Home Lex Fridman Notes
Lex Fridman · 2022-07-01 · 2h 10m

Demis Hassabis: DeepMind - AI, Superintelligence & the Future of Humanity | Lex Fridman Podcast #299

DeepMind's Demis Hassabis on solving intelligence, how AlphaFold cracked protein folding, and using AI to understand the universe itself.

Demis Hassabis: DeepMind - AI, Superintelligence & the Future of Humanity | Lex Fridman Podcast #299
The guest

Demis Hassabis — CEO and co-founder of DeepMind, the AI lab behind AlphaGo, AlphaZero and AlphaFold. A former chess prodigy and games designer turned neuroscientist, he is one of the most influential figures in modern artificial intelligence.

The gist

Lex Fridman talks with Demis Hassabis about DeepMind's mission to 'solve intelligence and then use it to solve everything else.' Hassabis traces his path from child chess master and 1990s games designer to building general learning systems, explaining the progression from AlphaGo to AlphaZero to MuZero and Gato. The bulk of the conversation centers on AI for science: how AlphaFold 2 solved the 50-year protein-folding problem and was open-sourced to half a million biologists, plus DeepMind's work on nuclear fusion plasma control and quantum chemistry. The discussion then widens into deep questions about consciousness, sentience in language models, the simulation hypothesis, alien civilizations, the corrupting nature of power, and the meaning of life.

Big reveals

  • AlphaFold 2 predicts a protein's 3D structure in seconds, a task that previously took a PhD student their entire PhD for a single protein.
  • Hassabis recounts that in 2010 nobody believed in AI; professors thought he was 'mad' and they hid the 'solve intelligence' mission to avoid eye-rolls.
  • AlphaFold has been used by over 500,000 researchers, which he estimates is essentially every professional biologist in the world.
  • Hassabis states his personal belief that we are alone in the universe, arguing we should have heard a 'cacophony of voices' but heard nothing.
  • He says none of today's AI systems have 'one iota' of consciousness and dismisses the Google LaMDA-sentience claim as premature projection.
  • He advocates building AI as tools first, not conscious entities, until we understand them well enough to deploy safely.
  • Asked the one question he'd put to a superintelligent AGI, he answers: 'What is the true nature of reality?'

Things worth remembering

  • Hassabis bought his first chess computer, a ZX Spectrum, at age eight using winnings from a chess competition.
  • At age 12 he was the second-highest-rated chess player in the world for his age, behind Judit Polgar.
  • Kasparov called chess 'the drosophila of intelligence'; Turing once ran his own chess program by hand with pencil and paper.
  • Levinthal's paradox: an average protein could fold in roughly 10^300 ways, yet nature solves it in milliseconds.
  • DeepMind's fusion work controlled plasma hotter than the sun (~million degrees C) by adjusting magnetic fields within milliseconds.
  • Von Neumann probes could in theory populate every star system in the galaxy within about a million years.
  • The human brain uses about 20% of the body's energy; high-level chess burns as much as a Formula One driver in a race.
  • Hassabis is a night owl who works a 'second day' from ~10pm to 4-5am for deep thinking, reading, and research writing.
  • He marvels that computers can be made of sand (silicon), one of Earth's most common elements, rather than something rare like diamond.

Recommended in this episode

Books, products and media the guest or host genuinely endorsed here — with the buy link.

Affiliate link — we may earn a commission at no extra cost to you.

Guest’s ownMedia

Theme Park

Bullfrog Productions (inferred)

“writing a game called theme park which had ai as a core gameplay component as part of the simulation and it sold millions of copies around the world” — Demis Hassabis 00:13:26
Find it on Amazon
Guest’s ownMedia

Black & White

Lionhead Studios (inferred)

“there was a game called black and white which was one game i was involved with in the early stages of which i still think is the most impressive example of reinforcement learning in a computer game” — Demis Hassabis 00:15:28
Find it on Amazon
Guest’s ownMedia

Republic: The Revolution

Elixir Studios (inferred)

“with games like republic i tried to have games where we designed whole cities and allowed you to play in” — Demis Hassabis 01:28:43
Find it on Amazon
RecommendedBook

The Ten Great Inventions of Evolution

Nick Lane

“there's a great book called the 10 grand great inventions of evolution by nick lane and he speculates on 10 of these you know what could be great filters” — Demis Hassabis 01:36:03
Find it on Amazon