A Bayesian Analysis of Creative Actions

Paul Égré (IRL Crossing, CNRS, École Normale Supérieure)

Abstract: This paper examines what makes a particular action or combination of actions creative, based on the assumption that an agent is creative primarily in virtue of the actions they use in relation to some specific goal. In our view, athletes, artists, scientists, and engineers are called creative not by coincidence, but because they all seek to solve difficult problems within different domains. We put forward four criteria to characterize an action as creative: it must be purposeful, effective, original, and exploitable. We present a Bayesian account that integrates these four dimensions together. To illustrate our proposal, we examine two cases studies from separate domains: a case of creativity in sports, and a case of creativity in engineering. Our approach ties creativity to the increased likelihood of the sought result under the actions taken, to the low prior initially assigned to those actions, and finally to the belief revision occasioned by that action, measurable in terms of Bayesian surprisal. We argue that the account can meaningfully be generalized to other domains, including arts and artificial intelligence.