AI Snake Oil

The book provides a good introduction to many different aspects of AI, an umbrella term for very different techniques. They cover 3 different domains: predictive AI, generative AI and AI for content moderation. There are many examples and plenty to agree with. In particular, people should treat the claims from predictive AI with the same degree of skepticism as any "traditional" solution: the vendor must prove first that the system works before one considers it. This seems like the least to ask but as soon as the word AI is sprinkled over a program, it seems that all critical thinking goes out of the window. AI is not special and this is also the stance of the FTC as well: even if it's AI-based, it is still subject to laws regarding false advertising. Despite their critics and their exposition of the limits of AI, the authors are still weirdly (and disappointingly) very positive about generative AI, seeing even more potential in it than in predictive AI. This is not going age well. The chapter about content moderation is fairly interesting and the argument that it is actually a good thing that this issue cannot be solved with any technology, including AI, is a convincing one. Those are political and moral decisions and they require an informed debate. They certainly can't be left to private companies or software engineers. (Incidently, their characterisation of content moderation on the Fediverse completely misses the mark ("worst of both worlds"); the point of moderation there is to protect and nurture a community/instance, not to fix the rest of the network in one fell swoop. Scaling is not the point.)

For a more informed and biting review of the book, see Edward Ongweso Jr's review in The New Republic (21 Nov. 2024).