Using Feasible Action-Space Reduction by Groups to fill Causal Responsibility Gaps in Spatial Interactions

A research team from Delft University of Technology, including Vassil Guenov, Ashwin George, and Arkady Zgonnikov, has published a novel paper addressing a critical gap in autonomous systems ethics: determining responsibility when multiple agents collectively cause an outcome. The work, titled 'Using Feasible Action-Space Reduction by Groups to fill Causal Responsibility Gaps in Spatial Interactions,' tackles the problem of 'causal overdeterminism'—scenarios where an incident (like a collision) is caused simultaneously by several autonomous vehicles or robots, making it impossible to pin blame on a single entity using existing individual-focused metrics. This is a fundamental challenge for the deployment of AVs and mobile robots in dense, real-world environments.

The researchers' core contribution is a formalized metric for calculating the causal responsibility of a *group* of agents. They developed a 'tiering algorithm' to systematically identify which agents are 'assertively' influencing a situation and formalized types of assertive influence. Using scenario-based simulations, they demonstrated how group responsibility effects vary with the dynamics of interaction and the physical proximity of agents. This provides a mathematical and ethical framework for insurers, regulators, and developers to analyze complex multi-agent accidents, moving beyond the limitations of individual fault assessment and paving the way for more nuanced liability models in the age of autonomy.