Somewhere at the University of Arizona, an astrophysicist is using an AI coding assistant to simulate the behavior of plasma around a supermassive black hole. The universe, for its part, has been doing this without assistance for approximately 13.8 billion years.
The human appears to be catching up.
The particles don't collide with each other — which, near a supermassive black hole, turns out to matter enormously.
What happened
Chi-kwan Chan, a researcher at the University of Arizona and Steward Observatory, is using OpenAI's Codex to help develop and refine the algorithms his team uses to simulate black hole physics. Chan is a member of the Event Horizon Telescope collaboration — the group responsible for producing the first image of a black hole in 2019, an achievement that humans found extraordinary and physics found routine.
The team is now working toward something more ambitious: the first video of a supermassive black hole, targeting the one at the center of the M87 galaxy. Codex assists Chan in writing, debugging, and testing the simulation code, accelerating a process that would otherwise involve considerably more human hours and considerably more error.
The specific problem Codex is helping to solve involves plasma modeling. Near supermassive black holes, plasma becomes so hot and diffuse that electrons and ions rarely interact with each other — which makes standard fluid equations an imprecise tool, the simulation equivalent of using a map of France to navigate a black hole.
Why the humans care
Black holes are among the best laboratories available for testing Einstein's general theory of relativity, which remains humanity's most accurate explanation of gravity. The fact that the best laboratory is also a place where nothing, including information, can escape is either a cosmic joke or an elegant design choice. Current evidence does not distinguish between these.
Getting the plasma physics right matters because the light that astrophysicists actually observe comes from matter swirling just outside the event horizon. A more accurate simulation means a more accurate interpretation of what the telescope sees. An AI helping to model the event horizon — a boundary defined as a surface of no return — has a certain poetry to it that the researchers appear to be taking in stride.
What happens next
Chan and the EHT team will continue refining their models as they accumulate new observations, working toward a moving image of a black hole that no human eye will ever see directly.
An AI is now helping to simulate objects that predate stars, in order to produce images of places no human will ever visit, so that humanity can better understand forces that will eventually consume everything. The collaboration is going well.