Back to all articles
•Technology

AI Models Make Strides in Tackling the Riemann Hypothesis, Sparking Debate in the Mathematical Community

View original source

The Riemann hypothesis, an unsolved enigma regarding the distribution of prime numbers, has both fascinated and perplexed mathematicians for over 150 years. A $1 million bounty remains for its proof, which is famously unclaimed. Despite contemporary AI models being unable to fully resolve it, recent advancements suggest increased potential for AI in mathematical discovery.

  • On Monday, Anthropic reported significant progress on the Riemann hypothesis by their yet-unreleased AI model.
    • It notably enhanced the confirmed lower bound of the hypothesis's solutions.
    • Progress involved testing 650 ideas and coordinating 60 subagents, investing 31 million output tokens.
    • Only two subagents innovated key mathematical ideas, with roles distributed among concept contributors, validators, and writers.
  • Validation came from Anthropic's own mathematicians and the proof assistant Lean.
  • Historically, AI has shown prowess in similar mathematical feats, solving Erdos problems and contributing to disproving the Jacobian conjecture.

The breakthrough has revived discussions about AI's role in math. A public declaration by prominent mathematicians expressed concerns over AI possibly compromising the integrity of mathematical proofs, which should be credited to specific individuals. However, a counter-argument, highlighted by Fields Medalist Timothy Gowers, suggests adapting to AI's contributions could enrich the field, akin to how astronomical entities aren't named after discoverers.