The More you Know About Everything, the Less you Know About Anything: What makes it easier to determine what is relevant in problem solving?

Abstract

Formalizing how to decide what is relevant information for solving a problem and what is not (the relevance problem; Fodor (1987)) is notoriously hard. Yet, it is vital for constructing autonomous agents that they can make this distinction. In this thesis I will argue, using a conceptual analysis understanding when it is difficult for an agent to determine (ir)relevant information will provide a basis for designing artificial agents that can survive in the real world. I will use the computational framework of expected relevance by Kwisthout (2012) and present an overview of constraints which may make this framework feasible.