MIT researcher Neil Thompson said AI models are set to transform scientific theories, including mathematics, statistics, economics, and physics. The models' ability to solve theoretical problems has increased from 2 percent in mid-2024 to 35 percent by mid-2026.
Thompson's study, conducted with Gabriel Manso and Claudio Mayrink Verdun, tested OpenAI models like o1, o3, and GPT-5.5 on problems from research papers published between December 2025 and May 2026. The models had to solve these problems without relying on training data or internet access.
The study focused on fields where AI models could figure things out without simulations or experiments. The results suggest that the shock waves currently felt in mathematics are likely to spread to other theory-heavy scientific areas.
"Within just a few years, we expect a lot of theory questions to be totally transformed in terms of AI's ability to do them," said Thompson. He noted that theoretical advances can drive practical progress in science and that AI may help accelerate this process.
Last month, OpenAI made headlines by announcing an unreleased model had solved the Navier-Stokes equations, a long-standing math challenge. This sparked controversy after a mathematician accused OpenAI of gazumping them. OpenAI and others have since announced hundreds more solutions, causing a crisis in the field.
Thompson said AI advances will have both positive and negative effects. He emphasized that experts will still be needed to verify solutions and ensure their accuracy.
"We also need ways to verify what was done, because it's very easy for a model to tell you ‘this is the right answer,’ and when you look at the entire solution, you start to see maybe it is not," said Manso.
Source: wired
