Machine learning, which lets researchers determine if two processes are causally linked without revealing how, could help stabilize the plasma within doughnut-shaped fusion devices known as tokamaks. Such learning can facilitate the avoidance of disruptions—off-normal events in tokamak plasmas that can lead to very fast loss of the stored thermal and magnetic energies and threaten the integrity of the machine. A paper by graduate student Matthew Parsons published in June in the journal Plasma Physics and Controlled Fusion describes the application of the learning to avoiding disruptions, which will be crucial to ensuring the longevity of future large tokamaks.