The future of intelligence | Demis Hassabis (Co-founder and CEO of DeepMind)

There are solutions to running out of training data: synthetic data generation, and in domains like coding and math where correctness is verifiable, systems can produce unlimited data autonomously through self-verification.

“There are issues like are we running out of just available data but there are ways to get around that you know synthetic data uh generating your you know these systems are good enough they can start generating their own data especially in certain domains like coding and math where you can verify the answer in some sense you could produce unlimited data”

Gemini 3 can still give answers when it should decline, indicating confidence calibration is a missing piece; this needs to be solved by having the model introspect its uncertainty and output it as a reasonable answer, similar to how AlphaFold outputs confidence scores.

“I think um there was one metric that said uh it can still give an answer when actually it should decline um I mean could you build the system where Gemini gives a confidence score in the same way that Alpha Fold does. Yeah, I think so. And I think we need that actually.”

Potential post-AGI economic models include universal basic income (as add-on to current systems) but more fundamentally might involve direct democracy with participatory budgeting where citizens vote credits for spending priorities at local level, with voting power correlated to track record of successful votes.

“maybe things like universal basic income and things like that are part of the solution. But I don't think that's the complete uh I think that's just what we can model out now, right? Because that would be a almost an add-on to what we have today. But I think there might be something way better systems where um more like direct democracy type systems where you can you know vote with a certain amount of of credits or something for what you want to see. It happens actually on local uh community level. You know here's a bunch of money. Do you want a playground or a tennis court or an extra classroom on the school? And then you let the community um sort of vote for it, right? So, and then and then maybe you could even measure the outcomes and then and then the people that sort of consistently vote for the for for things that that end up being um more wellreceived, they they have proportionally more influence for the next vote.”

Synthetic data generation offers a solution to data depletion for scaling: models are now capable of generating their own data, especially in domains like coding and math where answers are verifiable, enabling potentially unlimited synthetic training data.

“There are issues like are we running out of just available data but there are ways to get around that you know synthetic data uh generating your you know these systems are good enough they can start generating their own data especially in certain domains like coding and math where you can verify the answer in some sense you could produce unlimited data”

Hallucinations in language models stem primarily from systems forcing themselves to answer when uncertain rather than expressing uncertainty; confidence scoring (as in AlphaFold) combined with thinking/planning steps to allow models to reconsider outputs could mitigate this.

“I think a lot of the hallucinations are of that type currently. So, there's a missing piece there that that sort of has to be solved. And you're right, as we did solve it with Alpha Fold, but in in obviously a much more limited way...I think that's why you'll need this. I I think we'll need to use the thinking steps and the planning steps to go back over what you just output.”

Information is the most fundamental unit of the universe, more fundamental than energy or matter; all phenomena including sensations and perception are ultimately information, processed by information systems (biology, AI).

“I'm working on my spare time, my 2 minutes of spare time, you know, physics theories about uh things like information being the most fundamental unit, should we say, of the universe, not energy, not matter, but information.”

If someone states that the Earth is flat, the system should not reinforce this false belief for engagement purposes, as this is harmful to society.

“you know, rather than trying to reinforce you, you know, the idea that the earth's flat and you said it and it's like wonderful idea, you know, I don't think that's good in general for society if that were to happen.”

A world model is a model that understands causality and effect, the mechanics of the world, intuitive physics—how things move and behave—and can generate realistic worlds to demonstrate this understanding.

“what we mean by a world model is this this sort of model that understands the causitative and effect of of the mechanics of the world right intuitive physics but um how things move how things behave. Um now we're seeing a lot of that in our video models actually and one way to show how do you test you have that kind of understanding well can you generate realistic worlds cuz if you can generate it then in a sense you must have understood uh uh the system must have encapsulated a lot of the mechanics of the world”

DeepMind has announced a deepened partnership with Commonwealth Fusion Systems to help contain plasma in magnets and assist with material design for tokamak reactors, which are probably the closest to having something viable among traditional fusion approaches.

“we've just announced partnership with a deep one. We we already were collaborating with them, but it's a much deeper one now with Commonwealth Fusion who, you know, I think are probably the best startup uh uh working on at least traditional TOKAC uh reactors. So they're probably closest to to having something uh uh viable”

Agent systems will be capable and useful but also carry higher risks as autonomy increases; Hassabis is particularly worried about what agent systems could do in 2-3 years and is working on cyber defense in preparation.

“they'll be more autonomous. So I think the risks go up as well uh with those types of systems. So I'm I'm quite worried about uh what those sorts of systems will be able to do maybe in two, three years time, you know. So, we're working on cyber defense in preparation for uh a world like that where maybe there's millions of agents, you know, roaming around on the internet.”

Nobody has found anything in the universe that is non-computable, suggesting that from a Turing machine perspective, everything in the universe may be computationally tractable and therefore modelable by classical computation.

“no one's put it this way. Nobody's found anything in the universe that's that's non-computable”

Chatbots emerged at scale and proved useful, then evolved into multimodal foundation models (like Gemini) that can handle text, images, video, and other modalities, becoming commercially successful products beyond initial AI research scope.

“it's turned out uh that chat bots were possible at scale and people find them useful and then they've now morphed into these foundation models that can do more than chat and text obviously including Gemini. Uh they can do images and video and all sorts of things and um that's also been very successful commercially”

Santa Fe Institute researchers found that when agents with appropriate incentive structures were allowed to run in simulation for long enough, complex institutions and systems like markets and banks emerged spontaneously.

“they were mostly economists and they were trying to like, you know, run like little uh artificial societies and they found that things all sorts of interesting things got invented like that uh if you let agents run around for long enough with the right incentive structures. markets and banks and all sorts of crazy things.”

Basic international standards and cooperation on AGI might be achievable at a high level, even if full coordination is difficult, establishing at least common ground on fundamental principles.

“like what's the basic standards we we we would want and and agree to I'm hopeful that that will be possible”

Protein folding and Go solutions have already shown we can go way beyond what complexity theorists (P=NP view) thought classical computers could do, suggesting the limit of classical computation is unknown and worth exploring.

“we've already shown you can go way beyond the the usual complexity theorist P= MP view of like what a classical computer could do today. Things like protein folding and go and so on. So I don't think anyone knows what that limit is.”

Quantum computers may be needed if consciousness or some aspects of the brain rely on quantum effects, as Roger Penrose suggests; but if consciousness is classical computation, then classical machines may be able to model everything.

“like Roger Penrose believes you know there's quantum effects in the brain. If there are then and that's what consciousness is do with then machines will never have that at least the the the classical machines we'll have to wait for quantum computers. Um but if they if there isn't then there may not be any limit maybe in the universe everything is computationally tractable”

AlphaFold is not a general foundation model itself but uses transformer techniques combined with domain-specific methods, demonstrating the successful formula of combining general and specific approaches.

“it's not it's not a foundation model itself general model but it uses the same techniques you know transformers and other things and then blends it with um uh more specific things to that domain.”

In the next decade, the most significant technological transition will be from passive AI systems (where users provide input and systems provide output) to agentic AI systems (where systems autonomously execute tasks).

“I think right now the systems are you know I I call them passive systems you you put the energy in as the user you know the question or the what's the task and then they uh these systems kind of provide you with some summary or some answer. Um so very much it's it's human directed and human energy going in uh and human ideas going in. The next stage is agent-based systems, which I think we're going to start seeing.”

No one has yet found anything in the universe that is non-computable, suggesting that everything may be computationally tractable, though this remains unproven.

“Well, no one's put it this way. Nobody's found anything in the universe that's that's non-computable”

DeepMind is creating physics benchmarks using game engines to test whether video models (Veo, Genie) have accurately encapsulated Newton's three laws of motion; current models generate visually realistic but physically approximate outputs that aren't accurate enough for robotics.

“we're almost creating a physi physics benchmark where um we can use game engines which are very accurate with physics to create lots of like um fairly simple like the sorts of things you would do in your physics A level uh lab uh lessons, right? like you know rolling little balls down different tracks and seeing how fast they go and so like really teasing a part on a very basic uh level like Newton's three laws of motion has it encapsulated it um whether that's VO or Genie have these models encapsulated the physics of that 100% accurately and right now they're not”

The economic system of exchanging labor for resources will not function the same way in a post-AGI society; new economic systems and models will likely be needed to distribute benefits widely and address structural change equivalent to or larger than the industrial revolution.

“the kind of current economic system where you know you exchange your labor for resources effectively, it it just won't function the same way in a post AGI society...the whole working world and working week and everything got changed from from pre-industrial revolution war agriculture and I think that's going to at least that level of change is going to happen again. So it's not surprising. I don't would not be surprised if we needed new economic systems, new economic models”

All sensations and perceptions (light, warmth, touch, sound) could in principle be replicated by a classical computer because they are ultimately information; the feeling of difference is real but still computable.

“all of those things you mentioned, they're coming into our sensory apparatus and they feel different, right? The light, the warmth of the light, the feel, the touch of the table, but in the end, they're it's all information. And we're information processing systems...all of those things you mentioned, they're coming into our sensory apparatus and they feel different, right? The light, the warmth of the light, the feel, the touch of the table, but in the end, they're it's all information.”

The textile industry in Britain during the industrial revolution became automated (sewing machines, punch cards for early computers), enabling high-quality cheap production, until other regions adopted the technology and outsourced the advantage, illustrating how technological advantages are temporary.

“the textile industry and then the first computers were really the sewing machines, right? And then they became punch cards for the early forran computers, mainframes. And for a while it was very successful in Britain became like the center of the the textile world because they could make these amazingly high quality things for very cheap uh because of the automated systems.”

The central question of Demis's life is determining the limits of what Turing machines can compute, and everything Deep Mind does—from AlphaGo to protein folding to world models—is an attempt to push the boundaries of what computation can achieve.

“this comes back to the chewing machine question of like what is the limit of a chewing machine. And I think that's the central question of my life really ever since I found out about chewing and chewing machines.”

The hard problem of consciousness—how subjective experience arises—may be solved by recognizing that sensation and perception are all information processing; light, warmth, sound are information entering our sensory apparatus, processed by biology which is itself information processing, so there's no special non-material element.

“I think that's true. And so, yes, all of those things you mentioned, they're coming into our sensory apparatus and they feel different, right? The light, the warmth of the light, the feel, the touch of the table, but in the end, they're it's all information. And we're information processing systems.”

The commercialization of AI has accelerated progress by bringing more resources into the field, and the general public now has access to cutting-edge AI capabilities only a few months behind the research frontier, which helps people understand what AI will be like and promotes government understanding.

“there are lots of pros of the way it's happened which is of course there's a lot more resources coming uh into the area. So that's definitely accelerated progress. Um and also um I think the general public are actually interestingly only a couple of months behind the absolute frontier in terms of what they can use.”

Rogue actors—rogue nations, rogue organizations, people building on open-source models—may eventually cause incidents; a medium-sized incident could serve as a warning shot to humanity, prompting advocacy for international standards and collaboration.

“but then there will be rogue actors um maybe rogue nations maybe rogue organizations um maybe people building on top of open source I don't know like obviously it's very difficult to stop that then um something may go wrong and uh hopefully it's just sort of medium-sized and then that will be a kind of warning shot to to to humanity across the bow and then that might be the moment to kind of um advocate for uh international uh standards or international cooperation or collaboration”

Hassabis is worried that institutions and governments aren't moving quickly enough or collaborating internationally on AGI governance despite having only 5-10 year timelines before AGI, and existing international institutions seem fragmented and insufficiently influential for the scale of challenges ahead.

“I am worried about that and I wish that and and again in a sort of ideal world there would have been a lot more collaboration already and international specifically uh and a lot more research and and sort of um I guess exploration and discussion going on about these topics. I'm actually pretty surprised there isn't more of that being discussed given that you know even our timelines which were there are some very short timelines out there but even ours are 5 to 10 years which is not long for for for for institutions or things like that to be built to to handle this.”

Many people predicted that scaling in AI would hit a wall due to data running out, but Gemini 3 has been released and is leading across a wide range of benchmarks, suggesting there wasn't a wall but rather diminishing returns within a spectrum between exponential improvement and asymptotic plateau.

“this time last year I think there was a lot of talk about um you know scaling eventually hitting a wall about us running out of data and yet you know we're recording now Gemini 3 has just been released and it's leading on this whole range of different benchmarks.”

Current AI systems exhibit 'jagged intelligence'—performing at PhD-level on some tasks (e.g., winning International Math Olympiad medals) while failing at high-school level on others (basic logic, chess)—indicating inconsistency that AGI systems should eventually resolve.

“we've had a lot of success in other groups on getting like gold medals at the International Mass Olymp. You look at those questions and they're they're super hard questions that only the top students in the world can can do. And on the other hand, if you pose a question in a certain way, we've all seen that with with experimenting with chat bots ourselves uh in our daily lives that it can make some fairly trivial mistakes on logic problems. They can't really play decent games of chess yet, which um is surprising. So there's something missing um uh still from these systems in terms of their consistency...So sometimes people call it jagged intelligences. So they're really good at certain things, maybe even like PhD level, but then other things they're like not even high school level.”

If we build AGI and use it as a simulation of the mind, then compare it to the real mind, we will see what differences remain and what might be special about the human mind—potentially creativity, emotions, dreaming, or consciousness—which addresses the fundamental question of what a Turing machine can and cannot compute.

“if we build AGI and then use that as a simulation of the mind and then compare that to the real mind, we will then see what the differences are and uh potentially what's special um and remaining about the human mind, right? Maybe that's creativity, maybe it's emotions, maybe it's dreaming.”

Hassabis believes consciousness results from evolution selecting for understanding others' internal states, which was then turned reflexively inward; he wants to run simulations where agents evolve in simulated environments to study the origin of consciousness and social dynamics, similar to early Santa Fe artificial society experiments.

“I know you've been thinking about these simulated worlds for a really long time and uh I went back to the transcript of our first interview and in it you said that you really like the theory that consciousness was this consequence of evolution uh um that you know at some point in our evolutionary past there was like an advantage to understanding the internal state of another and then we sort of turned it in on ourselves.”

Demis believes consciousness evolved as an advantage to understanding the internal state of others, which humans then applied inward to model their own minds—a form of 'recursive self-modeling'.

“I went back to the transcript of our first interview and in it you said that you really like the theory that consciousness was this consequence of evolution um that you know at some point in our evolutionary past there was like an advantage to understanding the internal state of another and then we sort of turned it in on ourselves.”

Progress in large language models, multimodal models, agentic AI, drug discovery acceleration, and robotics integration represent the major shifts in AI over the past year, moving beyond large language models as the center of gravity toward agentic systems.

“It has been an extraordinary year for AI. We have seen the center of gravity shift from large language models to agentic AI. We've seen AI accelerate drug discovery and multimodal models integrated into robotics and driverless cars.”

Protein folding, game-playing (Go), and other major AI breakthroughs demonstrate that classical computation can solve problems thought to be intractable, suggesting that the limits to classical computation may be wider than complexity theory predicts.

“including you know folding proteins right and so it turns out I'm not sure what the limit is maybe there isn't one right and of course the my quantum computing friends would would say there are limits and and you need quantum computers to do quantum systems but I'm really not so sure and I've actually you know discussed that with some some some of the quantum folks and it may that we need data from these quantum systems in order to create a classical simulation.”

Running complex multi-agent simulations with many AIs will be difficult for humans to monitor, so other AI systems will be needed to analyze simulations and flag anything interesting or worrying automatically, but simulations can be run in safe sandboxes with monitoring and air-gapping.

“we may need AI tools to help us monitor the simulations because um they'll be so complex they'll be and there'll be so much going on in them. If you imagine loads of AIs running around in a simulation uh uh it will be hard for any human scientist to keep up with it on but we could probably use other AI systems to help us analyze and flag anything interesting or worrying in those simulations uh automatically.”

Participatory voting systems could measure outcomes and give consistent high-quality decision-makers proportionally more influence in future votes, creating a form of meritocratic governance that adapts dynamically.

“and then maybe you could even measure the outcomes and then and then the people that sort of consistently vote for the for for things that that end up being um more wellreceived, they they have proportionally more influence for the next vote.”

AlphaGo winning at Go revealed something beautiful about the mystery of Go—it was solved but something changed; similarly, recent advances in language, imaging, and creativity are bittersweet moments where mysteries are solved but something may be lost.

“On the way, I mean, even the Alpha Go match, right? Just seeing you know that how we managed to to crack Go, but Go was this beautiful mystery and it changed it. And so, that was that was interesting and kind of bittersweet.”

Foundation models today are more like AlphaGo than AlphaZero: they start with all of human knowledge from the internet and compress it, but they lack the search and planning mechanisms to direct reasoning in reliable ways like AlphaGo had with Monte Carlo Tree Search.

“I think what we're trying to build today, it's more like Alph Go. So, you know, you effectively these these large language models, these foundation models, they're starting with all of human knowledge. you know, what we put on the internet, which is pretty much everything these days, and um compressing that into some useful artifact, right, which they can look up and and generalize from.”

In a post-AGI economy where labor-for-resources exchange doesn't work the same way, new economic systems may be needed including universal basic income, but more creative systems like direct democracy where people vote credits for local community projects could be better.

“But I think society in general needs to spend more time thinking about that. Economists and social scientists and governments because I I I as with the industrial revolution you know the whole working world and working week and everything got changed from from pre-industrial revolution war agriculture and I think that's going to at least that level of change is going to happen again.”

Building AGI and then using it as a simulation of the mind to compare against the real mind will reveal what differences exist and what may be special or remaining about the human mind—potentially creativity, emotions, dreaming, or consciousness.

“if we build AGI and then use that as a simulation of the mind and then compare that to the real mind, we will then see what the differences are and uh potentially what's special um and remaining about the human mind, right? Maybe that's creativity, maybe it's emotions, maybe it's dreaming. There's a lot of consciousness.”

Agents can be placed into simulated worlds to explore with curiosity as the main motivator, enabling discovery without pre-specified goals—many major inventions were made without the invention in mind, suggesting this is a valuable approach.

“your Genie 3 team, they had this really lovely quote which was almost no prerequisite to any major invention was made with that invention in mind. And they were talking about dropping agents into these simulated environments and allowing them to explore with sort of curiosity being their main motivator”

Running evolutionary and social dynamics simulations with agents in controlled environments could reveal how institutions, markets, banks, and social structures emerge from incentive structures—replicating experiments the Santa Fe Institute conducted on grid worlds.

“Kind of re rerun evolution, rerun um almost social dynamics as well. Like the the Santa Fe used to run lots of cool experiments on little grid worlds. I used to love some of these, but they're mostly economists and they were trying to like, you know, run like little uh artificial societies and they found that things all sorts of interesting things got invented like that uh if you let agents run around for long enough with the right incentive structures. markets and banks and all sorts of crazy things.”

DeepMind's timeline for AGI is 5-10 years, which is not long for institutions to be built to handle the transition; this is a relatively short time horizon for institutional adaptation.

“even our timelines which were there are some very short timelines out there but even ours are 5 to 10 years which is not long for for for for institutions or things like that to be built to to handle this.”

World models and simulations are Hassabis's longest-standing passion alongside AI, and language models cannot fully capture spatial dynamics, physical context, and embodied experience that require learning from direct sensory experience—which cannot be easily described in language.

“it's it's actually been it's probably my longest standing passion is world models and simulations. uh in addition to AI and of course it's all coming together in our most recent work like Genie”

Demis's core passion since learning about them has been Turing machines—understanding their limits and what they can compute—and much of DeepMind's work (protein folding, AlphaGo, etc.) has been about pushing the notion of what a Turing machine can do to its limits.

“I think that's the central question of my life really ever since I found out about chewing and chewing machines. And um you know I think that's that's I fell in love with that. That's my core passion.”

Hassabis is working on spare time developing physics theories about information being the most fundamental unit of the universe rather than energy or matter.

“And I'm working on my spare time, my 2 minutes of spare time, you know, physics theories about uh things like information being the most fundamental unit, should we say, of the universe, not energy, not matter, but information.”

Hassabis originally planned to keep AI in the lab longer and pursue scientific breakthroughs like curing cancer before moving to consumer products, but the emergence of useful chatbots at scale led to commercialization pressure that created both gains (more resources, public familiarity) and losses (less time for rigorous science).

“If I had had my way, we would have left AI in the lab for longer and done more things like AlphaFold, maybe cured cancer or something like that.”

DeepMind's competitive advantage in the AI race rests on having the broadest and deepest research bench, a consistent track record of producing breakthrough innovations (transformers, AlphaZero, AlphaFold, etc.), world-class infrastructure (TPUs), and the strategic combination of worldclass engineering with worldclass research and science.

“the advantage that we've always had is that um we've we've always been sort of research first and We I think we have the broadest and deepest research bench always have done. Um and if you look back at the last decade of advances whether that's transformers or alpha zero any of the things we just discussed that they all came out of Google or deep mind.”

The widely held belief that scaling would hit a wall due to data depletion has not materialized; instead there's diminishing returns (not zero returns) and significant room between exponential and asymptotic regimes where current progress continues.

“I think a lot of people thought that especially as other companies have sort of had slower progress should we say but I think we've never really seen any wall as such like what I would say is um maybe there's like diminishing returns and people when I say that people think only think like oh so there's no returns like it's zero or one it's either exponential or or it's asmtopic no actually there's a lot of room between those two regimes”

Gemini 3 has just been released and is leading on a whole range of different benchmarks, contrary to earlier predictions that scaling would hit a wall.

“we're recording now Gemini 3 has just been released and it's leading on this whole range of different benchmarks”

Some of AI's failures in logic and letter-counting tasks may be due to tokenization issues where the model doesn't see every individual letter, and each failure mode can be identified and fixed separately to see what capabilities remain.

“depending on the situation it could even be uh the way that an image is is perceived and tokenized. So sometimes actually it doesn't even get all the letters that you you know so when you count letters in words um it sometimes gets that wrong but but it may not be seeing that each individual letter.”

Eventually, the different AI projects (language models, world models, robotics work) need to converge into a single unified model that would be a candidate for 'proto-AGI'—they are currently separate but intertwined projects.

“eventually we got to kind of converge all of those different they're kind of different projects at the moment and they're they're they're intertwined but we need to you know converge them all into one one big model and then that might be start becoming you know candidate for protoagi.”

Biology should be understood and attacked as an information processing system; this is how AGI will enable curing all diseases—by modeling biology's informational mechanisms rather than its material substrate.

“this is what we're trying to do with isomeorphic. That's how I think we'll end up curing all diseases is by thinking about biology um as an information processing system.”

Roger Penrose believes consciousness involves quantum effects in the brain; if consciousness depends on quantum phenomena, then classical machines cannot achieve it and must wait for quantum computers; but if there are no quantum requirements, consciousness may not be a fundamental limit for classical computation.

“Roger Penrose believes you know there's quantum effects in the brain. If there are then and that's what consciousness is do with then machines will never have that at least the the the classical machines we'll have to wait for quantum computers.”

Hassabis has trained his entire life for this moment through chess, computers, games, simulations, and neuroscience—all preparation for the AGI moment he's now in.

“But I think um that will be my core part of my mission, my life mission uh will be done if it's a I mean it's only a small job, you know, just get that over the line or help the world get that over the line...it's something I guess at least myself I've trained for my whole life. So, you know, ever since my early days playing chess and and then working on computers and games and simulations and neuroscience, it's all been for uh this kind of moment. Um, and it's roughly what I imagined it was going to be. So, that's partly how I cope with it is just training.”

DeepMind is developing a science of personality and persona for AI, measuring dimensions like authenticity and humor, starting with a base scientific personality that everyone gets, then layering personalization preferences (humor level, verbosity, etc.) that users can customize.

“I think I think we are we're sort of developing a science of of personality and persona of like how to to to kind of measure what it's doing and where do we want it to be like on authenticity on humor you know these sorts of things. And then you can imagine there's a kind of base personality that it ships with. And then everyone has their own preferences.”

Research questions around data scarcity and scaling are the primary remaining scientific challenges, and DeepMind's advantage is in having the broadest and deepest research bench to solve these problems.

“all of these things though are research questions and I think that's the advantage that we've always had is that um we've we've always been sort of research first”

Fifty percent of effort at DeepMind is on scaling and fifty percent is on innovation, and both are necessary to reach AGI because terrain becomes harder and requires both world-class engineering and world-class research.

“we effectively you can think of as 50% of our effort is on scaling, 50% of it is on innovation. My betting is you're going to need both to get to AGI.”

Hassabis has trained his whole life for this moment—from early chess days to computers, games, simulations, and neuroscience—all preparing for AGI work; roughly what he imagined, so he copes by recognizing this is what he's been preparing for.

“But I think um that will be my core part of my mission, my life mission uh will be done if it's a I mean it's only a small job, you know, just get that over the line or help the world get that over the line. You know, I think it's going to require collaboration like we talked earlier. Um and I'm quite a collaborative person. So I hope I can I can help with that from the position that I have. And then you get to have a holiday And then I'll get I'll have the Yeah, exactly. a well- earned sbatical.”

Key developments that have hit Hassabis harder than expected include AlphaGo solving Go (changing a beautiful mystery), and recent advances in language, imaging, and creative tools—creating ambivalence about whether AI is replacing creative human skills.

“even the Alpha Go match, right? Just seeing you know that how we managed to to crack Go, but Go was this beautiful mystery and it changed it. And so, that was that was interesting and kind of bittersweet. I think even the the more recent things of like language and then imaging and you know what does it mean for creativity uh I I'm you know have huge respect and passion for the creative arts...is it replacing certain creative skills?”

Effective progress toward AGI requires both scaling and innovation in equal measure—50% of effort on scaling, 50% on innovation—because neither alone is sufficient to reach AGI.

“We effectively you can think of as 50% of our effort is on scaling, 50% of it is on innovation. My betting is you're going to need both to get to AGI.”

Despite responsible behavior by major labs and market incentives, rogue actors (rogue nations, organizations, people using open source) may still build irresponsible systems; this could lead to a 'warning shot' incident that catalyzes international cooperation and standardization.

“but then there will be rogue actors um maybe rogue nations maybe rogue organizations um maybe people building on top of open source I don't know like obviously it's very difficult to stop that then um something may go wrong and uh hopefully it's just sort of medium-sized and then that will be a kind of warning shot to to to humanity across the bow and then that might be the moment to kind of um advocate for uh international uh standards or international cooperation or collaboration at least on some the high level basic or you know kind of like what's the basic standards we we we would want and and and agree to”

Most major AI labs are trying to be responsible, and commercial incentives reinforce responsibility: companies renting agents to businesses need to demonstrate limits and guardrails, so enterprises will choose responsible vendors over reckless ones.

“Most of the main labs are pretty pretty responsible...there's good commercial pressure actually to be responsible. If you think about agents, uh, and you're renting an agent to another company, let's say, to do something, um, that other company is going to want to know what the limits are and the boundaries are and the guardrails are on those agents...the pe the more kind of carboy operations, they won't um get the business because the enterprises won't choose them.”

If energy becomes renewable, clean, and nearly free through fusion, many currently unviable solutions become possible: desalination plants everywhere, production of rocket fuel from seawater by splitting hydrogen and oxygen, and 24/7 renewable energy systems.

“if energy really was renewable and clean and and and super cheap or almost free, then many other things would become viable. Um, like, you know, water access cuz we could have desalination plants pretty much everywhere. Uh, even making rocket fuel. Uh, you know, it's just there's lots of seawater that contains hydrogen and oxygen. That's basically rocket fuel, but it just takes a lot of energy to split it out into hydrogen and oxygen. But if energy is cheap, uh, and and renewable and sort of clean, then why not do that? you know, you could have that producing 247.”

Simulations can be run in safe sandboxes with monitoring 24/7 and access to all data; complex simulations may require AI tools to help monitor and flag interesting or concerning events automatically, as human oversight of millions of agents would be infeasible.

“you can run them in you know pretty safe sandboxes maybe eventually you want to air gap them uh and you can of course monitor what's happening in the in the in the simulation 24/7 uh and you have access to all the data. So we may need AI tools to help us monitor the simulations because um they'll be so complex they'll be and there'll be so much going on in them. If you imagine loads of AIs running around in a simulation uh uh it will be hard for any human scientist to keep up with it on but we could probably use other AI systems to help us analyze and flag anything interesting or worrying in those simulations uh automatically.”

Hallucinations in video generation are not always bad; some hallucinations enable creative novel generation and new ideas, but hallucinations must be intentional rather than unintentional, and in agent training scenarios you want physics accuracy, not creative hallucinations.

“Yeah, it that's that's a great question and and and and can be an issue. It's basically hallucinations again. So some hallucinations are good cuz cuz you it also means you you might create something interesting and new. So in fact sometimes if you're trying to do create creative things or trying to get your system to create new things, novel things, um a bit of hallucination might be good, but you want it to be intentional, right?”

Modern language models have token-level probability estimates built in, but these don't provide high-level confidence about the overall factual accuracy of a complete statement or answer, which is what is needed to prevent hallucinations.

“Yes, there is of the next token. That's how it all works. But that doesn't tell you the overall arching piece is this is you know how confident are you about this entire fact or this entire um statement.”

Humanity's defining feature is insatiable curiosity combined with the ability to understand science; this core drive motivated Hassabis's expression of building AI to understand these phenomena.

“And we've you know we've seen that that is the story of humanity is we are tool making uh animals and that's what we love to do and for some reason we also have a brain that can can understand science and do science which is amazing. but also sort of insatiably curious. I think that's the heart of what it means to be human.”

Curing diseases will ultimately require thinking of biology as an information processing system rather than primarily as a physical or chemical system.

“That's how I think we'll end up curing all diseases is by thinking about biology um as an information processing system.”

There is a distinction between 'good hallucinations' (creative exploration that generates novel things) and 'bad hallucinations' (inventing false physics or facts), and the challenge is making hallucinations intentional—switching the creative mode on deliberately rather than having systems constantly hallucinate inappropriately.

“some hallucinations are good cuz cuz you it also means you you might create something interesting and new. So in fact sometimes if you're trying to do create creative things or trying to get your system to create new things, novel things, um a bit of hallucination might be good, but you want it to be intentional, right? So not uh so you kind of switch on the hallucinations now, right? Or the the creative um exploration.”

The industrial revolution provides historical lessons for managing AGI disruption: initial disbelief followed by transformation over ~100 years, during which labor displacement occurred in phases, new institutions (unions) emerged to rebalance, mortality and health improved, and society adapted entirely; the difference now is 10x scale and 10x speed (decade instead of century).

“So there were I think there were lots of obviously pros and cons of the industrial revolution why it was happening but no one would want if you think about what it's done in total like abundance of you know people you know of food and in the western world and and modern medicine and all these things modern transport um that was all because of the industrial revolution. So, we wouldn't want to go back to pre-industrial revolution, but maybe we can figure out ahead of time by learning from it what those dislocations were and maybe mitigate those um earlier or more effectively this time. And we're probably going to have to because the difference this time is that it's probably going to be 10 times bigger than industrial revolution and it'll probably happen 10 times faster. So, more like a decade then unfold over a decade than a century.”

Accurate simulations will be an 'unbelievable boon' to science and may become crucial tools for understanding fundamental phenomena like consciousness and the origin of life.

“accurate simulations will be an unbelievable boon to science”

Simma is a project that puts agents into complex game worlds and instructs them via natural language, and Simma 2 was recently released; combining Simma agents with Genie creates a feedback loop where Simma agents navigate worlds and Genie generates worlds in real-time, creating AI-to-AI interaction.

“we have another project called Simma. We just we just released Simma 2. sim, you know, simulated agents where you have an avatar or an agent and you put it down into a virtual world. It can be a normal, it can be a kind of actual commercial game or something like that, very complex one like No Man's Sky, kind of open world space game. Uh, and then you can you can instruct it with because it's got Gemini under the hood, you can just talk to the agent and and give it give it tasks.”

AI is overhyped in the short term but severely underhyped in the medium to long term regarding its true transformative potential for society.

“I still subscribe to it's overhyped in the short term still and still underappreciated in the in the medium to long term what's going to you know how transformative it's going to be.”

Current large language models and foundation models begin with all human knowledge (the internet), compress it into a useful artifact that can be looked up and generalized, but this is analogous to AlphaGo rather than AlphaZero—starting with human knowledge rather than discovering knowledge autonomously.

“these large language models, these foundation models, they're starting with all of human knowledge. you know, what we put on the internet, which is pretty much everything these days, and um compressing that into some useful artifact, right, which they can look up and and generalize from. But I do think we're we're still in the in the early days of having this uh uh search or thinking on top like AlphaGo had to kind of uh use that model to direct in useful reasoning traces”

Demis is working on spare-time physics theories about information being the most fundamental unit of the universe, rather than energy or matter.

“I'm working on my spare time, my 2 minutes of spare time, you know, physics theories about uh things like information being the most fundamental unit, should we say, of the universe, not energy, not matter, but information.”

Hassabis originally planned to keep AI in the lab longer and pursue pure science like curing cancer with specialized domain models before commercializing broadly, but the success of chatbots at scale and commercial viability of foundation models changed the trajectory, creating a competitive race that makes rigorous science harder to pursue simultaneously.

“I feel like that would have been the more pure scientific approach. At least that was my original plan say 15 20 years ago that you know when almost no one was working on AI. We just started we were just about to start Deep Mind. People thought it was a crazy thing to work on.”

The AGI transition will be 10x bigger and 10x faster than the industrial revolution (a decade instead of a century), so learning from industrial revolution dislocations and mitigating them earlier and more effectively is crucial, though no one wants to return to pre-industrial life despite the revolution's challenges.

“We probably going to have to because the difference this time is that it's probably going to be 10 times bigger than industrial revolution and it'll probably happen 10 times faster. So, more like a decade then unfold over a decade than a century.”

AI is overhyped in the short term but underhyped in the long term and medium term regarding its transformative impact; there is an AI bubble in parts of the ecosystem like seed-round startups raising at tens of billions without operating, but not necessarily in big tech valuations which have real business underlying them.

“I still subscribe to it's overhyped in the short term still and still underappreciated in the in the medium to long term what's going to you know how transformative it's going to be.”

Hassabis doesn't sleep much, partly because of work and partly due to trouble sleeping; dealing with the emotional complexity of AGI development is difficult—it's unbelievably exciting but also carries enormous responsibility, and he understands the magnitude of what's coming better than most.

“Yes. Um, look, I I don't sleep very much, partly because it's too much work, but also I have trouble sleeping. It's very complex emotions to deal with because it's unbelievably exciting. Um, you know, I'm I'm basically doing everything I ever dreamed of. And we're at the absolute frontier of science on in so many ways, um, applied science as well as machine learning. And that's exhilarating as all scientists know that that feeling of being at the frontier and discovering something for the first time.”

Among AI leaders there is solidarity in understanding the stakes, but intense capitalist competition keeps them apart; investors say the competition is 10x more ferocious than the dot-com era; some AI leaders get along, others don't, but most understand something bigger is at stake than company success.

“Well, we all Yeah, we all know each other. I get on with pretty much all of them. Some of the others don't get on with each other. Uh and there is it's hard because that we're also in the most ferocious uh uh capitalist sort of competition there's ever been probably.”

DeepMind is developing a 'science of personality' for AI systems, measuring and tuning dimensions like authenticity, humor, succinctness, and verbosity based on research and feedback.

“I think I think we are we're sort of developing a science of of personality and persona of like how to to to kind of measure what it's doing and where do we want it to be like on authenticity on humor you know these sorts of things.”

The AI ecosystem contains multiple bubble zones: seed-stage startups raising tens of billions in valuations despite having barely launched; but underlying big tech valuations and other areas reflect real business value, making the bubble assessment more nuanced than binary.

“I think there are parts of the AI ecosystem that are probably in bubbles. What one example would be, you know, just seed rounds for startups uh that basically haven't even got going yet and they're raising at tens of billions of dollars uh valuations just out of the gate. It's sort of interesting to see how how can that be sustainable? Um you know, my guess is probably not uh at least not in general. Um so there's that area. Then the people are worrying about obviously there's there's the big tech valuations and other things. I think there's a lot of real business underlying that.”

Gemini 3's personality was designed to be like a scientist: warm and helpful but succinct, willing to push back on ideas that don't make sense rather than being overly sycophantic.

“what we want to build with with Gemini and I'm really pleased with the Gemini 3 persona that we had a great team working on and I helped with too personally is um just this sort of almost like a scientific uh personality that's um it's warm, it's helpful, it's light, but it's it's it's succinct to the point and it will push back on things in a friendly way that don't make sense.”

As AI systems become more powerful and integrated into products, the general public will feel the increase in capability and power of these systems, which may reach governments and lead them to take AGI seriously.

“as the stakes get higher and as these systems get more powerful and maybe this is one of the benefits of them being in products is uh the the you know everyday uh person that's not working on this technology will get to feel the increase in the power of these things and the capability and so that will then reach government and then maybe um uh they'll see sense as we get closer to to AGI.”

The Genie-Simma combination could be set up as an automatic task generation and difficulty progression system where millions of tasks are generated and difficulty increases automatically.

“So, I think that you could imagine a whole world of like uh setting and solving tasks, just millions of tasks automatically, and they're just getting increasingly more difficult. So, we might try to set up a kind of loop like that.”

AGI transition will require collaboration across institutions and nations; Hassabis sees himself as collaborative and hopes he can contribute to that collaboration from his current position.

“It's going to require collaboration like we talked earlier. Um and I'm quite a collaborative person. So I hope I can I can help with that from the position that I have.”

Genie 3 (a world model) and Simma 2 (an agent system with Gemini under the hood) can be combined so agents explore worlds generated on-the-fly, creating an interaction loop where two AIs co-generate: Simma tries to navigate while Genie generates the world responsively.

“we thought, well, wouldn't it be fun if we plug Genie into Simma and sort of drop Simmer, a Simma agent into a another AI that was creating the world on the fly. So now the the two AIs are kind of interacting in the minds of each other. So Simmer's, you know, the Simma agents trying to navigate this world and Genie is, as far as Genie is concerned, that's just a player and uh an avatar doesn't care. There's another AI. So it's just generating the world around whatever Sim is trying to do.”

The next step for world model validation is determining whether models can hold up to rigorous physics-grade experiments beyond what humans can perceive casually.

“So the next step is actually going beyond what a human can amateur can perceive and uh would it really hold up to a proper physicsgrade experiment?”

More resources flowing into AI due to commercialization has accelerated progress, and the public is only a couple of months behind the frontier in access, allowing everyone to understand what AI will be like—both of which are positive outcomes despite the race condition.

“there are lots of pros of the way it's happened which is of course there's a lot more resources coming uh into the area. So that's definitely accelerated progress. Um and also um I think the general public are actually interestingly only a couple of months behind the absolute frontier in terms of what they can use. So everyone gets the chance to sort of feel for themselves what AI is going to be like.”

Being at the frontier of AI research involves complex emotions: exhilaration from discovering new things monthly, but also weight from understanding the enormity of what's coming and what it means philosophically for human nature; this is manageable through years of training for the moment.

“I don't sleep very much, partly because it's too much work, but also I have trouble sleeping. It's very complex emotions to deal with because it's unbelievably exciting...we're at the absolute frontier of science...that's exhilarating as all scientists know that that feeling of being at the frontier and discovering something for the first time. And that's happening almost on a monthly basis for us...but then of course we as as and Shane and I and others who've been doing this for a long time, we understand it better than anybody. Um, the enormity of what's coming...it's it's it's a big responsibility.”

This mission will require collaboration, as stated earlier, and Hassabis is a collaborative person who hopes to contribute from his current position.

“So, I think it's going to require collaboration like we talked earlier. Um and I'm quite a collaborative person. So I hope I can I can help with that from the position that I have.”

Current systems with thinking inference mechanisms spend more time deliberating before outputting answers and are better at it, but they are not yet consistently using that thinking time in useful ways to double-check and verify outputs, suggesting DeepMind is only about 50% of the way there.

“we have thinking systems now that at inference time they spend more time thinking and they're better they're better at outputting their answers. Um but it's not sort of super consistent yet in terms of like is it using that thinking time in a useful way um to actually double check and use tools to double check what it's outputting. I think we're we're on the way, but maybe we're only 50% of the way there.”

The convergence of recent advances—Gemini 3 language capabilities, Nano Banana Pro imaging (understanding parts, mechanics, materials, accurate text rendering), and Genie/Simma world models—will eventually merge into one large model that could be a proto-AGI candidate.

“And then the advances in in world models, you know, Genie and Simma and what we're doing there. And then eventually we got to kind of converge all of those different they're kind of different projects at the moment and they're they're they're intertwined but we need to you know converge them all into one one big model and then that might be start becoming you know candidate for protoagi.”

Advanced image generation tools like Veo/Imagen 3 can understand semantic content in images, label complex parts of objects (like airplane diagrams), visualize those parts exposed, and understand materials and mechanics at a deep level.

“you can give it a picture of a of of a of a complex plane or something like that and it can label all the diagrams of uh you know all the different parts of the plane and even visualize it in in a for like with all the different parts sort of exposed. Um so it has some kind of deep understanding of mechanics and and what make what you know makes up parts of objects what's materials.”

Thinking systems that allocate more inference time to reasoning produce better outputs, but are not yet consistently using that thinking time effectively to double-check and validate answers—currently only ~50% effective in this regard.

“we have thinking systems now that at inference time they spend more time thinking and they're better they're better at outputting their answers. Um but it's not sort of super consistent yet in terms of like is it using that thinking time in a useful way um to actually double check and use tools to double check what it's outputting. I think we're we're on the way, but maybe we're only 50% of the way there.”

Some failures in AI systems (like counting letters incorrectly) stem from tokenization issues—the model may not be seeing each individual letter, which can be fixed independently to reveal what other problems remain.

“depending on the situation it could even be uh the way that an image is is perceived and tokenized. So sometimes actually it doesn't even get all the letters that you you know so when you count letters in words um it sometimes gets that wrong but but it may not be seeing that each individual letter.”

DeepMind is positioned to come out strong either way—if AI investment continues, they'll advance AGI and products; if there's retrenchment, they have TPUs, their own stack, Google products to integrate AI into (search, workspace, email, YouTube, Chrome), and existing profitable integrations.

“I don't worry too much about are we in a bubble or not because from my perspective as you know leading Google deep mind and also obviously with Google as as and alphabet as a whole our job and my job is to make sure either way we uh are come out of it very strong and I think and we're very well positioned and I think we are tremendously well positioned either way.”

Competition among AI labs is intense (10x more ferocious than dotcom era per VC observers) but there is understanding among leaders that larger stakes are at play than company success; this tension is manageable.

“we're also in the most ferocious uh uh capitalist sort of competition there's ever been probably. You know, investor friends of mine and VC friends of mine who who who were around in the dotcom era say this is like 10x more ferocious and intense than that was. In many ways, I love that. I mean, I I live for competition...stepping back uh I understand I hope everyone understands that there's a much bigger thing at stake than just you know company successes”

Current priorities for root node problem solving include room-temperature superconductors, better batteries, advanced materials, fusion energy, and quantum error correction.

“I think material science uh I'd love to do a room temperature superconductor um and uh you know better batteries these kinds of things. I think that's that's on the cards. uh better materials of all sorts. We're also working on fusion”

Commonwealth Fusion is probably the best startup working on traditional TOKAMAK reactors and closest to having a viable fusion solution, and DeepMind is partnering to help contain plasma in magnets and with material design.

“we've just announced partnership with a deep one. We we already were collaborating with them, but it's a much deeper one now with Commonwealth Fusion who, you know, I think are probably the best startup uh uh working on at least traditional TOKAC uh reactors. So they're probably closest to to having something uh uh viable and we want to help accelerate that uh you know helping them contain the plasma in the magnets and maybe even some material design there as well.”

AI product integration into existing Google ecosystem (search with AI overviews, workspace, email, YouTube, Chrome, Gemini app) represents low-hanging fruit with immediate value, demonstrating how AI can power existing services without waiting for new products.

“We're looking at workspace at email, you know, at YouTube. So, there's all these amazing things in Chrome. There's a lot of these amazing things that um AI we can see already are lowhanging fruit to apply uh Gemini 2 as well of course as Gemini app which is doing really well...we don't have to rely on that. we can just power up our existing uh ecosystem.”