Will AI be better every human at proving Math theorems by the end of 2030?

Ṁ100Ṁ158

2030

50%

chance

ALL

This market resolves YES if the following statement is true:

For a typical nontrivial open mathematical conjecture stated within the standard axioms of mathematics, the probability that a publicly available AI system solves the conjecture autonomously within a fixed time horizon is greater than the probability that a human mathematician solves it within the same time horizon without assistance from AI.

Resolution will be based on my judgment. I am a PhD student in mathematics. I will not place any bets in this market, as the resolution depends on my own judgment.

Clarifications

Scope of conjectures

“Nontrivial open conjecture” means a conjecture that:

is recognized by the research community as open, and

would plausibly merit publication in a peer-reviewed mathematics journal if solved.

Trivial, pedagogical, or adversarially constructed conjectures are excluded.

Which AI systems qualify

Any publicly available AI system qualifies.

The AI must not be a secret or proprietary system whose performance cannot be independently evaluated.

Evidence of the AI’s mathematical capabilities must be public and credible.

Meaning of “autonomous”

An AI is considered autonomous if it produces the core ideas and proof without step-by-step human guidance.

The AI may use standard tools (e.g. proof assistants, code execution, symbolic computation), but not human mathematical reasoning beyond prompt-level task specification.

Human benchmark

“Human mathematician” refers to professional research mathematicians.

Humans may collaborate with other humans and use standard non-AI tools (e.g. papers, textbooks, computers), but may not use AI systems for mathematical reasoning.