very sloppy. Furthermore, if you have a bell curve and you're out here, 10 minutes later, you go back to the mean. Now, this is weird, because you'd think hundreds of reactions are required in protein synthesis. It all averages itself out, and it'd be all tight, but it isn't. It's very, very floppy. Now, I think this may be telling us something in all of this, because we biologists...
[0:58:59] Too much think everything is highly controlled. But if it's highly controlled, if you get into a funny part of control space, you may never escape from it. Look, how often does our computer do something weird? What do we do? We switch it off and switch it back on again. We don't know why it's gone wrong. We just switch it off and switch it on. Now, cells can't do that. But if they're floppy and floppy...
[0:59:23] If you see what I mean. Maybe they never get stuck in the wrong place. This could be all hooey, okay? But thinking like this, with assistance, with the sort of stuff that Dennis is doing, is the way I would try and approach this problem. Sorry, this is the key, bringing thinking like this, or thinking like you did when you did your...
[0:59:45] Well, what you're saying is you have to be a bit crazy. And maybe you need to have an input on your models. You know, the craziness switch. Dial up the craziness.
[1:00:15] Just the right amount of crazy. So I think Paul's right in that when we first talked about this, maybe, I mean, gosh, maybe even 30 years ago, I would have, when I was an undergrad, I would have tried to do it in the maybe heuristic way of like a bunch of differential equations or some other rules. That's how AI systems were built back then, like chess computers, like Deep Blue.
[1:00:42] But it was very obvious to me actually during my undergrad that this was not going to work to get to general intelligence or even understand language or any of those things. You can't, you know, we were taught at Cambridge like first order logic is basically it's language is used first order logic rules. It's obviously wrong because half the time we don't speak, I don't speak grammatically. And, you know, we still understand each other even if it doesn't obey the logic system. So we're obviously doing something else, you know, without at least the human brain was.
[1:01:07] And then I sort of realized and I completely confirmed my suspicions of that with our early work at DeepMind of learning systems. And then eventually AlphaGo, like, well, you can't program directly heuristics for a Go system that would be world champion level. It's impossible because it's too esoteric and it's too intuitive. It's too about patterns. There's no easy things to write heuristics about like the value of material because every piece in Go is worth the same.
[1:01:34] So all the things that worked in chess do not work in Go. And so it had to be learned direct from experience and direct from data. The system playing against itself, figuring out its own intuitive heuristics, let's say, about the game and about what was useful, what were useful strategies. And in fact, it did discover...
[1:01:57] Famously, like new strategies that had never been discovered by human players, even though we played Go for 2000 plus years now. And so that was kind of extraordinary as well. Like, not only can it model something that was otherwise intractable, it can actually potentially you could use it to go beyond what
[1:02:14] We currently know as humans. And that makes sense because a chess computer or something that's programmed or scientific system is programmed directly with equations we already know about. It's quite hard for that to go then beyond the knowledge that the creators of that system had.
[1:02:30] Because obviously, what capacity has it got to do that? So that's one of the very exciting things that we got very excited. Obviously, everyone's realizing now about these new modern systems. They're learning, they're general. So that means they have potentially the capacity to go beyond, with the help of human experts, beyond our current level of knowledge, which was never true before of any other tools we've ever built as a species. Your car isn't suddenly going to just spontaneously fly.
[1:02:57] When you've designed the car, the engine of the car, it might not work in how you imagined it when the early car makers, but it's not going to suddenly do some capability you didn't design for it. So it's a very unusual type of machine and system. I would say it's unique, which is why I was so fascinated by it since I was a kid, because it's obviously different.
[1:03:19] And then with simulations, I do think the issue there and with the cell is, and what we've wrestled over the many years about discussing is, well, what is a good circumscribed system? It tumbles down to two things. Unless you're going to simulate the entire planet down to quantum level, which maybe one day we will do, then you're going to have to describe some...
[1:03:40] Some siloed self-contained system that you can approximate everything outside of it, right, in some way. And then the second question you have to answer is what granularity do you need to model the system in inside for it to be valuable for your predictions you're interested in? And chemistry is a great example of that. Like, it's amazing to me that there's room between physics and biology.
[1:04:04] And you can sort of separate it or approximate it up and down. It's kind of amazing.
[1:04:29] Actually, if you stop to think about it. And maybe there are other ways to do that with other parts of science. So my current thinking, which we haven't had time to discuss yet, but might as well tell you now, is to maybe we can do like cell nucleus. Perhaps that's a good starting point that Zego and I are working on. And we'll talk about that afterwards. But maybe that's a good subunit. Very complex. Yeah, probably still too complex. So many places to go, so little time.
[1:04:58] Alison, one thing that Demis mentioned was the potential for everybody around the world to benefit from these technologies that can be distributed, but you also mentioned the barrier to access of compute power. We have to be brief with the subjects we're going to cover now, but how do you ensure that the rest of the world can participate rather than just be carried along?
[1:05:27] I think, well, people are working on this. For example, if the work depends on computers that are high performance, then how do you do computation differently so it doesn't depend on the large computes? Then providing access as well. And then using techniques which still requires computation, but using things like federated analysis where you can share.
[1:05:55] The compute and then bring the results back. But some of this depends on the question you're looking at, the fidelity that you need to be able to build a model. But I think there are a variety of ways that people are moving forward. And that is just evolution of technological solutions.
[1:06:19] Yeah, I think we've got to sort out the computing problem. And I think it's also part of it as an energy cost problem in this country. I think we've got to get energy basically means intelligence now in effect, you know, via chips and data centers. You can sort of, it's going to basically be a one-to-one correlation. So we have some of the most expensive energy in the world here, which is a tricky, you know, for us to then scale up what we need.
[1:06:44] Having said all of that, I do think this is a creativity and imagination problem. So yes, you need tens of billions of dollars of computing if you want to build the latest frontier model. But if you're in academia or an institute, there's no way you should be thinking about doing that. This is pointless. You won't be able to keep up with that and all the people that you need. And also, it's not the thing to be doing. There's enough companies doing that. There's five in the US. There's three in China.
[1:07:13] What it is, is like we've been talking about black boxes and other things. I've said this to many departments and with varying degrees of success, but like they always ask me like, well, what should we work on here as professors of computer science at whichever top university? And if I was in academia, I would be looking at the analysis, the understanding of the black boxes, the stress testing of what they can do, the limits of that, how can we actually put...
[1:07:37] Monitoring tools. There's so many things that need to be done, benchmarks for that, that don't require a lot of compute. Because what I would do is, there's open source models, like we have Gemma 4, there's Chinese models, very good ones. They're only like six months off to a year off the frontier. Literally, like this time last year, they would have been frontier capability. They're tiny in size, so our model Gemma 4, for example, is built to run on a single laptop, even.
[1:08:03] Let alone a small university cluster. Easily could run many instances of these things. Same with the smaller Chinese models.
[1:08:10] And you could do plenty of great science on AI and also use it for plenty of great science, applying that kind of AI to another field. I mean, you literally wouldn't have had that a year ago. So it's just a bit of sort of FOMO of missing out and there's something exciting over here. But the best scientists, in my opinion, if you believe in what you're doing, you block out the noise. Just like we did back in 2010 when no one was working on AI, literally nobody. And everyone thought we were mad. And even in academia, it was considered to be a kind of like...
[1:08:39] We tried AI in the 90s. It doesn't work. But if you really believe in what you're doing and you're passionate about it, you should be able to block out that noise and just go in your lane where you think you have unique advantages to contribute. And I think multidisciplinary is one of those areas. And I actually believe that's one of the great things of the new era is it's never going to be an easier time to do proper multidisciplinary work because we can use these tools. You could use them to get up.
[1:09:08] I will turn to some of these questions. There were a lot of people interested in AGI and weather.
[1:09:37] Systems will become conscious and what that means. And it's a very complicated thing to talk about. Could I, I mean, I know that you set out to create AGI and to use AGI to solve world problems. That was your sort of, that's your sort of direction. Let's, maybe that can be talked about in another time. But now do you see, is there anything that worries you about any of this? We've talked about small worries, but what worries you?
[1:10:06] Well, there's huge worries. So obviously, we've talked about the opportunity. Obviously, the reason I spent my whole career building this is I want to advance science and especially medicine, actually, with AI. So that's clear with the work I've directly used these tools for myself, right, with AlphaFold and Isomorphic.
[1:10:21] But there are a couple of worries, and I'm happy to touch on the consciousness question as well. But the two main worries you can think of like this. Bad actors repurposing general purpose technologies for harmful ends. And those bad actors could be individual rogue people, bioterrorists, this type of thing to rogue states. So that's one. Second one is the technical about AGI risks. So as the systems become more autonomous...
[1:10:47] And obviously, they'll be more useful. That's why we're moving towards a gentic era. That's the first step of that. But they'll become more autonomous, more powerful. Can we make sure the guardrails we set them are strong enough to constrain the behavior of those systems to what we originally intended? And that's a super hard problem, too.
[1:11:07] So that's two problems. I kind of classify those as technical problems. And then beyond that, even if we solve that, there's the economics problem of how do we share the benefits as widely as possible to as many people, to benefit as many people and countries as possible. And then finally, if we solve that, there'll be the philosophical question about meaning and purpose.
[1:11:28] So there's just a lot of things to solve. I'm still very optimistic because I'm a huge believer in human ingenuity. And if we put our minds to it, especially in acute times of stress, I think humanity has always stepped up to the plate. But it's going to be, I think, first of all, we need to recognize what the challenges are. And I'm surprised, for example, there aren't more of my economist friends working, taking this seriously and working on what would be a post-AGI economic system that would work, for example.
[1:11:56] Do you think we can ever build a system where the guardrails will be simply removed? Do you think it is possible? That's the big question. So assuming a system gets more intelligent than us, which is what AGI would be, how does one keep guardrails and something like that?
[1:12:14] It's an extremely hard but also very interesting research problem and is one example of, I think, the sort of research that could be being done in academia or civil society. In fact, it might be better to be done there rather than the big tech companies sort of marking their own homework is that exact question. I suspect just from looking at how steerable our current systems is, I'm sort of optimistic about that. But it's an unsolved problem. It's called the alignment problem.
[1:12:41] For sure not solved. And anyone thinks it's like simple as like, where's the off switch? You just do some reading around it. It's going to get around that. This is what Asimov wrote and warned about with all his robot stories. The whole point about that was a warning, like the three laws of robotics don't work. And adding a zeroth law also doesn't work because it gets misinterpreted in certain contexts.
[1:13:04] But is there the bandwidth for people to actually do this work while everything's developing so fast? Is a slowdown needed? And if there was a slowdown needed, what would it look like? This isn't the way I imagined it 20 years ago in terms of the technologies where it's at, but what I always hoped it would do in things like AlphaFold and trying to advance medicine. But it would be much better, in my opinion, if...
[1:13:30] We've been doing the basic research in a more scientific way, like a CERN, like international collaboration like that. And then you could still move very fast with the applications. So things like AlphaFold or you could try and cure cancer for that before AGI arrived. But we wouldn't have to deal with the existential question until we were ready as a society. The technology hasn't gone like that because the language, what changed all that was the language bots, the chat bots becoming possible and also very commercial. So that's not going back in the box now.
[1:13:59] We have to think of some other ways of how can we get international collaboration around maybe standards or certification processes. And it has to be international, which is really hard right now with the geopolitics and the fragmentation of our international institutes. So it's a sort of bad confluence of just at the moment we need strong international institutions and collaboration, we're sort of in the nadir of that phase. So I have some ideas. I don't think, I think, but it's going to be, it's a tricky needle to thread here.
[1:14:29] Sorry, Devis, I don't want to keep coming to you. But since you mentioned you talk about consciousness and people do like to hear about it, just very quickly, do you think that everything that the human brain can do is computable? Well, my all-time hero is Turing. So I would say, and I loved his Turing machines and studying all of that at college. And I think the evidence, I mean...
[1:14:50] I've talked a lot with people like Roger Penrose, and obviously he would disagree, and he thinks there's something quantum in the brain, and Stuart Hameroff. I don't think there's any evidence of that so far. I feel like it's sort of, here's two strange things we don't fully understand. Let's put them together. But anyway, I don't want to, you know, we shouldn't.
[1:15:07] Roger's an amazing guy, so maybe he's right. But I think I haven't seen any evidence in all my neuroscience work. So my assumption would be that most things in the brain, everything is computable in the limit. But I think this adventure we're on, one of the reasons I wanted to build AI was to...
[1:15:26] Help use it as a tool to help neuroscience, but also maybe as a comparator to help us as a control really to see what was special about the mind, if anything. And so we'll come to that. Sorry, one problem here is that we don't really know what consciousness is. I mean, so we don't even know.
[1:15:41] So it's not a well-posed problem. But we have some aspects of it that are probably required, necessary but not sufficient, like self-awareness, sense of identity, these types of things. And my view is, and has always been, AGI, the reason we coined the term AGI rather than use the old term strong AI, so it used to be called strong and weak AI, and I didn't like strong AI because it implied consciousness.
[1:16:09] So it was conflated, like a sort of general intelligence with consciousness. I think those are dissociable properties, would be my guess. If you look also at animals, you know, our pet dogs and cats, there's a lot of things about them that feel pretty conscious, but they're not as smart as humans. So it feels like there's a gradation, and in my view, it could be separable. And I don't feel like at least the tools we're building are in any way...
[1:16:33] by whatever we mean by that definition. I don't feel they have a semblance of that, but I think they could. And so my recommendation, if I could wave a magic wand, is let's cross one Rubicon first, which is the AGI and these really smart and precision, super capable tools. And then use that, give ourselves time as a society and use those tools as well to maybe better pose this question of what is consciousness, do some neuroscience with it.
[1:17:03] Given 10 years with those kind of tools, we would be able to. And then maybe that would be the next question for societies. Do we want to cross the second Rubicon of actually trying to create conscious entities? And in my view, I don't think we want to. Those are big enough challenges in themselves. Do we really want to conflate that and cross both of them at the same time? Alison, I guess that all scientists get involved in these conversations about consciousness at some time. Do you want to chip in here?
[1:17:30] Well, I guess my work, I try and steer away from that, actually, from working in the applied area. But yes, of course, it comes up in the conversation. And also, how would you use that capability going back in the context of science? How would that actually help you as well? I think it's a full circle one as well. Yes, yes, indeed. Yes, indeed. One of the advantages of getting questions from the audience is they take you in places where you wouldn't necessarily have gone.
[1:17:59] You mentioned philosophy already, but a nice question is, what is the role of philosophy of science in the future of science? And a lot of conversations between scientists about where science is going happen without philosophers being involved. So does anyone want to? Well, I think philosophy is quite important in thinking about science, I have to say, and what knowledge is and what you can test and what you can't test. So I find...
[1:18:27] As I've already said, proper useful here in the sense that you don't really prove something's right. You only prove that it's wrong. And after a while, you sort of accept that it might be right because of the problem of what is deduction versus induction and so on. So I think you do end up in a bit of a tangle. But what we don't teach properly is the...
[1:18:53] is thinking about science and thinking how you do science, which does have a philosophical basis. Now, the philosophers get a bit, you know, tangled up with things. I mean, I think we should read them, but not actually, you know, be totally driven by them. But we don't act, it's difficult to quite say.
[1:19:14] We don't know what a scientific method is. I mean, you know, we can say general things, you know, that we should collect data, it should be reliable, it should be reproducible, we should challenge our own ideas. We can do a list of attributes, I mean, for sure. But I think that, and I think actually we need to teach that because it helps us in...
[1:19:36] But I really do push back in that there is some sort of gold standard scientific method of doing things because it differs from area to area. How you do science in physics is different from how you do it in biology or climatology. Some of the greatest physicists in the world were hopeless in thinking about climate because they simply thought about, they had to have complete understanding from A to Z.
[1:20:02] And so they criticize it because you weren't producing that. But that is because they tried to translate how they did their science, and that's what Roger does to some extent, to how people do science in other areas. So it's a complicated thing. It is something that we can deal with. We just have to recognize it and describe it. I think the philosophy's time has come.
[1:20:28] I would say if I was a philosopher right now, this would be the most exciting time ever because we not only do need a new philosophy of science or updated one because we touched on all the things that need updating, like how are we supposed to use simulations or black boxes, reverse engineering them. That needs to be included as part of the scientific method. Yes, and what is understanding now? Because also these black boxes can be, I think, potentially backed out into mathematical equations if that's what we wanted to do. So maybe it's a two-step process.
[1:20:57] So I think there's a new philosophy there. And then more broadly, I think we need some, the time is now for some great new philosophers like Kant or Wittgenstein or Spinoza, some of my favorites from the ancient times of like, well, okay, so what is going to be new ethical philosophy and virtue and purpose and all these things that we just discussed? If we get all the other technical things right, those are going to be the most important questions. And I think we're going to need some new philosophies of the human condition.
[1:21:27] If it's Wittgenstein, we'll just be silent all the time. Okay, two last things. One more question from here, and this one is for Alison and Paul. Possibly from a young scientist. Is there a possibility that AI will help us improve research culture and especially the kind of harassment of young scientists? What do you mean by research culture? I mean the culture of research that happens in laboratories around the world. Could you see a way that AI can...
[1:21:57] Just make it nicer to be a young researcher. You know, it's not so horrible as people may think. We've talked about it all in glowing terms for an hour and a half, and I just feel you do hear people complain. Well, look, you know, scientists are bad just like everybody else, so sometimes they misbehave and so on. I just don't think we should sort of get so tangled up with it. I do think the media love saying that they're all making stuff up and so on. It's rare. It's rare.
[1:22:27] Our culture isn't too bad. It's always improvable. I don't want to sound as if, of course it is, but let's not beat ourselves up too much over it, I would say. Alison? Well, if they mean in terms of there seems to be lots of pressure and to produce results now, and they're worrying because on archive their work might have been published, you can see that there's a lot of stress on the current...
[1:22:54] Young scientists. But I think we've also talked about this now. They should maybe pause and reflect and think about what being a scientist is and what scientists should be doing. But that's a tension that if publications matter and what's on their CV all the time.
[1:23:16] Then it's up to more senior people to stand up and say, no, I want you to be the best scientist, regardless of the number of publications. And to come out with a few outstanding papers rather than think about the volume of work. Volume doesn't matter at all. Exactly. What matters is quality of what you produce all the time, not quantity. And it's so obvious. It's astonishing that people don't realise it, really.
[1:23:44] Thank you very much indeed. I'd like to reiterate that the questions that you submitted have really informed this discussion. And so I hope I've represented at least some of your thinking in this conversation. Just to finish, Demis, when AlphaGo beat Lee Sedol in 2016, he retired from the game saying it's not the same game anymore. And I just wanted to... Well, it's a bit more complicated. But...
[1:24:13] Okay, maybe a constructed premise for a last question. But is science the same game with AI attached? Has something changed? And I'd like you all to just talk about that.
[1:24:27] Yeah, I think something will change. Look, I saw Lee Sedol recently when I went out to Korea. It was really nice to catch up with him. He's doing great. He's, you know, he's talking about, I mean, in a way he has visceral experience of what it's like for AI to encroach on his area. And I think the whole world can probably learn from that now. He was near the end of his career anyway. That's just something to, I mean, he was, I used to describe him as the Roger Federer of Go, which is what he was. He was, you know, 18 time world champion, but he was the strongest person of the, you know, player of the previous 10 years. Still at top of it.
[1:24:56] But he was getting towards the end of his career.
[1:24:59] So those things got a little bit conflated. But there was some bittersweetness and sadness for me, even with the AlphaGo match, especially as a games player myself. I used to play chess very professionally when I was young. And we had a chat actually on the eve of the match where we both joked, and he likes technology and computers, we could have been, but for some different events in our lives, maybe we could have been on the opposite sides of the board, right? Because at one point I was going to be a professional chess player.
[1:25:27] So I really respect that art, and it is art, and that skill.
[1:25:32] And there's no doubt that when an AI system comes along, it changes that, like it did with chess. And it changes the relationship. But having said that, chess is more popular than ever now, you know, to watch. And people don't care about watching chess computers play each other, just like we don't care about, you know, we still care about the 100 meters sprint in Usain Bolt, even though we have cars and things that can go faster than a human can. But, you know, we don't care about that. We want to see what our fellow human beings can do. So there's, I think in those areas,
[1:26:02] There's part of that. The only other thing I would say, which is probably relevant to science, is go players.
[1:26:08] I mean, they still, I think, describe it. And Go is regarded as more than just a game in Asia. It's a sort of mystical art. And some of the mysteries of the universe are thought to be embodied in the game because it's just such a beautiful kind of yin-yang game. And, of course, the best Go players used to tell me, and I have a lot of friends who are professional gamers in various sports, Go as well, that they want to know the mystery of the game. But do they really?
[1:26:38] It's the question, right? So, and I think that's the same of science. I have an insatiable curiosity. That's what's driven me. It's maybe pathological. It's like driven me my whole life. And AI is, and most of us in the room who are scientists have that, otherwise we wouldn't be scientists. And so we're striving, I think, to understand the laws of nature and the laws of the universe. And in my opinion, like my personal mission, nature of reality.
[1:27:01] These big questions, we would love to understand those answers. And AI will definitely help with that, but it will come at a cost. Thank you very much, Dennis. Alison. I was going to say something similar in the sense AI is definitely a tool that can help you, will be helping you do science in different ways. But that's a positive, it's not a negative. The driver in science, I think, for...
[1:27:29] As scientists, it's to understand something that we didn't understand before. Our motivation is curiosity about the unknown. Frankly, the tools we use to really back up what's just been said are important, but I don't think they change the character of that fundamental aspect of science. It's understanding the world. Maybe I can just finish on an optimistic note. I mean, I am very sure about in the next 10 years,
[1:27:59] We are going to be entering what I feel like is a new renaissance. It will be a new golden age of scientific discovery. I don't know what will happen beyond that. Maybe something else will happen. There will be a new sort of era. But I think at least we're going to sort of live through a kind of golden age in the next 10, 20 years. And hopefully when we look back, AlphaFold will just be one example of the things that we were able to crack with the help of AI. Thank you very much indeed. Thank you, all of you. It's been an enormous pleasure listening to you.
[1:28:28] And it's been a pleasure being here with this audience. I'd just like to say thank you all very much. Progressive's accident response can contact emergency services if you're in a major accident. The only thing worse than someone you love being in an accident is not being able to help them. So Progressive is here to help when you can't. So get peace of mind knowing your loved ones are protected with Progressive's accident response.
[1:29:06] Progressive Casualty Insurance Company and Affiliates. Limited availability in California. Driver is responsible for emergency and towing charges not covered by the policy. Coverage subject to policy terms and conditions. The secret society of semi-homemade? Listen up. Sam's Club has pre-made dishes for a Thanksgiving feast that's ready in two hours. All while tasting suspiciously homemade. Come join us. Sam's Club. Individual oven times may vary.
[1:29:31] Your sausage McMuffin with egg didn't change. Your receipt did. The sausage McMuffin with egg extra value meal includes a hash brown and a small coffee for just $5. Only at McDonald's for a limited time. Price and participation may vary. Promotion pricing may be lower than meal pricing.