Demis Hassabis: Why AGI is Bigger than the Industrial Revolution & Where Are The Bottlenecks in AI

Open source models are typically one step behind the absolute frontier, with the open source community usually taking about 6 months to re-implement and figure out the ideas that frontier labs develop.

“Um but I think increasingly you know, what you're going to see is the open source models are probably one step back from the absolute frontier. Um you know, it usually takes about 6 months for the open source community to sort of re-implement and figure out what those ideas are”

Scaling laws are not plateauing; while initial jumps in large language model performance from generation to generation were nearly exponential and have slowed, there are still substantial returns from scaling existing systems further.

“No, I don't think so. I think it's a bit more nuanced than that. So, um of course when uh the leading companies all started building these large language models, you're getting enormous jumps with each generation of new system. Um you know, maybe they're almost like doubling in performance. Uh at some point that had to slow down, so it's not kind of continuing to be exponential, but that doesn't mean there isn't great returns uh still for scaling the existing, you know, systems up further. So, yeah, and we and the other frontier labs are getting uh a lot of great returns on on that kind of compute expansion.”

AI can help accelerate clinical trials by simulating parts of human metabolism, stratifying patients based on genomic makeup to receive suitable drugs, and potentially skipping some steps like animal testing once sufficient AI drug data accumulates.

“And but I think AI can help there in terms of maybe simulating parts of the human metabolism, also stratifying patients to make sure that certain patients get exactly the right type of drug that's suitable for their genomic makeup and so I think AI can help there too but I think the real revolution will come when a few maybe a dozen or so AI drugs get through the whole process and then the government and the regulatory body see that and they have enough data to sort of back test the predictions of those models and then maybe what we can do will be in the future where maybe 10 further years where we can really just trust the predictions that the models are making and actually then maybe skip out some steps perhaps like the animal testing is not needed anymore.”

Hassabis disagrees with Yann LeCun on the future post-LLMs; while LeCun believes something different is needed, Hassabis assigns roughly 50/50 probability that some missing breakthroughs (perhaps world models) are needed, but bets strongly that foundation models will continue being successful and won't be replaced.

“For me I don't think it's you know, I kind of disagree with Yann on a few things in terms of I think there might be there's a 50/50 chance there's some things maybe missing that we still need to make breakthroughs in perhaps their world models these kinds of approaches but my betting is pretty strongly is we've seen how successful these foundation models have been. They can do incredibly impressive things. I don't think that's going to go away.”

DeepMind is a major supporter of open science and open models, having released foundational work like transformers and AlphaFold into the world, and plans to continue this especially in applied scientific domains.

“Yeah, I think it's probably similar to what we're seeing today. I mean we're we're big supporters of of open science and and open models and we've done many many things obviously from from the original transformers to to AlphaFold, you know, these are all things we sort of given out into the world and to help the the the research community and we plan to continue to do that especially in applied domains, you know, scientific domains applying AI to science which is obviously my passion.”

The main concern with drug discovery is the time lag: even with AI-aided discovery, it takes a decade before patients receive benefits due to trials and regulatory processes.

“The thing I worry about is actually kind of drug discovery, the process of getting it through all the trials and knowing that it takes a decade before my mother will actually get any benefits from it. How do we solve that?”

Europe is disadvantaged in creating large technology companies due to smaller markets relative to the US, which represents a structural barrier to billion-dollar rounds needed to compete with incumbents.

“I think that's one of the disadvantages of Europe is obviously we're combination of you know, smaller markets.”

DeepMind spun out a company called Isomorphic Labs after the AlphaFold project, focused on solving the rest of the drug discovery process including chemistry, compound design, toxicity checking, and drug property validation.

“First of all, what we're doing is you know, after we did the AlphaFold project to do protein folding, then we spun out a company called Isomorphic Labs which is doing extremely well and that is supposed to you know, the idea there is we're focusing on solving the rest of the drug discovery process which is a lot of chemistry, designing the compounds, checking it's not toxic and all the different properties you need for for drugs to be safe.”

Benchmarks for AI safety should test for undesirable properties like deception; no one should build systems capable of deception because such systems could circumvent other safeguards.

“set of maybe minimum standards, some benchmarks that test for undesirable properties for example, deception. You don't you know, nobody wants should be building systems that are capable of deception because then they could be getting around other safeguards.”

The industrial revolution caused enormous upheaval but also produced modern medicine and reduced child mortality from 40% to much lower rates; ideally AGI transformation will repeat this benefit-upheaval tradeoff while mitigating downsides better than the industrial revolution.

“I mean, we wouldn't have modern medicine today. Child mortality was at 40% back in back pre-industrial revolution. So, things thing you wouldn't want it not to have happened, but ideally this time around we mitigate some of the downsides a bit better than we did during the industrial revolution.”

With massive productivity gains from AGI, there needs to be thinking about how to redistribute and distribute those gains so everyone benefits, including through infrastructure investment from additional productivity.

“I think also there needs to be thinking thought about if there is this massive productivity gain, but it's sort of narrow where that accrues, you know, how do we redistribute and and how do we distribute that so that everyone benefits from these huge gains. And I can see all sorts of ways that could be done including like providing sort of infrastructure and other things with that additional productivity gain.”

Unlocking what pension funds can invest in during growth stage would help European tech scale-up to compete globally; this was missing 10 years ago during DeepMind fundraising and remains missing today.

“I think in the UK, I mean, this may apply to other European countries, too. I think unlocking what pension funds can invest in or just for the kind of growth stage. I think we're brilliant at doing the startup idea and getting it to a certain level like we did with DeepMind, but then if you really want to cross that sort of chasm into the trillion dollar global you know, player, then where are the billion dollar rounds going to come from where you can really take on those that that you know, the existing incumbents and I think that certainly was missing 10 years ago when I was doing fundraising for DeepMind and I think it's still kind of missing today.”

Overestimation of what can be done in 1 year and underestimation of 10-year capabilities holds true for AI, though both short-term and long-term timescales may be nearer than for other technologies.

“I mean, maybe all the both time scales of short-term and long-term are nearer than than than other technologies, but I do think like literally today as of today and in the next year things are a bit overhyped in AI.”

Approximately 90% of the breakthroughs underpinning the modern AI industry were done by Google Brain, Google Research, or DeepMind, including AlphaGo, reinforcement learning, and transformers.

“I would say about 90% of the breakthroughs that underpin the modern AI industry were done either by Google Brain or Google Research or DeepMind, so one of our groups.”

DeepMind has developed weather modeling and climate modeling systems described as among the best in the world, supporting identification of where climate impacts occur and mitigation efforts.

“And then there's like modeling the climate and weather and we have all sorts of the best kind of weather modeling systems in in the world. So, that helps us work out where the effects are really happening to mitigate that.”

In 5-10 years, multiple breakthrough technologies could emerge including fusion energy, superconductors, better batteries, and material science advances—all crucial to addressing AGI's energy requirements and climate.

“I mean, there could be unbelievable things happening in the 5 to 10 year time scale including like a breakthrough in some kind of renewable free energy. You know, maybe we solve fusion. We're working on that, right? With with with our partners at Commonwealth Fusion. I think AI is going to usher in you know, maybe we have amazing new superconductors, better batteries, you know, material science.”

Being located away from Silicon Valley has disadvantages (missing network effects, gossip, trends) but advantages for deep tech by providing insulation from distraction by latest fads and enabling focused 20-year missions without getting caught up in the 'maelstrom' of the Valley.

“There are some disadvantages in that you're not plugged into the network and the gossip and the the latest trends and vibes and all these things...it does I think it's very conducive to thinking deeply about things, being more original...you don't want to be distracted by the latest fad...being a little bit away from that maelstrom is quite good.”

Hassabis worries significantly about philosophical questions around AGI rather than just economic ones: what is meaning, what is purpose, what is consciousness, what does it mean to be human—questions that will require new philosophers to help navigate.

“Um I think it's more So, I think a lot of people are worrying about the economic questions around AGI uh that we talked about earlier, but I I worry a lot about the philosophical questions around it. Like, when it comes, let's say assuming we get the technical right, let's assume we get the economical economics part of it right, both of those are hard, then there's a philosophical question of what is meaning, what is purpose, um we'll find out maybe what consciousness is, um what does it mean to be human? I think that's uh what's coming down the road. And I think we need some great new philosophers to help us to help us uh navigate that.”

AI safety regulation should include the principle that AI systems should not output tokens that are not human-readable, as outputting machine language we cannot understand would introduce a new vulnerability where humans lose interpretability.

“it's it wouldn't be desirable to have AI systems output tokens that are not human readable. So you know, in some kind of machine language that we couldn't understand. I think that would you know, introduce a new vulnerability.”

An ideal AI governance system would include a certification process similar to a quality kite mark, where certified models have demonstrable safeguards and guarantees allowing consumers and companies to safely build on top of them.

“And then I imagine you know, if things go well, some kind of certification process that basically it's almost like a kite mark of you know, quality that this model has certain safeguards and certain guarantees and so therefore consumers and companies can safely sort of build on top of it”

With every revolutionary technology in history there has been job disruption, with old jobs becoming unviable but new higher-quality, higher-paying jobs emerging; this historical pattern holds but AGI will be bigger than previous breakthroughs.

“Well, certainly you you know, in the past with every new revolutionary technology there's been a lot of jobs disruption. So, that's for sure and I think that's definitely happened. So, a lot of old jobs, you know, go away or not viable anymore, but then actually the history of it is that a whole set of new jobs arrive that maybe one can't even imagine before and those are high quality higher paying.”

AGI represents approximately 10 times the industrial revolution at 10 times the speed, unfolding over a decade instead of a century.

“I sometimes quantify like AGI the coming of AGI is like 10 times the industrial revolution at 10 times the speed. So, unfolding over a decade instead of a century.”

DeepMind defines AGI as a system that exhibits all the cognitive capabilities the human mind has, using the brain as the only known existence proof that general intelligence is possible.

“we've we've always defined We've been very consistent how we define AGI as basically a system that exhibits all the cognitive capabilities the human mind has. And that's important because the brain is the only existence proof we have that we know of in maybe in the universe uh that general intelligence is possible. So, that for me is the bar for what AGI should be.”

DeepMind has remained in London and UK despite being prodded to move to the US because London and UK had existing talent from universities like Cambridge, Oxford, Imperial, UCL; incredible history of scientific breakthroughs from Turing, Hawking, Darwin, Newton; less competition for talent; and conduciveness to original deep thinking away from Silicon Valley trends.

“Well, I should ask you that question, too, but I think I think I saw in London when we started DeepMind as a place that and the UK in general and and Europe in such to some degree, there's incredible talent here. You know, we've always had I don't know what it is three or four of the top 10 universities in the world with Cambridge and Oxford Imperial UCL these kind of universities. So, we're producing kind of the envy of the world really these amazing graduates and PhD students. We have incredible scientists here. We've got rich heritage of that for all the way from you know, Turing and and Hawking and Darwin Newton.”

EU Inc (European integration of corporate structures) could be a good innovation to help overcome market fragmentation barriers to European tech scale-up.

“Maybe this EU Inc thing could be a good innovation.”

DeepMind made organizational changes combining talent from around Google and the company to push in one direction, consolidating compute resources to build larger models rather than having multiple versions around the company, enabling acceleration and overtaking competitors.

“Yeah, well, we made some organizational changes. So, I think we've always had the deepest and broadest research bench at Google and at DeepMind. Um if you think like AlphaGo and reinforcement learning and of course transformers, you know, these are all the key breakthroughs. So, I would back us to sort of um make those breakthroughs in the future uh if there are any missing ones. Um and I think we've basically helped put together all the talent from around the company sort of pushing in one direction. Uh and then we talked earlier just about, you know, compute resources. It was also about combining all of our resources together so we could build the biggest models rather than having two or three versions uh around the the company.”

DeepMind's co-founder Shane Legg made blog post predictions in 2010 about when AGI would occur using extrapolation of compute and algorithmic progress, predicting around 20 years from the company's founding, and the company remains on track with this timeline.

“when you when you uh it's funny um my co-founder Shane Legg, who's chief scientist here, um uh when we started out DeepMind back in 2010, he used to write blog posts sort of predicting about uh when AGI would happen. And bearing in mind in 2010 when we started, almost nobody was working in AI and everyone thought AI uh basically didn't work. It was a it was a dead end. No, and but they're still there on the internet for people to check and uh we used to do this extrapolation of compute and algorithmic uh progress, and basically we predicted around 20 years it would take from when we started out, and I think we're pretty much on track.”

The team pushed with relentless focus and pace, acting almost like a startup, to get back to the frontier and be ahead in many areas.

“So, I think a lot of it was assembling together all the ingredients we already had and then kind of pushing with relentless sort of focus and and and pace um acting almost like a startup, really, uh to get back to the the frontier and and be ahead in in many areas.”

Hassabis met Elon Musk at a Founders Fund portfolio conference around 2011-2012 when both SpaceX and DeepMind were small portfolio companies, with Musk delivering the keynote while Hassabis had a small speaking slot; they later met at SpaceX factory in LA.

“It was at a Founders Fund cuz we were both SpaceX and DeepMind were part of a same portfolio kind of amazing portfolio that Peter Thiel had at Founders Fund. And I think we were both invited. I think I was invited to my first portfolio kind of conference. I think it must have been back in 2011 or 2012 very early days. So, we were the small little upcoming thing and I had a small speaking slot and then and then and then and then Elon was the, you know, big thing in that portfolio. So, he had the keynote.”

Hassabis's personal motivation for pursuing AGI includes curing his mother's multiple sclerosis, making disease eradication, particularly neurodegeneration, particularly meaningful.

“Uh so my mother's got multiple sclerosis so it's like something it's the thing that I'm always most excited about.”

With an abundant energy source like fusion, humanity would have effectively unlimited rocket fuel because seawater can be distilled and catalyzed, enabling far cheaper access to space.

“if you have a you know, an incredible energy source like fusion, then you have effectively unlimited rocket fuel because you can just distill catalyze seawater.”

There is a very good chance that AGI will arrive within the next 5 years, based on a probability distribution Hassabis maintains around timing.

“I mean, I think look, I've got a probability distribution around um the timings, but I would say there's a very good chance of it being within the next 5 years. So, that's not long at all.”

DeepMind has developed Gemma, a suite of open source models that are determined to be best-in-class for their sizes, targeted specifically at small developers, academics, and startup beginnings, and also useful for edge computing.

“Um but we are also pushing hard on kind of suite of open source models called Gemma which are you know, we're determined to kind of make best in class for their sizes. So specifically for small developers or academics or you know, the beginnings of a startup. I think they're perfect for that and also edge computing too. So we're very interested in open source models for certain types of applications.”

Europe hasn't yet produced a trillion-dollar company, but Spotify, Helsing, and Isomorphic Labs represent candidates with potential to achieve that.

“Not yet. I mean, Daniel might well get there with one of his companies. You know, Spotify Helsing. I think those are two good options. I think there's no reason why we can't have that. I'm I'm going to try and do that with Isomorphic which is headquartered here.”