AI is automating theoretical science

Neil Thompson, who studies computing progress at MIT, presented evidence that AI will soon upend probability, statistics, economics, physics and cosmology.

Dramatic progress in two years

Between mid-two thousand twenty-four and mid-two thousand twenty-six, the best AI models improved from solving about two percent of theoretical problems to about thirty five percent.

Thompson's study tested real solutions

Thompson, Gabriel Manso and Claudio Mayrink Verdun tested OpenAI models o one, o three, GPT-five and later versions on problems from papers published after the models' training cutoff.

Math world already shaken

In September, OpenAI announced an unreleased model had solved Navier-Stokes, a long-standing grand challenge in mathematics concerning fluid motion equations.

Mathematicians worry about the process

Some mathematicians have called for limits on AI use, warning that the process of solving problems is often as useful as having the final answer.

“the process of solving problems is often as useful as having the final answer”

Experts still needed to verify

Gabriel Manso says theorists will remain essential to generate ideas and verify solutions, as AI models can make mistakes and present wrong answers confidently.

“We are still going to need people to generate good ideas to connect the dots”