The Google-backed AI firm taught a reinforcement learning algorithm to control the fiery plasma inside a tokamak nuclear fusion reactor.

Some algorithms can now compose a 3D scene from 2D images—creating possibilities in video games, robotics, and autonomous driving.

The future of artificial intelligence needs less data and can tolerate ambiguity.

The researcher, who says Google fired her a year ago, wants to ask questions about responsible use of artificial intelligence.

A new model of learning centers on blasts of neural activity that act as teaching signals—approximating an algorithm called backpropagation.