Can an AI Agent Safely Run a Government? Existence of Probably Approximately Aligned Policies

Frédéric Berdoz, Roger Wattenhofer

While autonomous agents often surpass humans in their ability to handle vast and complex data, their potential misalignment (i.e., lack of transparency regarding their true objecti…