The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World

These two dimensions of the reinforcement-learning problem have come to be known by the technical terms of the policy—what to do when—and the value function—what rewards or punishments to expect.

Inverse reinforcement learning is, famously, what mathematicians call an “ill-posed” problem: namely, one that doesn’t have a single, unique right answer.

Learning, in both brains and machines, thus requires searching for an optimal combination of parameters that, together, define the mental model in every detail.