In a recent paper in PRX Life, Prof. Nemenman and collaborators explore intrinsic motivation — the ability of biological or artificial agents to learn new skills without an explicit reward. They introduce a new method that can compute a certain quantity, called empowerment, which has been proposed as a good proxy for intrinsic motivation. Empowerment maximizes the capacity for potential actions to influence future states of the system. Here, it is calculated for continuous dynamical systems, linking behavior to fundamental properties for systems introduced in dynamical systems theory. The paper shows that empowerment maximization, indeed, can be used to make agents do nontrivial tasks. Read more here!