Are we heading for an AI apocalypse or are Terminator-style prophecies overblown? In 2023, the Future of Life Institute called for ‘all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4’. Among the 33,000 signatories to this statement were AI experts Max Tegmark, Stuart Russell and Elon Musk. Meanwhile, Sam Altman, CEO of Open AI (the company behind ChatGPT), mentions in interviews that he shares a birthday with Robert Oppenheimer. Why are so many tech experts warning us about AI while simultaneously investing in it? Are we living in an actual dystopia or an age of dystopian fantasies? All sorts of people provide all sorts of answers to these questions, but the trouble is that so many of them sound (a) quite weird, (b) quite evil or (c) both.

For these reasons, I was delighted to read Christopher Summerfield’s thoughtful and illuminating book. Summerfield is a professor of cognitive neuroscience at the University of Oxford, and is neither a wild-eyed utopian nor a doleful Cassandra. His book is mostly concerned with large language models (LLMs) such as Open AI’s ChatGPT, Anthropic’s Claude and Google’s Gemini. These LLMs are ‘trained’ using masses of text – taken from websites, articles, books – until they can generate copious answers to questions. Effectively, they are prediction machines, attempting to guess the next ‘token’ in a sequence of words, numbers or symbols. They used to be amusingly cranky and prone to spouting gibberish but this is rapidly changing. Summerfield argues that the dawning of an era in which ‘AI can speak’ is a ‘watershed moment’ comparable to the invention of writing, the printing press and the internet. He poses three questions: ‘How did we get here?’, ‘Do language models think?’ and ‘Are we all doomed?’

In pursuit of answers to the first two questions, Summerfield presents a lucid survey of themes including Cartesian dualism, logical positivism, the possible origins of language and why even the brainiest chimps haven’t yet written Hamlet. The explosive growth of the internet in the 1990s changed the landscape for AI,