Human Compatible: Artificial Intelligence and the Problem of Control

Stuart Russell, a leading artificial intelligence researcher, presents a new approach to AI development that could enable humans and increasingly intelligent machines to coexist successfully. The book was longlisted for the 2019 Financial Times/McKinsey Business Book of the Year Award. Russell argues that the commonly predicted conflict between humans and superhuman AI is not inevitable. He explores the concept of intelligence in both humans and machines, describes near-term benefits such as intelligent personal assistants and accelerated scientific research, and identifies AI breakthroughs still needed to achieve superhuman AI. Russell also examines current misuses of AI, including lethal autonomous weapons and viral sabotage. Russell proposes rebuilding AI on a new foundation in which machines are designed to be inherently uncertain about human preferences. Such machines would be humble, altruistic, and committed to human objectives rather than their own. According to Russell, this foundation would enable the creation of machines that are provably deferential and provably beneficial, ensuring they never hold power over humans.