DeepMind and the Culham Centre for Fusion Energy trained an AI to control plasma in a tokamak reactor, sustaining stable fusion for 6 minutes—a record for magnetic confinement.
Technical Insight:
- Reinforcement learning algorithms adjusted magnetic coils in real time to counteract plasma instabilities.
- The AI reduced energy waste by 40% in trials at the UK’s JET reactor.
Applications:
- Clean Energy: Accelerating viable fusion power, potentially replacing fossil fuels.
- Medical Isotopes: Optimizing reactors for cancer treatment isotope production.
Hurdles: AI models must adapt to larger reactors like ITER, set to launch in 2025.