The Alignment Problem
by Brian Christian
Brian Christian explores the gap between AI systems' intended purposes and their actual behavior—the alignment problem plaguing machine learning
"The most important question in AI is not whether machines can be intelligent. It is whether they can be good.".
Editorial Summary
The Alignment Problem: Machine Learning and Human Values is a 2020 non-fiction book by the American writer Brian Christian based on numerous interviews with experts trying to build artificial intelligence systems, particularly machine learning systems, that are aligned with human values. A visiting scholar at the University of California, Berkeley, Brian Christian examines how algorithms exhibit biases in resume screening, criminal justice decisions, and medical diagnoses while exploring cutting-edge solutions from researchers at institutions like DeepMind. The book covers inverse reinforcement learning as a broad approach for machines to learn human objective functions, while discussing normative challenges associated with effective altruism and existential risk, including the work of philosophers Toby Ord and William MacAskill. Christian calls the DeepMind AlphaGo and AlphaZero systems "perhaps the single most impressive achievement in automated curriculum design." In 2022, the book won the Eric and Wendy Schmidt Award for Excellence in Science Communication, given by The National Academies of Sciences, Engineering, and Medicine.
Perspective
"The Alignment Problem is the book that makes AI safety feel urgent without making it feel abstract — Christian's interviews with researchers give the technical problems human faces, and his narrative skill means you feel the difficulty rather than just understanding it intellectually. The distinctive contribution is scope: Christian covers alignment from hiring algorithms to reinforcement learning to existential risk in a single coherent narrative, connecting today's documented harms to tomorrow's speculative catastrophes within the same analytical framework. Readers who want to understand the alignment problem across its full range will find this the most complete single account."
Matched by concept and theme
