The end of maths as we know it? Seismic impact of AI has mathematicians worried

Tech firms are boasting of a future of mathematical solutions with zero human involvement

AI companies are claiming to have cracked some of mathematics longest unsolved problems with zero human involvement
OpenAI, Anthropic and other AI companies are competing directly with mathematicians, rushing to prove open questions. Photograph: Imen Ben Youssef/Hans Lucas/AFP/Getty

Three months ago I wrote about the dramatic progress in Artificial Intelligence (AI) and some of its consequences for mathematics. Since then, startling new accomplishments of AI have been announced on a weekly basis. Clearly, further important open mathematical questions will be resolved in the near future.

The end of maths?

Is a golden age at hand, or do these developments foreshadow the end of mathematics as currently practised? Reactions of research mathematicians vary from keen enthusiasm to deep concern for the future of the discipline.

It is clear that AI is having a massively disruptive effect on mathematics. Teachers and students realise that the old ways must change rapidly. Mathematicians are already grappling with how they should advise research students, and what qualities and qualifications they should require of candidates for open positions.

OpenAI, Anthropic and other AI companies are competing directly with mathematicians, rushing to prove open questions.

READ MORE

Currently, AI proofs can be very difficult to understand and often fail to show how new results depend on or arise from earlier work. Many of the AI proofs are verified by a logic-checking system called Lean, but this is unconvincing unless objectively checked and verified.

Most mathematicians anticipate that AI-generated papers will improve and become more elegant but, for now, they have much work to do disentangling sloppy AI Proofs of important problems.

Disentangling proofs

On October 6th, OpenAI announced proofs of hundreds of outstanding theorems across many different fields of maths. Researchers greeted this deluge with a mixture of delight and alarm. At a recent seminar at UCD, convened by Prof Gary McGuire, several of these AI-generated papers were reviewed by mathematicians with relevant expertise. These papers were of a broad spectrum of quality and significance, some spectacular and some utterly impenetrable and of dubious validity.

One paper of particular interest claimed to prove a weak form of the Riemann Hypothesis, one of the greatest open problems in mathematics, which has resisted solution for more than 160 years.

It concerns the distribution of the prime numbers, the building-blocks of the number system, and is of such interest that the Clay Mathematics Institute has offered a $1 million prize for a full solution. The proof has been carefully checked and is believed to be correct. This is the most substantial mathematical achievement of AI to date and a notable mathematical advance.

[ A golden age of mathematics may be at hand as AI enters the fieldOpens in new window ]

The Navier-Stokes equations are the basis for modelling fluid flow and are of central importance in weather forecasting, aircraft design and many other applications. But we have been unable to show that the solutions of these equation always remain “well behaved” or whether they may “blow up”.

OpenAI has found a solution where the flow rate becomes infinite within a finite time. This result is of profound theoretical interest.

Axiom reasoning

The AI company Axiom is building a reasoning engine – an AI mathematician – capable of discoveries, as it claims, “at unimaginable scales and speed”. Last month, Axiom released a complete computer-verified proof of the Classification of Finite Simple Groups Theorem. This theorem, one of the greatest achievements of 20th-century mathematics, is considered to be the longest mathematical proof ever written by humans, extending over roughly 15,000 pages, written by about 100 mathematicians.

Axiom argues that its reasoning engine presents an unprecedented opportunity, a system that can conjecture and prove theorems with “zero human involvement”. No wonder mathematicians are anxious.

Peter Lynch is emeritus professor at the School of Mathematics and Statistics, University College Dublin. He blogs at thatsmaths.com