UKAEA data could train AI models on path to fusion energy
A new essay highlights the potential of AI data stocktakes to unlock fusion's data.
The Joint European Torus (JET) at Culham generated record-breaking fusion energy and vast quantities of scientific data for 40 years. Google DeepMind’s new essay asks a compelling question: could we use the data to train AI models and accelerate the path to fusion power?
The essay introduces the concept of ‘AI data stocktakes’ using interviews with 25 leading experts. This is a more structured approach to identifying where high-quality scientific data exists, where the gaps are, and what interventions could help unlock AI’s potential across scientific disciplines.
The picture is promising and complex for fusion energy. Much of UKAEA’s data remains raw, unvalidated, or inaccessible for commercial use. One expert described it as a ‘stranded asset’. Addressing this will require collaboration across the fusion energy community.
The essay highlights positive momentum including the UK government’s AI for Science strategy. It points to fusion as a proof-of-concept for a broader challenge: without better data infrastructure, AI’s transformative potential risks going unrealised.
As UKAEA continues work on MAST Upgrade and the STEP programme, it’s important to make good use of the data generated from fusion research.
The essay includes:
- Why fusion? Why now?
- How to accelerate fusion with AI
- The challenges with fusion data
- Recommendations
- Six open debates