Every year computer scientists compete in for the Loebner Prize in artificial intelligence. The competition is based on the Turing Test whereby ‘blind’ judges interact with a computer chatbot and a human, and then have to decide which is which. The main prize, of course, is for the computer programme that comes closest to convincing the judges it is human - that it can pass for genuine intelligence. But there is another prize too: for the human who convinces the most judges they are not a computer. This is the task Brian Christian sets himself in this enjoyable tour of the challenges of creating a computer that thinks and behaves like a human. Note, this may not be the same as creating real intelligence. Past successes in the Turing test have come through programmes that mimic irritability, changing the subject or pretending not to fully understand English.

One of the easiest ways to trip up a chatbot is to refer to something ‘site-specific’. You can have millions of lines of densely programmed code tripped up immediately by a reference to the weather outside, or by noting the test has started 15 minutes late. There are enjoyable digressions into the world of chess computers and the philosophy of language. The fact that one chapter opens with quotes from both Plato and Phil Collins shows that the author - a science journalist - does not take himself too seriously.

The ultimate lesson, though, lies not for the way computers can be improved in their artificial intelligence, but in how our own interactions with technology often have a narrowing effect on language. Auto-corrected text messages, auto-fill Google searches and other online communication often follows rigid patterns that are easier to mimic than we might like to admit.