Human Compatible
Stuart Russell
Given how fast things are moving forward, a book on AI published in 2019 may seem outdated. On the contrary – everyone interested in AI and needing an in-depth understanding should read this book. As well as discussing the benefits to be gained from AI, Russell also explains, in great detail and with clear examples, why our approach to AI today is not necessarily the best. His core question explores how we ensure machines remain beneficial to humans and addresses a fundamental challenge: increasingly intelligent machines that do what we ask but not what we intend.
Russell starts by demystifying AI and clarifying its capabilities and boundaries. He argues that AI may excel in structured environments like chess, where its logic navigates a finite set of moves. However, AI’s ability to predict outcomes falters in the dynamic tapestry of human decision-making, where actions are not necessarily bound by rationality. Today, dissecting complex…