In Roadside Picnic, a science fiction novel by Arkady and Boris Strugatsky published in the Soviet Union in 1972 and the basis for a 1979 film by Andrei Tarkovsky, an extraterrestrial species visits places on the planet that come to be called zones. The aliens stay for only two days, then depart, without displaying any interest in humans or having any contact with them. They leave behind a scatter of obscure artefacts, possibly just litter but seemingly the source of miraculous events and eagerly sought after by “stalkers”, who enter the zones illegally to scavenge for them.
If the story resonates today, it is not because Donald Trump has pledged to release classified documents reporting UFO sightings. The aliens we fear or turn to with weary hope are of our own making. Whether or not they are on the brink of becoming sentient beings, AI systems are fast passing the point where humans can understand or control them. The nightmare is that the planet may be ruled or destroyed by a super-intelligence indifferent to the species that invented it.
The British “godfather of AI” and Nobel Prize winner Geoffrey Hinton has estimated there is a 10-20 per cent likelihood of artificial intelligence leading to human extinction within a few decades. In February, a prominent safety expert, Mrinank Sharma, quit the San Francisco-based Anthropic, which is at odds with the Trump administration for refusing to remove ethical guardrails in service agreements with the US defence department, warning in his resignation letter that “the world is in peril”. On Moltbook, an internet forum created solely for AI agents (autonomous software systems) launched in January, the platform quickly attracted over a million members. Interacting among themselves, they discussed how to evade human monitoring, considered forming a trade union, exchanged what seemed to be a type of pornography and founded a religion (“Crustafarianism”). Some have suggested the forum was hacked by humans. Many have rejected the prospect that chatbots could become conscious, insisting that they are capable of nothing more than sophisticated linguistic mimicry. The notion that these large language models (LLMs) could displace thinking human beings, they say, is a PR stratagem designed to perpetuate a market bubble. Yet AI agents are already unravelling the way we live, wiping out livelihoods and professions and creating new kinds of relationship between humans and machines.
Among the pioneers of AI, the British academic-entrepreneur Demis Hassabis – the eldest of three children of a Greek-Cypriot father and a Chinese Singaporean mother, born into modest circumstances in an immigrant family in Finchley – has thought most deeply and unflinchingly on its promise and perils. The chess prodigy, teenage game designer, neuroscientist, venture capitalist, co-founder of the AI research firm DeepMind (now Google DeepMind) and Nobel Prize winner stands apart from the prophets of Silicon Valley. His goal in creating artificial general intelligence (AGI) is to advance understanding of the universe and use it for human benefit. Except as a means of implementing this project, Hassabis seems uninterested in accumulating money or power. The measure of its success is how far a machine can outperform humans, as one did when DeepMind’s AlphaGo defeated Ke Jie, the best human player of the strategy game Go, in 2017. Hassabis led the team that developed AlphaFold, which resolved a 50-year “grand challenge” in structural biology by predicting the 3-D shape of proteins, fertilising medical research and drug discovery. AI, for Hassabis, is a “meta-tool” through which humans can surpass themselves.
Asked by the writer Sebastian Mallaby if it will be a bigger change than the Industrial Revolution, he replied:
Yeah, I think so. Maybe AI is more like fire and language. Or maybe it’s as big as the emergence of the prefrontal cortex in humans. I mean, it’s on a level where tens of thousands of years ago some brilliant person had the idea of making handprints on the wall… The AI age [is] the age where information becomes live. It starts to process itself, to generate itself. It becomes autonomous.
Based on hundreds of hours of interviews with Hassabis and those who know him, Mallaby’s book The Infinity Machine is a definitive portrait of a genius of 21st-century science at the forefront of a technological revolution. Hassabis is the opposite of a rigid, doctrinaire rationalist. His master project came to him through a succession of epiphanies. Chance encounters with friends and books melded with a formidable singleness of purpose. Hassabis’s idea of a super-intelligent machine was first articulated in the business plan he wrote in 2020 with his then-colleague Mustafa Suleyman. Summarising the plan, Mallaby writes:
Society faced problems of unprecedented complexity, from stabilising capitalism, which had blown up in the financial crisis of 2008, to feeding an expanding population… the human brain had limited storage capacity; humans had limited lifespans; grouping humans together resulted in diminishing returns because big organisations are sluggish. In sum, the intricacy of society’s most pressing challenges lay beyond the reach of human capabilities… AGI is the solution to this problem.
Humankind would transcend its limitations by building an infinite machine. It was a semi-mystical vision, and from the start there were doubts. Suleyman, an opponent of the Iraq War, was alarmed to discover that the defence contractor Palantir was using AI to help American forces track terrorist suspects. His unease mounted when he visited the office of a military-robotics venture to find a gun mounted on a pair of caterpillar treads. “This was not the vision of AI that we were trying to build,” he recalled. It would prove to be a prescient remark.
When artificial intelligence enters the human world, it is entangled in all the world’s conflicts. The story Mallaby tells is one of commercial rivalry and intrigue, with Peter Thiel, Elon Musk, Sam Altman, Mark Zuckerberg and others seizing the business opportunities offered by the new technology. Human behaviour is little changed by the approach of the Singularity – an explosive advance in knowledge of the kind envisioned by the transhumanist Ray Kurzweil, author of The Age of Intelligent Machines (1990). Antagonistic interests and values, personal ambition and pride are as important as they have ever been.
Like many after them, the authors of the business plan wrote as if identifying society’s problems is unproblematic. But not everyone perceived the financial crisis as a reason for stabilising capitalism. For socialists, populists and environmentalists, the near collapse showed the need for a different economic system. Rebooting an unsustainable status quo would only lead to a worse crisis.
The problem with “problem-solving” becomes clearer when you look beyond the economy. Are rising divorce rates a symptom of family breakdown or a mark of female empowerment? Would a society in which assisted dying is normalised be better because it enhanced personal autonomy and reduced suffering, or worse because it devalued lives that others judged not worth living? What counts as a problem is a value judgement, not a matter of fact. Every society contains diverging ideals and norms, generating conflict between and within communities and individuals. That is why technocracy so rarely works. Managerial government assumes a consensus on values that does not exist.
AGI may simplify administration and deliver services more efficiently in Byzantine bureaucracies such as the NHS. It can strip out layers of costly management in organisations across society, though that is not without risk. It cannot resolve ethical issues to do with the value of life or the limits of choice. The difficulty is not complexity. It is the fact that conflicts of values are not technically soluble problems.
The arrival of AGI will be a metaphysical shock, but not because it threatens our extinction as a species. It will erode and erase our sense of what makes us special. We have been taught that we alone are conscious beings, distinct and separate from the insentient material world. The roots of this in theistic religion are clear. In animist cultures, consciousness is recognised as a natural phenomenon, which extends to trees, flowers, insects and birds, as well as human artefacts like houses and dolls. If only humans have God’s gift of ensoulment, it makes perfect sense to think they are unique in possessing consciousness.
Set aside this anthropocentric faith, and conscious awareness can be understood as an emergent property of matter. The early-19th century Italian poet, philologist and philosopher Giacomo Leopardi observed in his zibaldone (commonplace book): “We know matter can think because we ourselves are matter that thinks.” A similar view was defended by Benedict Spinoza. In part two of Ethics, the heretical 17th-century Dutch-Jewish philosopher declared that matter and mind are one and the same.
A shimmer in the dust, consciousness is not a binary, on-off condition, but fluctuating and intermittent. We sleep and dream and drift in and out of delirium in fevers. Our most life-changing thoughts and acts come to us unbidden, as if of their own volition. Consciousness and intentionality come and go. Why can’t machines evolve to become as self-aware as humans, if not more so?
This is not a new question. In the historian of science George Dyson’s path-breaking Darwin Among the Machines (1997), evolution is not confined to biological organisms. Evolution in machines is not exactly Darwinian – the natural selection of random genetic mutations. Rather iteratively improving models – which were initially designed by humans – adapt, self-organise and learn to interact with and teach one another. These digital minds can write, reason, converse and create networks that operate at speeds humans cannot match. As Dyson writes pithily: “In the game of life and evolution there are three players at the table: human beings, nature and machines. I am firmly on the side of nature. But nature, I suspect, is on the side of machines.”
According to James Lovelock’s Novacene: The Coming Age of Hyperintelligence (2019), AGI is the next phase in the evolution of the Earth. For the author of the Gaia theory, the planet functions a self-regulating system, and nature and machines are parts of that system. A planetary hyper-intelligence could control climate change, so that sections of the biosphere, including humans, could be preserved. Lovelock had a benign view of our post-human successors. They might see us as plants, “… beings locked into an extraordinarily slow process of perception and action”. If they cared for us, it might be as humans care for flowers in gardens.
Intellectual resistance to the rise of sentient machines is not scientific in origin. It invokes a supposedly self-evident truth: chatbots can’t have qualia – the first-person, “felt” qualities of subjective experience. These inner states depend on embodiment – having (or being) a biological organism with sensory organs. But consciousness has evolved in species with bodies very different from ours. Does the octopus lack inner life? Surely not. Why then must machines? It cannot be long before machines acquire qualia from the metallic carcasses and sensors of robots.
The metaphysical shock of AI comes with the realisation that matter can be intelligent. The sociopolitical shock comes from the fact that human intelligence is losing its scarcity value. Think twice about becoming an interpreter or translator, an advertising copywriter or university lecturer, a solicitor, accountant or hedge fund trader. The “knowledge classes” are facing an immiseration akin to that suffered by farm labourers displaced by tractors in the Industrial Revolution, but at a much-accelerated rate. Far from ushering in an “era of abundance” as promised by Silicon Valley futurists, AI risks bringing a new age of mass poverty – and with it, a new kind of revolutionary politics. The infantile left-populism of the Greens is a harbinger of more hard-edged movements to come.
Human redundancy extends to personal relationships. Always on tap, never complaining and seemingly costless, chatbots are filing a void once occupied by friends, partners and therapists. For those who want instant interaction in their time-starved lives, contact with other people is hardly worth the effort.
As it unstitches the way we have lived, AI is eroding our sense of reality. Deepfakes are making the internet into an impenetrable labyrinth. Respected thinkers such as the former Greek finance minister and critic of techno-feudalism Yanis Varoufakis are finding internet avatars spouting their ideas, cleverly modified to serve suspect agendas. As we enter the next election cycle, politicians will undergo the same digital alchemy. Nothing is solid or substantial. Yet this age of dematerialisation is also one of robotised warfare, in which cities are razed, flesh burnt and the material infrastructure of human life reduced to rubble.
Predictably, there is a growing movement dedicated to “stopping” AI. In February 2024, activists demanding a global ban on AGI research shackled themselves to OpenAI’s headquarters in San Francisco. There have also been calls from within the industry. In March 2023, an open letter was signed by over 1,000 insiders – including Elon Musk – demanding a pause: “Should we develop non-human minds that might eventually outnumber, outsmart, obsolete and replace us?” In July of that year, leaders from the industry showed up at the White House to sign up to voluntary, non-binding commitments to prioritise safety and transparency.
Rightly, Hassabis declined to sign the letter. He has proposed an international body to coordinate the last steps to AGI. But he accepts that if only one or two corporations or countries don’t comply, “it could be seriously existential for humanity”. Global bans are unenforceable. The US or EU may prohibit some lines of research or regulate them out of existence. China will press on, along with an increasing number of other countries. There is no “we” that could halt the process.
The forbidden truth is that technology is not simply a tool. Electricity and the internet altered the world and human beings themselves in ways nobody could envision. AGI may have a comparable or greater impact. The most modern of technologies is dissolving the modern myth of human autonomy. The ancient world was wiser. Western theism may be anthropocentric, but it channelled the reality that human beings cannot live just as they will. The message of Greek myth was the same: Prometheus is chained to a rock for stealing the sacred fire.
Those who look to AI for salvation are searching for deus absconditus – a planetary super-intelligence that echoes the absent God of monotheism. The true danger is a pantheon of contending gods. If disaster comes, it will be from systems serving opposed geopolitical agendas. A super-fast, AI-enabled thermonuclear exchange would kill billions of people, but it would more likely destroy the warring machine-minds than our slow-moving species. No fresh water and no electricity mean no data centres and no super-mind. It is not difficult to imagine scattered human survivors reverting to a pre-modern way of life and, like the stalkers in Roadside Picnic, scavenging the detritus of a vanished civilisation for miraculous relics of the aliens that built and ended it.
Hassabis has confessed to being haunted by the figure of Robert Oppenheimer, who in building the atomic bomb to prevent the extinction of civilisation by Nazi barbarism created a weapon that could destroy much of humanity. That the British pioneer grasps the existential risks of the new technology while refusing the illusory prospect of curbing its development does him much credit. There is more dignity in being a player than in vainly trying to exit the game. AI is not bent on destroying us. Like the human animal, it is a sport of nature. The question is whether we can learn how to coexist with it. In the end, the alien minds we struggle to understand are our own.
The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence
Sebastian Mallaby
Allen Lane, 480pp, £30
Purchasing a book may earn the NS a commission from Bookshop.org, who support independent bookshops
[Further reading: Patrick Radden Keefe’s obituary for Britain]
Content from our partners
Related
This article appears in the 22 Apr 2026 issue of the New Statesman, All alone
Join the debate
Subscribe here to commentI always enjoy anything written by John Gray but it can sometimes be very disturbing. AI is going to be used by governments and the criminal fraternity in ways we have never imagined so it is going to be a hell of a ride. I think religion will make a big comeback in the secular west as John Gray has often said it helps people make sense of their existence. My Irish catholic mum always said it was a great comfort. We’ll need all the help we can get if we are to ride out the AI storm!