A Notion of Relevance for Rational Decision Making

Abstract Decision theories have largely ignored the step of decision making in which the agent models the situation. Given that a decision can be represented with different models, and that these can lead to different recommendations, then without a principled way to assess them the agent’s choice is under-determined. As models require the agent to select the aspects that matter to the decision, an account of rational decision modelling must include a notion of relevance. I propose that the most rational model is the one taking into account all and only the considerations relevant for the decision. I define relevance for a decision as a matter of providing reasons for some option, and I identify four functional types of reasons leading to four corresponding types of relevance. I focus on what I call “constitutive relevance”, which provides the content of the decision model, and propose a formal definition of this concept: the central idea is that the model should include all and only those considerations the inclusion of which changes the utility of at least one of the alternative options.