Research Seminar in Decision and Action Theory: Jurgis Karpus (LMU)

Title:

Interactive decision making for real

Abstract:

Standard game theory postulates that rational decision-makers (“players”) seek to maximally satisfy their personal preferences over the possible outcomes of their interactions (“games”) in light of their beliefs about what other players in a game will do. However, this model of best-response reasoning fails to account for people’s ability to easily coordinate their actions to attain mutually beneficial results. Also, in many games in which interacting players’ preferences are partly misaligned, people often cooperate to attain outcomes that they do not individually prefer the most. Numerous theories have been proposed to explain these findings. However, several of these approaches, including the theory of team reasoning, have thus far been accepted as good descriptive accounts of what people often do, but not as extensions of the normative theory itself. In this paper, we present an alternative view. We argue that people’s preferences often depend on contextual properties of outcomes, which the standard theory is unable to account for. This drastically expands the set of rationalizable choices that are available to decision-makers in interactive decision situations. It also allows us to account for a series of unsurprising, but puzzling behavioural findings in the study of human interaction with artificial agents.