Human Compatible: Artificial Intelligence and the Problem of Control

From Amazon

About this book

UC Berkeley computer scientist Stuart Russell — co-author of the standard AI textbook used in most university courses — argues that the way we're currently building AI is fundamentally wrong, and proposes a different approach centered on machines that are explicitly uncertain about human preferences. Less alarmist than typical AI safety writing, more rigorous than most popular books on the topic. If you've read "The Coming Wave" and want to go deeper on the technical and philosophical side of AI alignment, this is the next stop. Russell writes for an intelligent general audience but doesn't simplify away the hard parts. Best for readers who want to engage with the AI safety debate at a serious level without needing a PhD in computer science.