I always complain about books that were better as magazine articles; this one is the opposite! I didn't like the excerpt in The Atlantic which struck me as superficial but the longer version is a nice exploration of AI. The author is a participant in the annual Loeb Prize contest that tests chatbots to see if they can convince human interlocutors that they are human; each judge chats with a variety of bots and human confederates, without knowing which is which. There is a Most Human Computer award each year for the bot that tricks that most judges. The twist is that the author wants to win the Most Human Human award for the confederate who gets the most votes.

Since I have read about AI (kind of a lot) (recently), I was more interested in the sections having to do with conversation--how conversational turns work, who says what when, conversation "state," etc. He describes one bot that convinces people it's human by arguing with them--arguments are essentially "stateless" (each turn has little to do with the one before) so this mode is easy for bots to do. He also has the only discussion I've ever read of the strange way that an IM conversation can branch into two directions at once, and then continue that way, as a result of messages crossing each other.