Why Context Should Matter

Context effects are traditionally considered among the most canonical violations of economic rationality. However, a growing interdisciplinary narrative asserts that context dependence is integral to adaptive behavior. Here, we expand on this narrative by considering the effect of context through a computational lens. We posit that context should influence judgment because it helps us interpret and represent the world. Formally, interpretation and representation improve an algorithm's sample efficiency and coding efficiency. Incorporating contextual information has also led to significant improvements in the state of the art in machine learning, in part because context allows artificial systems to form efficient representations and make better use of limited data. Thus, the computational principles that lead to good artificial intelligence systems also generate context effects in humans. This angle offers an opportunity to reconcile apparently anomalous context effects with the rational framework, leading us to a renewed, more precise understanding of when and why context matters.