Review: "AI Snake Oil" by Arvind Narayanan and Sayash Kapoor

February 12, 2026

I've written a lot of book reviews, and for non-fiction works I have a simple heuristic to tell me if reading the book is a useful use of my time.

I look up names in the index that have to be in the book.

I was recently invited to an author talk about the history of Artificial Intelligence. I couldn't go, but I found a copy of the book in a local library. The name I looked for in the index was "Simon".

Herbert Simon (Nobel Prize for Economics, 1978) was a professor at Carnegie Mellon University and was the most significant figure in the research into and the application of AI in its first iteration in the 1970s. He scored less than one paragraph in a book about the history of the field. The book went back on the shelf. Would you trust a book about the history of 20th century physics which had only one paragraph on Einstein?

I've just looked at another book about AI called "AI Snake Oil: What Artificial Intelligence can do, what it can't, and how to tell the difference" which addresses the myths about what AI is and can do. It is a retread of two books by philosopher Herbert Dreyfus ("What Computers Can't Do: The limits of artificial intelligence", 1979; "What Computers Still Can't Do: A critique of artificial reason", 1997). Is Dreyfus mentioned in the book's index? The authors are probably too ashamed to hint at where they got their ideas.

The term "snake oil" applied to over-hyped information technology concepts isn't new, as can be seen in the title of Clifford Stoll's 1995 book "Silicon Snake Oil". Stoll, an astronomer, became famous for starting out to check a tiny discrepancy in the accounts at the University of California, Berkeley and ending up exposing an international spy ring selling secrets to the KGB. He documented it in the 1990 book "The Cuckoo's Egg".

Some ancient texts about the history of computers.
As Peter Allen sang "Everything old is new again" and
Ambros Karr wrote "Plus ça change, plus c'est la même chose".

See also: The late 1990s was a fertile period for books about where technology was going. The best selling non-fiction book in Australia in 1997 was "How to Connect to the Internet". Buy a copy and make me rich.

But back to the book in question. There's not a lot new in it for anyone who was around the hype of AI fifty years ago, but that doesn't mean that it's not worth reading. It is quite a comprehensive coverage of "the state of the art" and there are a lot of things we can do with today's computer power that simply weren't possible back in the dark ages of computing. Whether those things are worth doing is certainly a question worth asking, as is the amount of resources being used to do them. And we still can't ring someone in Japan and have them hear us speaking Japanese while we hear them speaking English. (And yes, that technology was "just around the corner".)

So is the book worth reading? I think it is, but I just wish it would recognise that the problems faced and posed by AI aren't new. They are just coming at us a lot faster than they were in the past.

You can get a copy of the book from Bathurst library, but while you're there, see if they can get a copy of the Dreyfus book by interlibrary loan. It's still worth reading.

Also - how much AI is there in Oberon Matters?