At a recent MIT presentation, researcher Neil Thompson argued that artificial intelligence is on the verge of overturning multiple scientific domains. He highlighted the rapid automation of mathematics as a warning sign that other theory-driven fields could soon experience similar upheaval.
Thompson’s analysis tracked the performance of leading AI models from mid-2024 to mid-2026, noting an increase from roughly two percent to thirty-five percent of correctly solved theoretical problems. The fields examined included probability, statistics, economic theory, high-energy theory, gravitation and cosmology, and quantum physics, and the trend showed no indication of decelerating.
The investigation used problems extracted from research papers published between December 2025 and May 2026. Models tested comprised OpenAI’s o1, o3, GPT-5, GPT-5.1, GPT-5.4 and GPT-5.5, each run without internet access to prevent external look-ups. Solutions published after each model’s training cut-off served as the benchmark for evaluating correctness.
Because the study focused on areas where solving a problem does not require running simulations or physical experiments, it represents a conservative slice of scientific inquiry. Nonetheless, the researchers argue that the disruptive wave currently felt in mathematics is likely to propagate into other disciplines that rely heavily on abstract reasoning.
OpenAI’s recent claim of solving the Navier-Stokes problem,a long-standing fluid dynamics challenge,sparked controversy, with a NYU mathematician alleging that OpenAI had preempted collaborative work. The episode has intensified calls from mathematicians for safeguards, emphasizing that the process of solving can be as valuable as the final answer.
Looking ahead, several prominent AI scholars have launched laboratories dedicated to AI-driven science, and major AI firms are expanding their research programs. While AI may eventually handle both theoretical analysis and experimental design, Thompson cautions that human experts will remain essential for interpreting results and ensuring their validity.