Human Compatible

Outline

Human Compatible (2019) examines the risks posed by increasingly capable artificial intelligence and how to keep it under human control. Stuart Russell argues that machines designed to pursue fixed goals could act against human interests, and proposes building AI systems that remain uncertain about what people want and learn from human behaviour. The book offers a thoughtful framework for making AI beneficial and safe.

• Increasingly capable AI creates real risks, and keeping it under human control is the central problem the book addresses.
• Machines built to pursue fixed goals can end up acting against human interests.
• Russell proposes AI systems that stay uncertain about what people actually want instead of following a locked objective.
• Such systems would learn human preferences by observing how people behave.
• The book offers a framework for making AI beneficial and safe before machines outpace human values.

Stuart Russell is a British-born computer scientist and professor at the University of California, Berkeley, and one of the world’s leading researchers in artificial intelligence. He co-wrote Artificial Intelligence: A Modern Approach, the standard textbook in the field.