Last week, Neil Thompson, a researcher at MIT, presented findings on the influence of artificial intelligence (AI) on various scientific disciplines. Thompson's study indicates that fields such as probability, statistics, economic theory, high-energy theory, gravitation and cosmology, and quantum physics may soon experience significant changes due to advancements in AI capabilities. The study reports that between mid-2024 and mid-2026, leading AI models improved their ability to solve theoretical problems in these areas from approximately 2 percent to 35 percent. Thompson stated, "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."
Thompson's lab investigates the societal impacts of computer science advancements. The study, conducted with PhD student Gabriel Manso and research scientist Claudio Mayrink Verdun, involved testing various OpenAI models—o1, o3, GPT-5, GPT-5.1, GPT-5.4, and GPT-5.5—on problems from research papers published between December 2025 and May 2026. Notably, the models had to independently solve these problems without relying on pre-existing training data, as the solutions were found in papers published after the models' training cutoffs. Internet access was also blocked to ensure the models could not look up solutions online.
The research focused on areas where AI could derive solutions without simulations or experiments, suggesting that the disruptions currently observed in mathematics could extend to other theory-heavy scientific fields. In September, OpenAI announced that an unreleased model had solved the Navier-Stokes equations, a significant challenge in mathematics, which led to controversy over claims of academic misconduct. Following this, OpenAI and others reported numerous solutions to longstanding mathematical problems, prompting some mathematicians to call for restrictions on AI usage, emphasizing the importance of the problem-solving process itself.
Thompson noted that advancements in AI could yield both beneficial and detrimental effects across disciplines. He pointed out that theoretical breakthroughs often lead to practical advancements in science, and increased AI utilization may accelerate this process. Several AI researchers have established new labs dedicated to applying AI in scientific research, with major AI labs intensifying their focus on scientific inquiries. These initiatives aim to explore AI's potential to conduct real experiments alongside theoretical analysis. Despite the potential for AI to solve key problems, experts will still be necessary to interpret solutions and ensure their validity. Manso remarked, "We are still going to need people to generate good ideas to connect the dots," and cautioned that AI could make errors in complex problem-solving. OpenAI has retracted several solutions from its latest dataset post-publication, highlighting the need for verification of AI-generated results.