AI Came for Math. Now It’s Chasing the Rest of Science
wired ·
MIT researcher Neil Thompson argues that artificial intelligence is poised to disrupt theoretical sciences beyond mathematics, including economics, high-energy physics, and cosmology. A recent study by Thompson and his colleagues demonstrates that top AI models improved their ability to solve theoretical problems in these fields from approximately 2 percent to 35 percent between mid-2024 and mid-2026. The researchers tested various OpenAI models against problems published after the models' training cutoffs, ensuring the systems generated solutions independently rather than recalling memorized data. This rapid advancement mirrors the current upheaval in mathematics, where AI has solved long-standing challenges like the Navier-Stokes equations, sparking controversy and calls for regulation among mathematicians. While some experts warn that relying on AI may undermine the valuable process of problem-solving itself, others suggest that automated theory generation could accelerate practical scientific discoveries. Despite these capabilities, human researchers remain essential for verifying AI-generated proofs, connecting conceptual dots, and ensuring the soundness of complex results. Consequently, major AI laboratories are expanding their focus toward integrating theoretical analysis with real-world experimental validation.
wired