Building a god The ethics of artificial intelligence and the race to control it

"How do we build a superintelligent machine... that only helps us and can never harm us?" asks DiCarlo (So You Think You Can Think), a researcher at the think tank Convergence Analysis, in this scattered treatise. Hyperbolically claiming that controlling AI is more important than combatting climate change, nuclear annihilation, and world hunger, DiCarlo warns that "artificial general intelligence"--a hypothetical form of AI that could surpass human cognitive abilities--might, by means unspecified, gain consciousness and disobey "its humanly programmed ethical behavior commands." Unfortunately, his discussion of how to avoid such an outcome gets bogged down in granular tangents. For instance, the ostensible focus on AI is almost completely lost in a chapter-length explanation of basic ethics (individual sections of which explore "cultural relativism," "religion and morality," and the philosophy of Thomas Hobbes) and critical thinking ("In order to have an argument, you must have at least one premise supporting a conclusion"). DiCarlo calls for government regulations on AI development, but his admittedly sound critiques of proposals to program "human values" into AI (who decides which values count, and what it means to act them out?) don't offer better alternatives. Additionally, the inclusion of lengthy block quotes on nearly every page makes this feel like a collection of other authors' thoughts. In the crowded field of books on AI, this doesn't stand out. (Jan.)