AI moves from automating math toward theory heavy fields such as physics and economics

Theory problems become harder to ignore

AI is moving beyond math and into other theory heavy fields. Neil Thompson, who studies progress in computing and artificial intelligence, argued at MIT that probability, statistics theory, economic theory, high energy theory, gravitation and cosmology, and quantum physics may be next. His lab studies how advances in computer science affect society and innovation.

The work with Gabriel Manso and Claudio Mayrink Verdun points to a steep change. Between mid-2024 and mid-2026, the best AI models rose from correctly solving about 2 percent of theoretical problems in these fields to about 35 percent. The researchers saw no signs that the progress would slow.

Problems were checked after training cutoffs

The study used problems stated in research papers posted online between December 2025 and May 2026. It tested OpenAI models including o1, o3, GPT-5, GPT-5.1, GPT-5.4, and GPT-5.5. Solutions appeared in papers published after the training cutoff for GPT-5.5 and the preceding models, so the systems had to work out answers rather than repeat known results.

The researchers also blocked internet access so the models could not look up solutions. They limited the areas to fields where AI could reason without running simulations or experiments. Even so, the authors say the pattern resembles the shock waves now felt in math and may spread to other math and theory heavy areas of inquiry.

The human role shifts toward judgment and verification

OpenAI recently drew attention by saying an unreleased model had solved the long standing Navier-Stokes grand challenge in mathematics, which concerns equations for fluid motion. The announcement became contentious after a mathematician at NYU accused OpenAI of gazumping collaborators after learning of related progress. OpenAI and others later announced hundreds more solutions to longstanding math problems, prompting calls for limits from some mathematicians.

The researchers expect both positive and negative effects. Thompson notes that theoretical advances sometimes drive practical advances in science, so greater AI use may speed that path. Some AI researchers have founded labs focused on AI for science, and large AI companies are increasing scientific research efforts that pair analysis with real experiments.

Experts remain important. Manso said people are still needed to generate good ideas and connect the dots. He added that complex problems may still lead AI astray, and that claims of the right answer should be checked against the full solution. OpenAI also withdrew several solutions in its latest dataset after publication, underscoring the need for verification.

Sources: Wired