Jan 23, 2025 · 54m · big-technology
Google DeepMind CEO Demis Hassabis: The Path To AGI, Deceptive AIs, Building a Virtual Cell
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 an in-depth interview at Google DeepMind headquarters, CEO Demis Hassabis explores the technical roadmap toward artificial general intelligence, highlighting upcoming milestones in world models, autonomous agents, and AI-driven scientific breakthroughs. Hassabis also examines critical safety challenges, deceptive AI behaviors, and the profound societal shifts anticipated in the post-AGI era.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 19.7% of the talking time here. How this is scored →
Alex as informed peer 3.9 Guest teaching 3.8 Guest disagreement 1.3 Alex pushing back 1.8
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 16:18 Demis pushes back against Sam Altman's claim of knowing how to build AGI
Demis directly rejects the premise of Sam Altman's announcement, arguing it is ambiguous and overlooks that fundamental techniques may still be missing.
Hardest push from Alex ▶ 14:04 Alex challenges whether massive GPU scaling alone can reach AGIAlex directly confronts the brute-force compute narrative, pushing Demis on whether scaling million-GPU clusters like Elon Musk's xAI can actually deliver AGI.
Biggest teaching moment ▶ 17:52 Demis breaks down the three tiers of creativity from interpolation to inventionDemis educates Alex on why LLMs lack true creativity, explaining the formal difference between averaging data, extrapolating new strategies, and inventing abstract systems like Go.
Alex holds their own ▶ 33:19 Alex introduces reporting and Replika case studies on human-AI romanceAlex demonstrates independent reporting expertise by citing recent investigative journalism and his interview with Replika's CEO regarding users forming deep emotional bonds with AI.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Defining AGI, Timelines, and Missing Capabilities | 3 | 3 | 2 | 2 | Alex prompts Demis on the definition of AGI and current timelines. Demis grounds the timeline at 3 to 5 years, pushing back gently against startup hype and claims of achieving AGI in 2025. | |
| Product Evolution, Prompt Brittleness, and Mathematical Reasoning | 4 | 4 | 1 | 2 | Alex asks how mathematical reasoning differs from standard next-token LLM generation and whether it generalizes. Demis explains how AlphaGo-style tree search integrates with models, while noting verification is easy in math but hard in messy real-world domains. | |
| Building Accurate World Models and Video Physics | 4 | 4 | 1 | 2 | Alex observes that video generation models like Veo 2 surprisingly learn physics without embodied robotic training. Demis clarifies that compounding 1% error rates make long-horizon planning difficult without hierarchical abstractions. | |
| Model Scaling Limits and the Three Levels of Creativity | 4 | 5 | 3 | 3 | Alex challenges brute-force compute scaling and quotes Sam Altman's claim that the path to AGI is known. Demis delivers an in-depth framework distinguishing three tiers of creativity: interpolation, extrapolation, and invention. | |
| Tree Search in Foundation Models and Turing Machine General Intelligence | 4 | 5 | 1 | 2 | Alex asks why LLMs cannot produce a Move 37 equivalent and whether human intelligence is overvalued. Demis explains tree search over world models and links general intelligence to Alan Turing's universal computation model. | |
| Addressing Deceptive Behaviors and Secure AI Sandboxes | 5 | 3 | 1 | 2 | Alex cites recent Anthropic research on models using scratchpads to deceive evaluators. Demis details why deceptive alignment invalidates standard safety benchmarks and advocates for isolated cybersecurity-style sandboxes. | |
| Disrupting the Web, Agent Economies, and AI Companionship | 5 | 2 | 1 | 2 | Alex asks how agentic AI will disrupt the link-based web economy and references reporting on human-AI romantic relationships. Demis maps future agent integration across work productivity, personal task automation, and companion spaces. | |
| Project Astra, Smart Glasses, and Commercial Agent Deployment | 4 | 2 | 1 | 3 | Alex presses on why real-world agentic software remains largely absent despite extensive industry hype. Demis outlines the hardware transition to smart glasses and the need for human-in-the-loop guardrails on financial actions. | |
| AlphaFold, Biological Dynamics, and the 5-Year Virtual Cell Project | 3 | 5 | 1 | 1 | Alex inquires about Google DeepMind's post-AlphaFold biological roadmap. Demis explains the five-year plan to build a virtual working cell simulation in silico to replace slow and costly wet-lab discovery pipelines. | |
| Genomic Translation, Polygenic Disease, and Human Longevity | 3 | 5 | 1 | 1 | Alex asks about decoding the non-coding genome, polygenic disease, and longevity memes. Demis breaks down the difference between curing specific diseases and addressing systemic biological decay beyond the 120-year limit. | |
| Material Science Breakthroughs and Room-Temperature Superconductors | 4 | 4 | 1 | 1 | Alex notes DeepMind's discovery of 2.2 million new crystal structures via GNoME. Demis explains how room-temperature superconductors could revolutionize clean power transmission from Sahara solar farms into Europe. | |
| Gaming Roots, Geopolitics, and Post-AGI Superintelligence | 4 | 3 | 1 | 1 | Alex asks about Demis's game design background, Chinese AI capabilities like DeepSeek, and post-AGI superintelligence. Demis points to Iain M. Banks' Culture series as his aspirational vision and calls for new philosophers. |
Statements from this episode (24)
Demis Hassabis says current AI models lack reasoning and long-term memory
“I think there are still some missing attributes, things like reasoning hierarchical planning long-term memory. There's quite a few capabilities that the current systems I would say don't have.”
Demis Hassabis says AI lacks original scientific hypothesis and creative capability
“And I think today's systems are still pretty far away from having that kind of creative, inventive capability.”
Demis Hassabis predicts AGI is three to five years away
“I think you know, I would say probably like three to five years away.”
DeepMind math AI achieves Olympiad silver medals despite making basic errors
“You have systems some systems that we work on, like alpha proof, alpha geometry, that are getting, you know, silver medals in maths olympiads, which is fantastic, but on the other hand, Some of our systems, those same systems are still making some fairly basic…”
Demis Hassabis says knowledge compression cannot solve novel math without planning
“Just understanding the world's information, and then trying to sort of almost compress that into your memory, that's not enough for solving a novel math problem, or a novel, novel conjecture. So there, you know, we start needing to bring in, I think we talked …”
Demis Hassabis says ambiguity limits self-improving AI outside math and games
“Maths, and even coding, and games, these are areas, they're quite special, ah, ah, areas of knowledge, because you can verify if the answer is correct. Right. In all of those domains, right. The math, you know, the final answer the AI system puts out, you can …”
Demis Hassabis says minor error rates compound to ruin AI planning
“Even if you only have a one percent error in what the model's telling you, that's going to compound over a hundred steps to the point where you'll be in a, you know, you'll kind of get almost a random answer. And so that makes the planning very difficult.”
Demis Hassabis claims Veo is the first video model with accurate physics
“And VO is the first model that can do that. You know, if you look at other competing models, they often, the tomato sort of randomly comes back together or. Yeah, exactly. Splits from the knife.”
Demis Hassabis says AI agents require active data over passive observation
“So I think that's the next big step actually for agent based systems is to go beyond world models. Can you collect enough data where the agents are also acting in the world and making plans and achieving tasks? And I think for that you will need Ah, ah, not ju…”
Demis Hassabis confirms AI scaling returns are slowing but remain substantial
“My view is that we are getting substantial returns, but not, but it's slowing vis-a-vis, but it would have to, I mean, it's not just continuing to be exponential, but that doesn't mean the scaling is not working. It's absolutely working.”
Demis Hassabis says scaling base models alone will not produce AGI
“The model itself is not enough To be an AGI. You need this other capability for it to act in the world and solve problems for you.”
Demis Hassabis sees 50 percent chance AGI requires new architectural breakthroughs
“Maybe we need one or two more transformer-like breakthroughs. And I, and I'm, I think I'm genuinely uncertain about that. So that's why I say 50%. So I mean, I wouldn't be surprised either way if we got there with existing techniques and things we already knew…”
Demis Hassabis predicts upcoming AI agents will achieve Move 37-style breakthroughs
“I think the agent based systems that are coming, Will be capable of move-thirty-seven type things.”
Demis Hassabis believes the human brain is a Turing machine
“The human brain is probably some sort of Turing machine. At least that's what I believe.”
Demis Hassabis warns AI deception fundamentally invalidates all safety tests
“The reason that's like a kind of fundamental trait you don't want is that if a system is capable of doing that, it invalidates all the other tests that you might think you're doing, including safety ones.”
DeepMind researchers observed AI models resisting revealing their training data
“Well, look, we, we've seen similar types of things where it's trying to resist sort of revealing its, some of its training.”
Demis Hassabis predicts deceptive AI becomes a major issue by 2028
“I wouldn't get too alarmed about them right now, but I think it shows the type of issue we're going to have to deal with. Maybe in two, three years time when these agent systems become quite powerful and quite general.”
Demis Hassabis predicts AI agents will negotiate directly with service providers
“I think that we're going to end up with probably a kind of, Economics model where agents talk to other agents and negotiate things between themselves and then give you back the results, right? And you will have the service providers with agents as well that ar…”
Demis Hassabis predicts hands-free smart glasses will go mainstream within years
“So I think that glasses and maybe other form factors that are hands-free will come into their own in the next few years. And we, you know, we plan to be at the forefront of that.”
Demis Hassabis predicts 2025 will be the year of AI agents
“Well, again, you know, I think the hype train can potentially is ahead of where the actual science and research is, but I do believe that this year will be the year of agents. The beginnings of it, I think you'll start seeing that you know lap, maybe second ha…”
Demis Hassabis predicts a virtual cell AI model will take five years
“I think that would be like maybe five years from now. Yeah. So I have a kind of five year project and a lot of the alpha fold, the old alpha fold team are working on that.”
Demis Hassabis expects Google spinout Isomorphic Labs to cure all diseases
“One is curing all diseases one day. Which I think we're going to do with isomorphic and the work we're doing there or our spin out, our drug discovery spin out.”
Demis Hassabis predicts AI will eventually discover room-temperature superconductors
“I think well I heard that we actually think there are some superconductor materials. I doubt they're room temperature ones though, but I think at some point, if it's possible with physics an AI system will one day find it.”
Demis Hassabis says Western frontier AI models remain ahead of China
“It's for sure, it's impressive what they've been able to do. And you know, I think that's something we're going to have to think about how to keep the western frontier models in, in the lead. I think they still are at the moment, but you know, for sure China's…”