Review: The Reverse Centaur’s Guide to Life After AI, Cory Doctorow, Verso and God AI, Toby Walsh, Black Inc
Hysteria around AI continues to abound. Recently we have had wild claims about how AI will kill us all in five years, and Orwellian reports of AIs being trained on whole bookshops’ worth of books that are scanned and then destroyed. There has also been entirely justified consternation about the building of data centres, their pollution of adjacent residential areas and their greed for water and power. (Jeff Bezos has even suggested data centres should be prioritised over humans when it comes to water needs.)
Cory Doctorow is a sceptic when it comes to many of the grander claims about AI, but he is not completely negative. He advocates for the correct use of AI, building his book around the metaphor of a centaur – an augmented human. He writes that there are many good uses for AI that supplement our human creativity, as technology has in the past, but he warns that we are in danger of becoming ‘reverse’ centaurs – humans who exist to serve the needs of AI – or more importantly, the billionaires who are profiting from AI hype. (One example he gives of a reverse centaur is the exploited Amazon warehouse worker.)
His short, sharp book is an exercise in debunking the claims of AI founders. Recently, we have seen the curious situation of AI founders making statements about the existential crisis AI is drawing us into. Yet they refuse regulation and restrictions. Doctorow argues, not uniquely but passionately, that this is because AI founders are using this as a roundabout way of spruiking the power of their products, which otherwise don’t quite live up to the hype.
Talking up the inevitability of AI, says Doctorow, is just the old snake oil salesman’s trick. Talking about AI’s inevitable ability to replace human jobs encourages the desire of company leaders to do just that, and makes AI companies wealthy, even though we see time and again that AI is not capable of replacing human jobs, and the use of AI just means the need for humans to look out for AI errors (or ‘hallucinations’, as they are euphemised by AI companies). When company leaders want to use AI, it is because they want to reduce costs, not because they want better user outcomes, says Doctorow.
The hype around AI is aimed at investors, and so we see AI companies inflating the worth of their products and inflating the levels of user enthusiasm. Share prices are not dependent on how many users there are, but how many users investors think there could be. What look like attempts to convince you to love AI are actually attempts to convince investors to love AI. In this overheated market, the signs are pointing to an immense, dangerous tech bubble. Doctorow is worried about chatbots replacing workers, and not doing the workers’ jobs properly, then the bubble bursting and the work not getting done at all, after data centres have added catastrophically to our environmental problems.
AI can take the drudgery out of many tasks, but the claims of imminent AI consciousness are baseless, unless you think consciousness is a matter of text prediction. Claims of AI consciousness are driven by those who think that the human brain is merely a computer – ‘garbage in, garbage out’, as the saying goes – and Doctorow suggests debate about AI consciousness serves to distract from the significant problems of bias within LLMs, created by the ‘garbage in, garbage out’ model.
In contrast to Doctorow, Toby Walsh is equivocal about AI possibilities for good and evil, maybe because he is the head of an AI institute and being as sceptical as Doctorow might not be in his best interest. His book is a curious if entertaining combination of repeating AI booster claims and then wondering if they might be overstated – a case perhaps of giving equal weight to both sides of the argument.
Unlike Doctorow, Walsh is both optimistic about AI’s potential and now concerned about AI consciousness. Like Doctorow, Walsh observes that AI programmers also claim they are worried about existential risk, but he sees it as ‘odd’ that they are not speaking of environmental concerns. Doctorow would argue that this is not odd at all, because AI companies want to minimise the environmental risk while boosting the appearance of AI’s superiority in order to inflate share prices (something Walsh does later hint at).
Walsh suggests that AI has a real potential for exponential growth and therefore may become smarter than us – LLMs can train themselves to get better, and computers can run all day, self-improving all the while. He states that ‘superintelligent machines are a real possibility’ and they are ‘likely’ to surpass us.
Yet in adjacent pages he also suggests that ‘brains might not be anything like computers’. He writes that there is continual talk of ‘narrowing the gap’ between human and AI intelligence, but we have no agreement on how to measure that intelligence. Again, in thinking of the human brain as a computer, we assume actual computers will eventually match it. But, as Walsh emphasises, intelligence is more than just remembering facts and the speed of their regurgitation – intelligence is about how to arrange those facts and judge them, something AIs have thus far been notoriously ineffectual at.
‘Doomsday’ scenarios overestimate just how much potential there is for AI improvement, Walsh argues, but at the same time, AI is becoming so baked into our systems that just turning off the power switch if AI threatens is no longer an option. While stating that there is a danger AI will become our oppressor, Walsh admits there is more likelihood that AI will be used by our oppressors – to spread misinformation, to reduce wages, to wage war, to add to environmental woes.
Walsh emphasises how AI has been built on data theft, and echoes Doctorow’s concerns around bias. He also notes that AI might come up against natural limits – of resources, speed of computing. He says we can’t just keep building more (inefficient) data centres. Regarding the title of his book, although he lists reasons why an advanced AI might become God-like, and retains belief in the possibility of it becoming malevolent, much of what he writes suggests that rather than making for ourselves a god, we’ve made an idol, and we are sacrificing ourselves to it.
Nick Mattiske blogs on books at coburgreviewofbooks.wordpress.com and is the illustrator of Thoughts That Feel So Big.
