DiscussionTechnical

Humans vs AI: Can AI develop math we can't explain? | Lex Fridman

Lex Clips

The discussion explores whether AI can develop mathematics humans cannot understand, concluding that incomprehensibility would stem from scale rather than fundamentally alien reasoning. The speakers argue that mathematics should prioritize storytelling and understanding over mere theorem accumulation, with formal proofs serving as verification rather than the primary vehicle for insight.

Summary

The conversation begins with a thought experiment about AI-generated proofs that world experts cannot understand. One speaker argues that within formal mathematical systems with fixed axioms and inference rules, incomprehensibility can only arise from scale—the sheer length or complexity of proofs—not from genuinely alien mathematical reasoning.

A contrarian perspective is introduced about "unformalizing" mathematics. The speaker notes that while formalization builds confidence in correctness, human understanding actually operates at an unformalized level, focusing on interesting special cases, techniques, and intuition. If AI produces black box proofs, the response would be to ask simpler questions and develop understanding through dialogue rather than accept the proof as final.

The discussion then shifts to the distinction between proof "engineering" and proof "communication." A real analysis instructor shares how traditional textbooks spend weeks formalizing real numbers before doing calculus, even though students intuitively understand them. The instructor redesigned their course to jump directly into epsilon-delta calculus, only introducing formal constructions when students encountered problems that demanded deeper understanding—mirroring how mathematics actually develops historically.

This leads to a vision for future mathematical writing: papers should tell the story of discovery, explaining why certain paths were chosen over others, with formal Lean proofs accompanying the narrative as verification. Mathematics should not be merely a rising tide of theorems, but rather a system of beliefs and abstractions discovered through intellectual struggle with problems.

The final point emphasizes that mathematical progress is fundamentally about the process of fighting with problems and learning from them, not just accumulating results. Understanding and storytelling are now more important than ever as AI generates mathematical content.

Key Insights

  • Within formal mathematical systems with fixed axioms and inference rules, the only way incomprehensible mathematics could arise is through scale and complexity, not through fundamentally alien ideas.
  • Human mathematical understanding operates at an unformalized level focused on intuition, special cases, and techniques, whereas formal proofs serve engineering purposes before being communicated.
  • Traditional mathematics pedagogy frontloads formal construction (like defining real numbers via Cauchy sequences) for engineering reasons, but students understand intuitively first and should only learn formalization when their intuition fails.
  • Future mathematical writing should separate storytelling about discovery and reasoning from formal verification, using Lean formalizations as companion proofs rather than the primary exposition.
  • Mathematics is fundamentally about fighting with problems and discovering broader systems of meaning and abstraction, not simply accumulating theorems; this makes understanding and storytelling more critical when AI contributes to mathematical progress.

Topics

Black box proofs and AI-generated mathematicsFormalization vs. intuitive understandingMathematics as narrative and storytellingEngineering proofs vs. communicating mathematical ideasThe role of struggle and problem-solving in mathematical understanding

Transcript

[0:03] - So this is where I'll just take the tangent, and the thought experiment may be a bit philosophical at this moment, but what if you have that black box that gives you proofs that you can't understand? Let's say the more interesting case where, not because it's so big, but because the ideas just don't make sense to you, to any of the world expert mathematicians in that field. What do you do with that? - So the only way, sort of almost by definition, because the logical rules of inference are written down, and there, there's [0:34] only a few of them. There's a handful of them. The axioms, there's a handful of them. So the only…

Full transcript available for MurmurCast members

Sign Up to Access

More from Lex Clips

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.