Team: Niclas Boehmer, Markus Brill (project lead), Paul Goldberg, Davide Grossi, Jobst Heitzig, Wes Holliday.
Project description: The team will address the critical challenge of aggregating diverse and often conflicting human perspectives on AI safety. They will develop a framework for aggregating human safety specifications for AI systems, with the goal of contributing to a “gatekeeper AI” that mitigates risks posed by AI agents. The core technology involves novel methods in the tradition of social choice theory to elicit and aggregate stakeholder preferences regarding acceptable risk thresholds and safety conditions.