AI Snake Oil

Narayanan and Kapoor's central move is taxonomic rather than alarmist: they split "AI" into categories that behave completely differently, generative tools that genuinely work, and predictive uses (hiring screens, recidivism scoring, content moderation) where the underlying claims are frequently unfalsifiable or simply false, and argue most public confusion comes from treating those as one technology. Reviewers, including in scientific press, have praised the book precisely for this discipline: it doesn't traffic in either hype or blanket dismissal. The most substantive criticism raised is scope, the analysis is heavily anchored in US markets and US-style AI deployment.