Don’t panic! Robots will keep us on as pets

Steven

many would be enough. So the AI, being very intelligent, first disables its off switch. After all, if someone turns it off it won’t be able to fulfil its mission to make paper clips. Then it happily starts making paper clips, and doesn’t stop until everything on Earth has been mined for its constituent atoms – including all the humans – and turned into paper clips. Job done.

The problem this nightmare illustrates is that it is very difficult to specify any goal you give an intelligent machine to avoid misunderstandings. In Human Compatible: AI and the Problem of Control (Allen Lane, £25), machine learning pioneer Stuart Russell gives an excellent, nuanced history of the field, and suggests we build a kind of humility into our robots, programming them not to fulfil particular goals but to infer what we want from our behaviour, and to always be ready to change their minds.

A brighter future might be one in which AIs create beautiful art for us to enjoy, a possibility cunningly teased by the mathematician Marcus du Sautoy in The Creativity Code: How AI Is Learning to Write, Paint, and Think (Fourth Estate, £20), before he proceeds to elegantly demolish naive boosterism on the subject. He travels around talking to experts in machine learning, including computers that play Go and chess, but also dating algorithms and systems that analyse paintings by Rembrandt to create new ones. It turns out that, paper clips aside, human artists have little to fear yet. Computer algorithms, de Sautoy concludes for example, can generate muzak, but not “quality music”. Any creativity a computer system displays, he points out, has been put there by the creativity of the people who designed it.

The same, unfortunately, is also true of any destructiveness a computer system might exhibit, as pointed out in Matt Parker’s very funny Humble Pi: A Comedy of Maths Errors (Allen Lane, £16.99), a compendium of stories about mathematical failures; some are amusing, others alarming, as in the case of the passenger aircraft that ran out of fuel because it had been measured in the wrong units.

Modern machine-learning systems for facial recognition can turn out notoriously racist results (mistaking black people for gorillas), because they are trained on cultural material that is itself racist. Angela Saini’s magisterial Superior: The Return of Race Science (Fourth Estate, £16.99) is a forensic demolition of the racism that persists even in modern medicine, as well as in dodgy journals that lend a quasi-academic imprimatur to the prejudices of the globally resurgent far right. Such prejudices are, paradoxically, mirrored in much left-green writing about the natural world, with its horror of “invasive species”, as Dan Eatherley points out in his warm and nuanced Invasive Aliens: The Plants and Animals From Over There That Are Over Here (William Collins, £16.99). Almost all foreign plants and animals, he reminds us, innocently hitchhike on human trade: that’s globalisation for you.

Artificial intelligence might be a more plausible future if it turns out life as we know it runs on the same juice as computers: information. The physicist Paul Davies, in his The Demon in the Machine: How Hidden Webs of Information Are Solving the Mystery of Life (Allen verba” – take nobody’s word for it – can still come in handy today. Another gem in the history of science this year was Benjamin Wardhaugh’s Gunpowder and Geometry: The Life of Charles Hutton: Pit Boy, Mathematician and Scientific Rebel (William Collins, £20). The author colourfully narrates the life of Hutton, a miner turned schoolmaster who rose to become professor of mathematics at the Royal Arsenal at Woolwich. They tested ballistics equations back then by firing cannon down the Thames; more innocent times.

Sceptics say we have more to fear from natural stupidity than from artificial intelligence, and this is the view of James

Lovelock, the eminent scientist and inventor. In his wonderful Novacene: The Coming Age of Hyperintelligence (Allen Lane, £14.99), he hails our future robot overlords as the inevitable next step in the evolution of conscious life. Rather than enslaving or destroying us, he thinks they’ll keep us around as pets who

(like all living things) will help regulate the climate, which is partially reassuring. The book is illuminated by mischievous wit and offhand brilliance, and readers might particularly enjoy the suggestion that we could combat global warming by beaming all social media and advertising out into space. There, as we know, no one can hear you scream.

Steven Poole’s A Word for Every Day of the Year (£14.99) is published by Quercus