Walk up to anyone and ask if they know what ChatGPT is: they’ll very likely offer an explanation. Now ask if they know what GPT stands for. Shrugging shoulders and empty stares will follow.
There is a chasm between our enthusiasm for artificial intelligence and our understanding of it. Everyone has opined about AI over the past year, opining on whether it willl cut jobs or create new ones, whether it will transform our lives or kill us all. But next to no one is familiar with even the basic terminology, let alone how it all works.
And that is a problem. Not just because AI will be involved in a growing share of our lives, but because it will pose threats to our online world that only AI-literate folk can tackle.
Few books are more timely, therefore, than How AI Thinks, an accessible guide that walks the reader through the technology’s developmental history right back to the days before the computer.
Author Nigel Toon, the CEO of British semiconductor firm Graphcore that develop processors for AI, knows more about the subject than most, including — crucially — what GPT stands for (it’s Generative Pre-trained Transformer, to put you out of your misery).
But he is desperate to prove that getting to grips with the basics isn’t hard. Toon is dyslexic, he points out, and yet struggling to spell half the jargon he elucidates has not stopped him running a business making the kind of chips on which the technology relies.
So what does “think” mean exactly in the context of AI? Toon offers his own definition of intelligence: “The ability to gather and use information to adapt and survive.” The definition has always applied to animals, he argues, but most computers are dumb: they produce a series of outputs given a series of inputs and a set of rules. But machine learning is different: it adjusts its own rulebook based on changes in the information it receives, in the same way animals adapt based on changes to their environment observed through sensory data.
A pioneering moment in computing history was marked modestly in Tommy Flowers' diary: 'Colossus did its first job. Car broke down on way home.'
To trace the early development of AI, Toon takes us right back to the mid-19th century. A set of notes written by pioneering mathematician Ada Lovelace (who also happened to be Lord Byron’s daughter) in 1840 was thought to be the first expression of a rudimentary computer program, Toon tells us, and it was considered influential on the work of Alan Turing in developing his concept of a “universal machine”, the forerunner of what we know today as a computer.
But while Turing is often lionised as the British genius behind computing, Toon points to his colleague at Enigma-cracking Bletchley Park during the Second World War as the unsung hero. It was Tommy Flowers , not Turing, who in February 1944 created the first electronic computer, Colossus, that decoded encrypted German messages — the moment was modestly marked in Flowers’s diary entry at the end of a day’s work: “Colossus did its first job. Car broke down on way home.”
Women had an outsized role in producing the software that would later evolve into AI, we learn. Toon suggests that this is in a strange sense the by-product of misogyny: in the early days of computing firms, men worked on the hardware, what they considered the serious stuff, while women were left to test and fine-tune the software, perceived as largely cosmetic — a reason that it was a woman, Grace Hopper, who first referred to a computer glitch as a “bug” after finding a dead moth inside the machine.
And Toon tracks the birth of the internet and the huge progress of power and efficiency of semiconductors. The decades-long observation of Moore’s Law, the belief that computer power doubles every two years has laid the groundwork over decades to the present era, where we have the immense processing and data server capacity to build the first large language models — like ChatGPT.
This is a fascinating read. But it is not a moral framework. A small section at the end discusses AI ethics — but beyond some civil servant-style platitudes that we must “ensure that regulations are followed” and “have a process to monitor outcomes”, it doesn’t get far. Deeper analysis is wanting here, not just because the debate is lively and fascinating, but because the more anxious among us need reassurance that we can chart a course to avert the world’s destruction. “Every living organism on this planet… they are all depending on us,” says Toon. So what do we do?
How AI Thinks by Nigel Toon is out on February 8 (Torva, £22)
