AI controls fusion plasma faster than humans can react
Princeton's PPPL deployed machine learning systems that predict and stabilise tearing-mode instabilities in tokamak reactors on millisecond timescales — faster than any human operator could intervene. The result is validated in active tokamak runs, not only in simulation.
- ML model predicts instability onset and triggers magnetic corrections in under 100ms
- Validated in active tokamak operations, not simulation only
- Princeton team sees the system as the path to continuous high-density plasma operations