Compatible with Humans: Artificial Intelligence and the Control Problem

Compatible with Humans examines the critical challenge of ensuring advanced AI systems remain under human control. Leading AI researcher Stuart Russell (with contributions by Peter Norvig) explains the “control problem”—how optimizing powerful agents can produce unintended, even dangerous outcomes. The authors survey the current state of AI, propose rigorous safety frameworks, and outline strategies—such as value alignment, provable guarantees, and human-in-the-loop designs—to keep AI beneficial and safe as it grows increasingly autonomous.