AAP 2026 CONFERENCE

There has been a recent proliferation of “AI Agents”: systems or programs that can operate in an increasingly autonomous manner. These systems raise an important question: do improvements in the capabilities of autonomous systems change the requirements for how we hold such systems – or their operators or designers – responsible for their outputs? Responsibility can be fruitfully connected to explainability – having the ability to explain an outcome helps in determining who, or what, should be held responsible for that outcome.

My talk will highlight how increasingly agentic systems pose challenges for pre-existing criteria for explainability. Specifically, I will examine how the increasing agentic features of algorithmic systems complicates the explanatory picture, and how different accounts of artificial agency can help to clarify these added complexities. I will conclude by considering how we should better understand the role of explanation in relation to increasingly complex explanatory contexts involving AI agents.