Maths Meets AI With a $15 Million Question
Two Renaissance mathematicians debate geometric proofs on parchment and a modern tablet, balancing excitement with doubt. Made by Edward Frank Morris
The Simons Foundation has announced a $15 million funding and resource initiative with OpenAI and Anthropic to help mathematicians evaluate AI-assisted research and develop shared standards for using AI in mathematics. The programme is expected to support proposals, workshops and collaboration. OpenAI and Anthropic will provide tools and computing resources, while researchers are meant to help shape how the work is assessed. The aim is not to declare machine-generated proofs correct by applause. It is to give mathematicians ways to test them.
Use ChatGPT to interrogate a mathematical claim
When a paper or announcement makes a proof-related claim, give ChatGPT the exact text and ask it to extract the theorem, assumptions, method and stated verification. Then ask for a list of steps that a human mathematician or formal proof assistant would need to check. Keep the model’s output labelled as a reading aid. It is not a certificate of proof.
A research prompt that keeps its feet on the ground
Try: “From this source, separate the initiative’s confirmed commitments from its expected activities. Then propose three research questions about evaluating AI-assisted mathematics. For each, identify what data would be needed, how a result could be independently checked, and what would count as failure. Do not invent grant deadlines or eligibility rules.”
Turn a headline into a research plan
Researchers can ask ChatGPT to compare evaluation approaches, draft workshop questions or translate a technical abstract for a mixed audience. Mathematicians should retain control of definitions, proof standards and final judgement. Funding can buy compute, meetings and time. It cannot buy certainty. The useful question is whether the initiative helps the community create better ways to examine AI-assisted results, including the awkward ones that do not work.
Comments