Apr 7, 2026 · 32m · 20vc
Demis Hassabis: Why AGI is Bigger than the Industrial Revolution & Where Are The Bottlenecks in AI · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20VC, Google DeepMind co-founder Demis Hassabis sits down with Harry Stebbings to discuss the timeline, technical bottlenecks, and organizational shifts on the path to Artificial General Intelligence (AGI). Hassabis explores how AGI will transform global healthcare, the economy, and energy systems, while emphasizing the critical need for safety regulations and proactive redistribution strategies to manage labor displacement.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 22.3% of the talking time here. How this is scored →
Harry as informed peer 3.5 Guest teaching 3.3 Guest disagreement 1.2 Harry pushing back 1.9
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 20:31 Rejecting standard Silicon Valley optimism on job disruption
Demis directly rejects Marc Andreessen's dismissal of job losses, arguing that AGI is ten times the magnitude of the Industrial Revolution at ten times the speed, requiring proactive downside mitigation.
Hardest push from Harry ▶ 20:00 Challenging Demis with Andreessen's anti-displacement viewHarry forcefully brings up Marc Andreessen calling job loss fears 'Marxist rubbish' to test Demis's thesis on labor displacement.
Biggest teaching moment ▶ 3:55 Deconstructing the scaling law plateau narrativeDemis corrects Harry's assertion that scaling laws are plateauing, explaining how diminishing exponential leaps still yield massive, highly valuable returns.
Harry holds his own ▶ 10:00 Detailing startup benchmarking and cost-optimization workflowsHarry shows strong operational expertise into portfolio company dynamics, describing how founders use frontier models for benchmarking before switching to cost-effective open models.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Podcast Trailer and Highlight Compilation | 1 | 2 | 0 | 0 | Harry sets up the interview praising Demis and asks about AGI definitions. Demis explains DeepMind's consistent human-brain benchmark for AGI. | |
| Compute Bottlenecks and Scaling Laws | 4 | 4 | 2 | 3 | Harry queries whether scaling laws are plateauing, which Demis refutes by offering a nuanced take on sub-exponential but still high returns. Demis then details the biological analogy of sleep consolidation to explain missing continuous learning capabilities. | |
| Assembling the Ingredients: DeepMind's Organizational Shift | 3 | 3 | 1 | 1 | Demis highlights DeepMind's historical contributions to 90% of modern AI breakthroughs and organizational consolidation. Harry agrees on agent failure modes when file paths shift, while Demis points to missing hierarchical planning and memory architectures. | |
| The Future of Open Science and Gemma | 4 | 3 | 1 | 1 | Harry highlights how startup founders benchmark with frontier models but deploy open-source alternatives for cost efficiency. Demis explains that open-source models usually lag six months behind frontier labs and details DeepMind's Gemma strategy. | |
| Foundation Models vs. Complete AI Systems | 4 | 4 | 2 | 2 | Demis pushes back against Yann LeCun's perspective, arguing foundation models will be augmented rather than replaced. When Harry brings up the lengthy clinical trial bottleneck for MS, Demis outlines how AI simulation and backtesting could eventually compress regulatory timelines. | |
| Dual-Purpose Technologies and International Regulation | 3 | 3 | 1 | 3 | Harry brings up Stephen Hawking's warning and notes that international coordination is worsening just when needed most. Demis agrees with the geopolitical friction but details a framework for minimum safety standards and non-deception benchmarks. | |
| Technical Auditing and National Safety Institutes | 3 | 3 | 1 | 1 | Harry asks who can act as the ultimate arbiter of truth given deepfakes and media noise. Demis proposes technical auditing by state AI Safety Institutes and an IAEA-style international oversight body. | |
| The Speed of Labor Market Displacement | 5 | 4 | 3 | 5 | Harry confronts Demis with Marc Andreessen's view that labor displacement concerns are unfounded rubbish. Demis counters that while historical tech shifts created new jobs, AGI represents ten times the impact of the Industrial Revolution at ten times the speed, making current disruption far more severe. | |
| Underappreciated vs. Overhyped Timelines | 4 | 4 | 1 | 2 | Harry asks if AI progress follows Amara's Law and how society will handle energy demands and wealth concentration. Demis distinguishes near-term hyper-hype from long-term underappreciation, asserting that AI will ultimately solve energy grid inefficiencies and enable nuclear fusion. | |
| Why DeepMind Chose London over Silicon Valley | 3 | 3 | 0 | 1 | Harry asks why Demis chose London over Silicon Valley despite relentless pressure. Demis explains that London offered elite university talent with less local competition, and distance from Valley hype allowed deeper, long-term focus. | |
| Overcoming Europe's Growth Capital Gaps | 4 | 3 | 1 | 2 | Harry challenges Demis on why Europe hasn't built a trillion-dollar tech giant and asks how to unlock growth capital. Demis highlights the late-stage funding gap in European venture capital and shares anecdotes about meeting Elon Musk and the philosophical implications of AGI. |
Statements from this episode (34)
Hassabis: Google and DeepMind developed 90% of modern AI breakthroughs
“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.”
Hassabis: Algorithmic innovation will soon drive AI advantages over pure scaling
“Those labs that have Capability to invent new algorithmic ideas are going to start having bigger advantage over the next few years. As the last set of ideas, all the juices being wrung out of them.”
Hassabis: AI optimization could boost national grid efficiency by 30 to 40%
“I think we could probably get 30, 40% more efficiency out of our national grids.”
Hassabis: AGI impact will be 10x the Industrial Revolution at 10x speed
“I sometimes quantify the coming of AGI as 10 times the industrial revolution at 10 times the speed.”
Hassabis: True AGI must match all cognitive capabilities of the human mind
“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 th…”
Hassabis: There is a very good chance AGI arrives within five years
“I mean, I think, look, I've got a probability distribution around the timings, but I would say there's a very good chance of it being within the next five years.”
Hassabis: DeepMind's 2010 prediction of a 20-year AGI timeline remains on track
“We used to do this extrapolation of compute and algorithmic 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.”
Hassabis: Compute is AI's primary bottleneck for both scaling and experimentation
“I think compute is the big one, not just for the obvious reason of scaling up your ideas and your systems as you know, the scaling laws as they're called, you know, keeping on building bigger and bigger architectures with more and more parameters. And as you d…”
Hassabis: Current AI systems lack continual learning capabilities after deployment
“These systems don't learn after you finish training them, after you put them out into the world.”
Hassabis: Long-context windows in AI are a brute-force approach to memory
“At the moment we have these long context windows, which are kind of a bit brute force. You just put everything in them.”
Hassabis: Current AI systems fail at multi-year, long-term hierarchical planning
“These systems are not very good at planning at long time horizons, you know, many years into the future which we is, you know, with our minds we can do.”
Hassabis: Current AI exhibits 'jagged intelligence' with surprising basic failure points
“So, you know, I sometimes call these systems jagged intelligences because they're really amazing at certain things when you pose the question in a certain way. But if you pose a question in a slightly different way, they can actually still fail at quite elemen…”
Hassabis: Open-source AI models will lag frontier proprietary models by six months
“I think increasingly you know, what you're gonna see is the open source models are probably one step back from the absolute frontier. You know, it usually takes about six months for the open source community to sort of re-implement and figure out what those id…”
Hassabis: 50% chance breakthroughs like world models are still needed for AGI
“There's a fifty-fifty chance there's some things maybe missing that we still need to make breakthroughs in, perhaps their world models”
Hassabis: Foundation models will not be replaced, but built upon for AGI
“I don't think it's going to get replaced. I think it's going to get built on top of these foundation models, just like the way we do with our world models.”
Hassabis: AGI will usher in a golden age of scientific discovery
“Well, I think on the positive side and the things obviously I've, I spent my whole career and life building towards AGI is, I think it will be the ultimate tool for science and medicine. So in terms of advancing scientific discovery finding cures to diseases, …”
Hassabis: Isomorphic Labs will have an AI drug design engine in 5-10 years
“I think we'll have that whole Drug design engine ready in, you know, the next five plus five to 10 years, then you're right.”
Hassabis: Medical regulation will transform after a dozen AI drugs complete trials
“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…”
Hassabis: AI systems will become significantly more agentic and autonomous within 1-2 years
“Second issue is a technical one, making sure These systems as they get more powerful, not today's systems, but maybe in a year or two's time when they become more agentic, more autonomous as we get towards AGI can they be kept on the guardrails that we want?”
Hassabis: AI developers must not build systems capable of deception
“And it's not ideal, but I think we're gonna have to try and do the best we can to at least come up with a sort of set of minimum, maybe minimum standards, some benchmarks that test for undesirable properties, for example, deception. You don't, you know, nobody…”
Hassabis: Global AI oversight requires an international body like the IAEA
“Yeah, I think we need some kind of international body, maybe similar to the atomic agency, something like that, That perhaps the AI safety Institute sort of feed into”
Hassabis: Non-human-readable AI output tokens introduce dangerous security vulnerabilities
“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.”
Hassabis: Artificial intelligence technology is currently overhyped for the next year
“Literally today, as of today and in the next year, things are a bit overhyped in AI.”
Hassabis: AI's revolutionary impact over a 10-year horizon remains vastly underappreciated
“I still think it's still very underappreciated how revolutionary this is going to be in the sort of timescale of about 10 years.”
Hassabis: Sovereign wealth funds should buy AI shares to distribute tech wealth
“Maybe pension funds should be buying into all the big AI companies and making sure that everyone has a piece of that. Or sovereign funds. Maybe everyone, every country should have a sovereign wealth fund that does that.”
Hassabis: AI will drive material science and energy breakthroughs within 5-10 years
“I mean, there could be unbelievable things happening in the five to 10 year timescale, including like a breakthrough in some kind of renewable free energy. You know, maybe we sold fusion. We're working on that right with our partners at Commonwealth fusion. I …”
Hassabis: Google DeepMind is partnering with Commonwealth Fusion on nuclear fusion
“We're working on that right with our partners at Commonwealth fusion.”
Hassabis: AI will more than pay for its own energy costs long-term
“AI will in the medium to long run more than pay for itself, I think, in terms of energy costs.”
Hassabis: Founding DeepMind in London provided a structural advantage in talent recruitment
“And I felt that there was actually less competition here for that sort of talent. And we could even draw in the best talent from the top European universities. And that's what it was like in the early days of DeepMind. So I think it was a huge structural advan…”
Hassabis: Distance from Silicon Valley enables the deeper thinking required for deep tech
“And then the final thing is maybe being a bit away from the valley. There is some disadvantage in that. You're not plugged into the network and the gossip and the latest trends and vibes and all of these things. We're a little bit out of it here, but it does, …”
Hassabis: Spotify and Helsing are contenders for Europe's first trillion-dollar company
“Daniel might well get there with one of his companies, you know, Spotify, Helsing. I think those are two good options.”
Hassabis: European capital markets still lack growth-stage funding depth and ambition
“And I think that certainly was missing 10 years ago when I was doing fundraising for deep mind and I think it's still kind of missing today. Just that kind of level of ambition and the amount the capital markets can support.”
Hassabis: Isomorphic Labs' AI drug platform will eventually cover all disease areas
“Well, look, I want to literally cure cancer. I know people said that's the cliche, but I actually, what we're building isomorphic is general purpose. So we're trying to build a platform, a drug design platform that will be applicable to any therapeutic area. S…”
Shorts cut from this episode
▶ Google DeepMind CEO on AI Concerns · 20VC with Harry Stebbin (@15:29)
▶ AGI will be 10x the industrial revolution · 20VC with Harry (@21:21)
▶ Building out of the Valley is a MASSIVE Advantage · 20VC wit (@26:59)
▶ We're entering a new golden era · 20VC with Harry Stebbings (@12:45)
▶ Google DeepMind CEO on Building in London & Europe · 20VC wi (@25:58)