nocturama 's review for:

AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference by Sayash Kapoor, Arvind Narayanan

---

reviews of nonfiction books are irritating because they're just indicators of whether or not the reviewer agreed with the arguments/ideas/politics of the book in question, not an actual evaluation of the craft of the book. positive review = reviewer agrees with author, negative review = reviewer disagrees with author.

so in lieu of that, i'll list out the most interesting insights from my reading:

• after thoroughly debunking predictive ai, i was surprised by the authors' almost existential conclusion that we as a society simply have to accept the inherent randomness and unpredictability of life: "When people are forced to confront randomness and the illusion of control breaks down, they look for ways to take back control." predictive ai is so tempting because it presents the false promise of imposing order upon a fundamentally unknowable world.
• the authors are much more sanguine about generative ai like chatgpt and image generators, but they do list out the many (already very well known) problems with such technology: stealing writing/art without attribution, its propensity to spit out wildly false claims, reliance on the labor of underpaid contractors in the global south, etc. but as other reviewers noted, they don't at all mention the climate impacts of all that energy consumption.
• the long, 80-year history of generative ai was enlightening, understanding the boom/bust cycle ai has gone through helps to put this current ai boom in its proper perspective: "Every day there are news reports of AI-related product launches or improvements in capabilities. But ... this suddenness is an illusion: The tech behind generative AI has developed over the past eighty years."

anyway i continue to be generally suspicious/skeptical of ai, which i think the authors are as well, while also trying to not become an old man yelling at clouds.