The success of large language models like ChatGPT as part of the development of artificial intelligence has left the future looking even more uncertain than cliché normally paints it, adding fresh urgency to old questions. Are we set for a utopian future of abundance, or might we be facing a world in which we eventually fuse with machines? Could there be dark times ahead, where we worship false gods that reflect our worst biases back to us, or will these strange tools help us better define our own nature?
A tranche of new books may help navigate these waters. Let’s start with philosopher Nick Bostrom, who became well-known in AI circles for Superintelligence, his 2014 book in which he conjured the idea of an AI so much smarter than humans, across so many domains, that it could pose an existential threat to us.
The idea of a “superintelligent” and malevolent AI has been taken very seriously by some technology companies, notably OpenAI, and has elevated Bostrom to a key figure in the “AI safety” movement. But the approach has also been ridiculed by many as overly pessimistic and not grounded in reality.
Advertisement
It is unclear whether criticisms of being overly gloomy have got under Bostrom’s skin, but he has turned his attention to far rosier futures in his new book, Deep Utopia: Life and meaning in a solved world. Here, Bostrom draws the contours of what our possible futures might look like, assuming the breathless hype around AI turns out to be true. The book isn’t a conventional philosophical tome: it is structured as a series of lectures to fictional students, sometimes turning into a Socratic dialogue, with occasional digressions into letters written from the perspective of a worried fox.
The writing is often clunky and there are some painful comedic asides, but there are enough fresh ideas in the book to make it a stimulating read. Bostrom asks how the economics of a post-work society might function, what kinds of utopia could exist and whether it will be possible to find meaning and purpose in a world where machines can do everything for us.
Like Superintelligence, though, it falters in its core assumption that human-like artificial general intelligence (AGI) can be achieved, and superhumanly so. Many other eyebrow-raising suggestions, such as uploading our minds to the cloud, are also explored. But they, like the book, feel more like thought experiments than a handy guide to how the future might be. It should also be noted that, last year, emails written by Bostrom came to light that contained racist comments about IQ. Earlier this year, the University of Oxford closed the Future of Humanity Institute, the department Bostrom headed, and he subsequently resigned from the university.
Ray Kurzweil is another grandee among those predicting AI futures and a believer in the possibility of radical change. He is most famous for the precise predictions in his 2005 book The Singularity is Near, where he wrote that, by 2029, we will be able to create machines that pass the Turing test and that, by 2045, non-human intelligence will be so powerful that it will be able to self-improve and merge with humans – the so-called “singularity”.
Â鶹´«Ă˝ book club
Love reading? Come and join our friendly group of fellow book lovers. Every six weeks, we delve into an exciting new title, with members given free access to extracts from our books, articles from our authors and video interviews.
In his new book, The Singularity is Nearer: When we merge with AI, Kurzweil not only re-emphasises his predictions but adds that, according to the evidence, he may have been too cautious. Sudden jumps in apparent intelligence – such as in image recognition or, more recently, with large language models – are evidence that his 2029 prediction is still on track, he says. We are now in the “steep part of the exponential” curve, he writes, which means that all of Kurzweil’s prior predictions about superhuman intelligence will be rushing towards us ever faster.
While this is apparently good news for the accuracy of his predictions, there is little by way of fresh ideas in his new book. There is a useful overview of AI history, as well as brief accounts of many new, exciting technologies, from fusion to biotechnology, but the distinct lack of caution or effort in examining whether human intelligence is something that can be extracted into a machine leaves the book feeling one-dimensional.
Neil Lawrence, professor of machine learning at the University of Cambridge, thinks that Kurzweil and Bostrom’s ideas about superintelligence and singularities are flawed. “They misrepresent intelligence as a unidimensional quality and this doesn’t reflect the diversity of intelligences we experience,” he writes in The Atomic Human: Understanding ourselves in the age of AI.
Intelligence, according to Lawrence, is far grander and more subtle. It can be seen in the collective intelligence of vast groups of people, such as in the British Army, in which his grandfather fought, or through organisations like Amazon, where Lawrence was in charge of AI research for three years. AI is also toothless if it can’t deal with uncertainty and work out how wrong it might be compared with the real world – something that is straightforward for humans.
Lawrence’s aim is lofty: by explaining how AI came to be and rose to its current status, and by contrasting it with the ways humans function, the essential, “atomic” core of humanity and human intelligence will be exposed, separate from AI. The links and connections he makes in this mission, from the code-breaking activities of Alan Turing to Lawrence’s own early days on off-shore oil rigs, are sprawling. His writing is elegant, but the bigger picture can be confusing and unclear.
One of the more useful aspects of Lawrence’s book, however, is in pouring sufficient cold water on the hype from writers like Kurzweil and Bostrom. Perhaps by design, The Atomic Human offers little in the way of predictions.
Shannon Vallor, a professor of ethics specialising in AI at the University of Edinburgh, UK, shares Lawrence’s scepticism. In The AI Mirror: How to reclaim our humanity in an age of machine thinking, she goes further, asking why we can’t be more imaginative. “Why do so few depictions of AGI show us a superhuman intelligence that laughs more than we do?” she asks. “Where are the intelligent machines not on a mission, but mastering being silly, goofing off, exploring, playing? Most highly intelligent creatures do a lot of this!”
Vallor isn’t naive about future AIs, and she devotes a large portion of her book to the idea in its title – that AI systems are a mirror of their human makers. Mirrors, of course, don’t just reflect and reveal things as they are, Vallor writes, they “also magnify, occlude, and distort what is captured in their frame”. There has been a lot written about the human biases AIs can perpetuate, but to see this laid out at length, and linked to other problems like climate change or zoonotic viruses, is more than a little unnerving.
In the end, writes Vallor, the only way to guard against the worst downsides of future AI will be to rethink human values and how we design our tech. This involves shifting away from commonly prized traits like productivity and confidence, which, she argues, have steered us towards societal collapse, to more obscure virtues such as Aristotle’s phronesis, a kind of wisdom that concerns practical understanding and sound judgement.
Vallor is clear on how difficult this wholesale redirection of our moral compass would be, but is nonetheless optimistic about our capacity for change, especially with a powerful enough vision, giving us the “civil courage” to collectively start repairing and rebuilding the world for others.
The questions and viewpoints in these books aren’t easy to resolve: the gulf between them over what AI is and may become is frankly vast. But while this technology is difficult to understand and predict, it isn’t magic. We would do well to listen to experts like Vallor or Lawrence to discover how it really works, rather than succumb to fantastical visions of the future.
[image_container wp-image=2433175] [/image_container]
Nick Bostrom (IdeaPress Publishing (on sale now))
[image_container wp-image=2433176] [/image_container]
Ray Kurzweil (Bodley Head (US, 25 June) (UK, 27 June))
[image_container wp-image=2433174] [/image_container]
Neil Lawrence (Allen Lane (UK, 6 June) PublicAffairs (US, 3 September))
[image_container wp-image=2433177] [/image_container]
Shannon Vallor (Oxford University Press (3 June))
