Counterfactual Responsibility in Multi-Agent Systems

A new arXiv paper introduces a formal framework for allocating causal responsibility in probabilistic multi-agent systems. The work models such systems as concurrent stochastic multi-player games and defines a retrospective counterfactual notion of responsibility that quantifies an agent's accountability for outcomes under a given strategy profile. To distribute responsibility among agents, the authors employ the Shapley value and prove it satisfies fairness and consistency. The framework supports both verification and strategic reasoning, and by adopting Nash equilibrium as the solution concept, they demonstrate how to compute stable strategy profiles where agents trade off responsibility against expected utility.