Professor Calls for Using AI to Find Simpler Proofs of Old Theorems

University of Toronto professor Gautam Kamath took to Twitter on October 6 with a thread on a blind spot in AI for math: researchers are eager to use AI to produce new results, yet almost no one is using AI to find shorter, more elegant proofs of existing theorems. His proposed remedy: a dedicated publication channel for "book proofs" to change the status quo.

Confirmed

• Kamath attributes this neglect to publication culture: glory goes almost entirely to whoever first obtains a result, while simplification work rarely gets published. He cites Dinur's simplification of the PCP theorem as a rare notable exception.
• Drawing on his own experience, he notes that current AI models, even when "dehydrated," are not good at producing human-readable or elegant solutions; they tend to keep piling technical modifications onto prior technical approaches, becoming increasingly hard to grasp intuitively.
• He warns of a compounding problem: as multiple AI-driven works stack on top of one another, the complete solution may become incomprehensible—no single paper makes a glaring error, but confusion accumulates, potentially burying far more elegant solutions that may exist.
• He notes the math community's recent emphasis on "human understanding," suggesting this may be the right moment to revisit whether proofs of known results are clean and accessible, and proposes a dedicated "book proofs" publication channel to encourage simplified proofs of known results.
• Blogger srchvrs strongly agreed, saying science has accumulated massive "scientific debt"—many older results have questionable validity.

Why it matters

• As AI-assisted proofs multiply, the readability and comprehensibility of arguments is becoming a systemic risk that goes overlooked; without steering the community toward simplification work, the mathematical literature could become opaque and hard to navigate at an accelerating pace.
• The "book proofs" channel proposal points to a concrete incentive reform, aiming to reverse a publication culture that rewards only the "first prover."

2026-10-06 ~ 2026-10-06 · 8 related posts

• [source] Everyone uses AI for new math results; Kamath wants AI to simplify old proofs — thegautamkamath · 2026-10-06
• Gautam Kamath: math community's 'human understanding' push makes AI proofs worth revisiting — thegautamkamath · 2026-10-06
• Kamath blames publication culture for the lack of AI-driven proof simplification — thegautamkamath · 2026-10-06
• Kamath: current AI rarely produces elegant, human-readable math solutions — thegautamkamath · 2026-10-06
• Kamath: stacked AI-driven works could make entire solutions inscrutable — thegautamkamath · 2026-10-06
• Mathematician Proposes a "Book Proof" Venue as AI-Assisted Papers Stack Into Inscrutability — thegautamkamath · 2026-10-06
• [source] Researchers Call for AI to Clean Up Math Literature's Scientific Debt — srchvrs · 2026-10-06

1 near-duplicate retellings: thegautamkamath