Chi-kwan Chan, an astrophysicist at the University of Arizona, is applying OpenAI Codex to one of astronomy's most computationally demanding challenges. His work focuses on simulating the plasma surrounding black holes, which the Event Horizon Telescope collaboration studies to test Einstein's general theory of relativity. Chan helped develop tools used to interpret the first black hole image published in 2019.

The core problem lies in modeling plasma behavior around supermassive black holes. In extreme conditions near these cosmic monsters, particles rarely collide and instead spiral around magnetic field lines. Traditional simulations must calculate every microscopic turn, forcing computers to use impractically small timesteps that consume resources better spent on larger-scale physics.

Chan turned to Codex to explore mathematical approaches that could bypass these limitations. The AI generated numerous candidate algorithms—many incorrect, but all testable. Unlike opaque AI systems, Codex produces inspectable code that researchers can understand and validate physically. This matters because scientific ideas must survive rigorous testing regardless of their origin.

If successful, these new algorithms could simulate trillions of particles previously impossible to model. The approach represents a practical application of AI in scientific computing where generated hypotheses undergo traditional peer review and experimental validation before acceptance.