AI summary by TwinMind
AI Could Make Quantum Computing Obsolete, Nobel Prize Winner Says
Sabine Hossenfelder
8m 39s
Hassabis Interview and Main Thesis
- Demis Hassabis (DeepMind, Nobel Prize chemistry winner) believes AI will eat up many quantum computing use cases
- Potentially devastating news for quantum computing startups
- Speaker's controversial take: classical Turing machines capable of more than previously thought
- Classical systems may model quantum systems
- Discussed with Professor Zinger (Nobel Prize physics, quantum computing pioneer)
- David Deutsch called it "crazy but the right sort of crazy"
Quantum Computing vs Classical Computing Performance
- AI as quantum computing competitor isn't new concept
- Quantum advantage comes from algorithmic scaling, not raw speed
- Quantum mechanics provides larger space of possibilities through entanglement
- Fewer operations needed for some problems as they scale
- Speed comparison:
- Supercomputer clusters: 10^18 operations/second (exaflop range)
- Google's Sycamore quantum computer: ~10 million operations/second (best case, if coherent)
- Quantum computers have huge starting disadvantage due to slow operations
- Should catch up eventually, timing depends on classical computer/algorithm improvements
AI Success Cases: AlphaFold and AlphaGo
- Both systems tackled problems with impossibly large possibility spaces
- Protein folding example: 100 amino acid molecule has >10^40 folding possibilities
- Success because physical rules constrain the space
- AI learned these rules from data, vastly reducing search space
- Similar principle applied to Go game
Quantum Chemistry Applications
- Major quantum computing interest area
- Goal: calculate molecular/material properties without physical production
- Applications: drug development, battery technologies, catalysts
- Commercially very interesting
- Quantum computers believed useful because chemistry is quantum physics (electron bands around nuclei)
AI Competing in Quantum Chemistry
- Recent Science paper (last month) shows AI filling this niche rapidly
- Many molecules/materials are "quantum but not a lot"
- Don't need full quantum computer possibility space
- Jeppe Cario (computational physics professor, Switzerland, Science paper co-author):
- Quantum computers unlikely to provide advantage for most chemistry/material science problems
- MIT Technology Review interview confirms this trend
Speculative Implications
- Two interpretations of Hassabis's statement:
- Certain natural systems have physical rules making problems digestible for clever AI
- More speculative: All natural systems (including large quantum computers) may be more rule-determined than thought
- If true, could teach something deep about quantum mechanics
- Risk: AI making quantum computers superfluous before we discover these rules
Personal Reflection and Conclusion
- Speaker's reaction: "Why haven't I thought of this" - explaining Nobel Prize vs YouTube distinction
- Acknowledges Hassabis's insight while maintaining humor about platform differences
