Responsible Mechanism Design

Traditionally, the goal of mechanism design was to promote socially desirable behaviour of rational agents, to achieve fairness, or to promote efficiency. I would like to suggest a new subfield of mechanism design, Responsible Mechanism Design, focused on achieving individual accountability of agents for their contributions to the outcome of collective decisions. Why Now? From ancient democracies to the AAAI paper reviewing process, humans have long participated in collective decisionmaking, and just as long, they have assigned blame when outcomes go awry. Yet the mechanisms we use to make such decisions remain remarkably primitive. Most reduce to variants of majority rule, consensus, delegation, or fixed hierarchies of power. This simplicity is not accidental: humans tend to avoid mechanisms they cannot intuitively grasp, and our slow, error-prone communication makes complex coordination difficult. As a result, there has been little serious attempt to develop more sophisticated decision procedures. Compounding this is the fuzziness of responsibility itself. The final interpretation of legal responsibility is often left to courts, while moral responsibility has become the subject of conceptual studies in the humanities rather than mathematically grounded scientific analysis. The situation is changing as AI agents start taking a significant part in collective decision-making. Not only are such agents easily capable of engaging in complex decisionmaking protocols, but they can communicate with each other much faster. As a result, for example, we should expect that simplistic traffic-sign-based road rules designed for humans will be replaced by sophisticated coordination protocols that would allow self-driving cars to avoid collisions while never stopping at intersections. Additionally, recent works in AI introduced rigorous, mathematically precise definitions of responsibility that can be applied to humans and artificial agents and can also be used by AI agents to reason about their responsibility. Those definitions can now be used to adjust group decision-making protocols in a way that enforces agents' individual accountability.