#475 – Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games

Original Media · 2h 35m · English · Podcast · Lex Fridman Podcast

TLDR

Demis Hassabis discusses his conjecture that natural patterns are efficiently modelable by classical algorithms, the path to AGI, AlphaFold's impact on biology, video generation's intuitive physics, energy futures, consciousness, and the societal transformation AI will bring. He emphasizes cautious optimism, scientific collaboration, and the need for new governance to manage AI's 100x faster-than-industrial-revolution impact.

For you

PRO members can sign in and request more information — or their own version of this digest, tuned to what they care about.

Sign in to unlock

Surprising Insights

•

VO3's realistic physics emerges from watching YouTube videos, not from embodied interaction or explicit physics engines.

•

The origin of life might be simulatable as a search process in chemical space, with AI potentially discovering a 'move 37' for abiogenesis.

•

Demis's post-AGI dream projects are designing a video game and working on a fundamental physics theory.

•

The conjecture that any natural pattern is efficiently modelable because nature is not random but shaped by selection pressures over deep time.

•

Demis believes the P vs NP problem is a physics question if the universe is fundamentally informational.

•

AlphaEvolve combines LLMs with evolutionary search to discover algorithms, a hybrid approach that may overcome traditional evolutionary computing's inability to generate novel emergent properties.

•

Human adaptability is almost limitless — we thrive in a modern world with hunter-gatherer brains, suggesting we can adapt to AI-driven change.

Key takeaways

•

Natural systems have evolved structure that makes them efficiently modelable by classical learning algorithms, as demonstrated by AlphaFold and AlphaGo.

•

The universe may be fundamentally informational, making P vs NP a physics question; a new complexity class for 'learnable natural systems' might exist.

•

AI video models like VO3 learn intuitive physics from passive observation, challenging the notion that embodied interaction is necessary for physical understanding.

•

AGI could arrive around 2030 (50% chance), defined as matching all human cognitive functions consistently, with lighthouse moments like novel scientific conjectures.

•

Recursive self-improvement via systems like AlphaEvolve (LLM-guided evolution) may accelerate AI progress, but big architectural leaps (like Transformers) remain uncertain.

•

Energy breakthroughs (fusion, advanced solar) could enable a post-scarcity civilization within 100 years, solving water, food, and space access.

•

Societal adaptation to AI will be 100x faster than the Industrial Revolution, requiring new economic policies (universal basic provision) and governance.

•

Consciousness likely arises from classical computation; neural interfaces may eventually let humans experience silicon-based computation.

Notable quotes

“I sometimes call it survival of the stabilist or something like that, because, you know, it's, it's, of course, there's evolution for life, living things. But there's also, you know, if you think about geological time, so the shape of mountains, that's been shaped by weathering processes, right over 1000s of years, but then you can even take it cosmological, the orbits of planets, the shapes of asteroids, these have all been survived kind of processes that have acted on them many, many times.”

▸ Notable quotes · 10:37

“I think the universe as a kind of informational system. So when you think of the universe as an informational system, then the P equals MP question is a physics question.”

▸ Notable quotes · 13:39

“I think to the extent that it can predict the next frames, you know, in a coherent way, that's some, that is a form, you know, of understanding, right?”

▸ Notable quotes · 21:59

“I think we're going to have to kind of confront the question again of what is the fundamental nature of reality? Uh, what is there going to be the difference between these increasingly realistic simulations and, uh, multiplayer ones and emergent, um, and what we do in the real world?”

▸ Notable quotes · 34:15

“I think the people that are real, uh, believers in the mission of AGI and what it can do and understand the real consequences, both good and bad from that. And what's what that responsibility entails. I think they're mostly doing it to be like myself to be on the frontier of that research.”

▸ Notable quotes · 107:03

“I think it's going to be 10 times at least of the industrial revolution so I think he's right so I think he would have been i imagine fascinated by uh where we are now”

▸ Notable quotes · 119:17

“I think the best way of course one can kind of philosophize about it and have thought experiments and maybe even do actual experiments like you do in neuroscience on on real brains but in the end i always imagined that building ai a kind of intelligent artifact and then comparing that to the human mind and seeing what the differences were would be the best way to uncover what's special about the human mind”

▸ Notable quotes · 130:59

“I think it's almost limitless what we can potentially do with them. And then the other thing is our extreme adaptability.”

▸ Notable quotes · 136:38

Chapters

•

8:30 — Nobel Lecture Conjecture: Nature's Patterns are Efficiently Modelable: Demis discusses his conjecture that any natural pattern can be efficiently modeled by classical learning algorithms, using AlphaGo and AlphaFold as examples.

•

10:00 — P vs NP, Information as Fundamental, Complexity Classes: Discussion on P vs NP, the idea of a new complexity class for learnable natural systems, and the view of the universe as an informational system.

•

17:00 — VO3 Video Generation and Intuitive Physics: Demis explains how VO3's realistic physics simulation suggests AI can learn intuitive physics from passive observation, challenging embodied AI assumptions.

•

25:00 — Video Games, Open Worlds, and Post-AGI Dreams: Demis shares his passion for games, vision for AI-generated infinite open worlds, and post-AGI projects: game design and physics theory.

•

37:00 — AlphaEvolve: Evolutionary Algorithms Guided by LLMs: Description of AlphaEvolve, combining LLMs with evolutionary search to discover algorithms, and potential for hybrid AI systems.

•

47:00 — Virtual Cell, Origin of Life, and Modeling Biology: Demis outlines the path to simulating a whole cell (starting with yeast), and the possibility of simulating the origin of life.

•

54:00 — Consciousness, Life Definition, and Nature of Reality: Philosophical discussion on the hard problem of consciousness, the continuum from non-living to living, and the mystery of existence.

•

58:00 — AGI Timeline, Definition, and Testing: Demis estimates 50% chance of AGI by 2030, defines AGI as matching human cognitive functions, and proposes testing via expert red-teaming and lighthouse moments like novel conjectures.

•

66:00 — Recursive Self-Improvement, Scaling Laws, and Compute: Discussion on AlphaEvolve's recursive improvement, scaling laws (pre-training, post-training, inference), and the role of compute and energy.

•

72:00 — Energy Future: Fusion, Solar, and Kardashev Scale: Demis bets on fusion and solar as primary energy sources, enabling desalination, space travel, and a Type I civilization within 100 years.

•

84:00 — Google DeepMind Leadership, Product Culture, and Shipping: Demis describes merging Google Brain and DeepMind, relentless shipping culture, and balancing research with productization.

•

92:00 — AI Interfaces, Multimodality, and Personalization: Vision for AI-generated personalized interfaces, moving beyond chat boxes to collaborative multimodal interactions.

•

102:00 — Competition, Talent Wars, and International Collaboration: Thoughts on talent competition, collaboration with other lab leaders, and the importance of scientific cooperation across geopolitical lines.

•

109:00 — Impact on Jobs, Programming, and Societal Transition: Demis predicts 10x impact of industrial revolution at 10x speed, need for universal basic provision, and new governance structures.

•

115:00 — Von Neumann, Manhattan Project, and AI Risks: Reflections on von Neumann's foresight, the dual-use nature of AI, and the need for cautious optimism and international safety standards.

•

132:00 — Consciousness: Classical vs Quantum, Penrose, and Neural Interfaces: Demis disagrees with Penrose's quantum consciousness, believes consciousness is classical computation, and suggests neural interfaces may bridge substrate gap.

•

136:00 — Hope for Humanity: Ingenuity and Adaptability: Demis expresses hope in human ingenuity and adaptability, citing our ability to thrive far beyond ancestral environments.

•

138:00 — Lex's AMA: David Foster Wallace, MIT, Drexel, and Online Attacks: Lex reflects on David Foster Wallace's 'This is Water' speech, clarifies his MIT and Drexel affiliations, and addresses online criticism.

Speakers

•

Lex Fridman — Host

•

Demis Hassabis — Guest

⋮