In my research I am working on a formalized representation and quantification of responsibility in complex, multi-agent decision scenarios such as the global climate crisis. The focus currently lies on backward-looking moral responsibility for outcomes.
In this talk I will start out by presenting the game-theoretic framework which we use to represent the corresponding decision scenarios, before defining a general form of responsibility functions on this framework. With this general form of the function we make explicit a perception of responsibility for an outcome as an aggregation of point-wise responsibilities assigned for individual actions.
We employ an axiomatic method in order to evaluate candidate functions for each partial or final function: point-wise responsibility functions as well as aggregation functions and outcome responsibility functions. I will discuss certain desirable properties (axioms) and evaluate sets of proposed responsibility functions according to these.
An incompatibility result between two of the axioms regarding point-wise responsibility shows that we have to make a choice which we cannot justify by recurrence to axioms. However, for aggregation functions we find an axiomatic characterisation. Thus, I conclude by presenting a maximally axiom-compliant function for outcome responsiblity for each of the two choices we are able to make regarding point-wise responsiblity.