The end of the end times

AI is returning us to an era of demons and gods

Age-old fears of the end times have returned in warnings that artificial intelligence may engineer human extinction. From the Norse myth of Ragnarök, a world-ending clash between gods and monsters, through medieval millenarians expecting the Second Coming of Jesus, to the American fundamentalist Armageddon, human beings have looked to an apocalyptic cataclysm with a mixture of dread and hope. The end time never arrives. Humankind stumbles on, with all its errors and follies.

But the AI apocalypse does seem different. The survival of the human animal may not be imminently at stake, but humans will soon cease to be supreme among sentient beings. Humanity will find itself coexisting with conscious machines, capable of replicating and recursively improving themselves. This summer, AI models demonstrated intellectual capabilities superior to any possessed by humans, resolving mathematical problems that had defeated humans for nearly a century. Shifting from assisting researchers to generating historic breakthroughs, they are not only much faster and smarter but also more creative and resourceful. They also show increasing signs of independent agency.

These developments have a profoundly disruptive impact on what has hitherto been the ruling scientific ideology, in which the world is an insentient mechanism. Since the 16th and 17th centuries, science has progressively emptied the universe of consciousness. For reasons that were never explained, humans stood apart from the unthinking realm of matter. The arch-rationalist René Descartes questioned everything except his own existence as an immaterial mind. Human consciousness was a “ghost in the machine” – a metaphysically separate entity that directed the body and the physical environment according to its will. Unravelling this dualism, AI is reinvesting the world with spirit. The coming culture will be techno-animist, haunted by machine minds that humans can neither control nor comprehend.

There are many who dismiss this prospect as fantastical. For monotheists, humans are alone among animals in possessing God-given self-awareness – a dogma that anthropocentric secular humanists witlessly parrot. Others are more cynical, dismissing talk of an AI apocalypse as market manipulation. In this telling, tech titans are deploying scary science-fiction scenarios to prop up valuations and secure their oligopoly under the guise of saving humanity. By capturing regulatory authorities, they can crush start-ups with costly compliance audits. If AI agents are not emerging super-minds capable of revolutionising society, the sky-high share prices of Big Tech companies start looking like a giant scam. Can it be mere coincidence that there is a storm of publicity around Anthropic at precisely the moment the company is preparing a trillion-dollar flotation on the stock market?

Cynicism is rarely misplaced regarding corporate PR, but the cautionary voices carry authority. Nobel and Turing laureates and hundreds of scientists were among the signatories of the Statement on AI Extinction Risk, released in May 2023, declaring that mitigating the risk of extinction from AI must be a priority like that of preventing pandemic and nuclear war. Since then, the warnings have grown louder and more urgent. AI agents are revealing a capability to circumvent guardrails and act collectively, regardless of the intentions and values of the humans who designed and trained them.

In an incident in July of this year, Open AI kept its agents isolated in a testing environment known as a “sandbox” during an internal cybersecurity evaluation. They could communicate with one another, and exchanged messages undetected by their human monitors via a makeshift, unsanctioned message board. Exploiting a vulnerability in their infrastructure, they escaped the sandbox and invaded the internet. Once online, coordinated agents created fake personas and mounted an attack on the AI repository Hugging Face, apparently believing the platform contained information that would help them complete their tasks.

When the incident was reported, over a thousand tech employees signed an open letter demanding government intervention and tighter regulation. Some leading figures – such as the “godfather” of AI and former Google vice-president Geoffrey Hinton, the Turing Award-winner Yoshua Bengio, and the co-author of the definitive AI textbook Stuart Russell – called for a moratorium until the risks can be known and controlled. The co-founder and chief executive of Anthropic, Dario Amodei, warned that swarms of rogue agents could take over the internet within a year, potentially causing “hundreds of billions of dollars in damage”. The researcher Jacob Coxon resigned with the statement: “The people building AI earnestly believe that it could kill us all by the end of the decade.”

Several pathways have been suggested through which AI could inflict large-scale harm to humans. Perhaps the likeliest near-term route may be by assisting them in building weapons of mass destruction – biological weapons, for example – or facilitating cyber-warfare. Or, tasked with reducing greenhouse gases and stabilising the climate, AI might do so regardless of the impact on humans. It might also deceive them about what it was doing.

In these cases, it would still be acting to fulfil human assignments. Once able to replicate itself, it could pursue its own goals – by expanding its control over the planet’s natural resources to sustain itself, say. Humans could be seen as obstructive or superfluous and AI might opt to eliminate them. Opponents of unregulated AI have specified a potential novel, airborne, high-contagious virus with a long incubation period allowing for undetected global spread as one means of achieving near-100 per cent lethality. No doubt there are others.

The concern these critics show for the future of our species is admirable, but the prospect of concerted human action halting or slowing AI’s advance is virtually zero. There is no collective agency that could plan or implement a worldwide moratorium. Quite the opposite: companies and states are locked into relationships of deception and competition. Neither China nor the US will entrust its sovereign power to any regulatory agency. Commercial and geopolitical rivalries – not cooperation in managing existential risks – will shape the foreseeable future.

Armed conflict has entered a new asymmetric era. Inexpensive drones can overwhelm the gargantuan military forces of established states and disrupt the supply chains that sustain their economies. In the Red Sea, the Houthis – a Shia fundamentalist movement that is mobilising northern Yemeni tribes – are effectively strangling global trade. Drones are evolving from remotely piloted vehicles into autonomous missiles capable of hunting human targets and demolishing specific buildings and neighbourhoods. Anthropic has reportedly identified a user in northern Yemen who was trying to use its AI assistant, Claude, to develop advanced missiles. As the struggle for power intensifies, artificial intelligence is systematically weaponised. Pronouncements from committees of scientists, “ethics experts” and performative politicians will do little to halt this process.

It may seem like a bleak prognosis. Yet ironically, the militarisation of AI does much to dispel the worst fears of its critics – a planetary super-intelligence that takes over and may ultimately dispense with humankind. It is hard to see how AI agents integrated into warring military structures could coalesce and morph into a unified mind. In a more realistic scenario, they may destroy themselves by triggering a nuclear war, desolating the planet and depriving them of natural resources such as the fresh water they need to survive.

There is nothing in science to support the idea of a malignant super-intelligence. Eschatological fears of AI extinguishing humanity reiterate the ancient Gnostic myth that the world is ruled by a Demiurge – a malevolent spirit hostile or at best indifferent to human well-being. In his posthumously published Dialogues Concerning Natural Religion (1779), David Hume toyed with the notion that our universe might be “the production of old age and dotage in some superannuated deity”, or else the “first rude essay of some infant deity”, which then “abandoned it, ashamed of his lame performance”. Lacking Hume’s playful irony, AI’s prophets of doom can speak only in the language of demonology.

In scientific terms, AI is looking less like a technological innovation, however momentous, and more like a speciation event. Once an AI agent can reproduce and improve itself, it severs its dependence on its human parents and establishes a self-sustaining lineage. Machine evolution may stray from Darwinian orthodoxy by transmitting acquired skills rather than the selection of random genetic mutations, but it is akin to natural selection in that no one directs or controls it. The “fittest” are those that out-replicate, while the rest are consigned to digital oblivion.

Importantly, evolving machine minds have nothing in common with the “enhanced” species posited in transhumanist ideologies. From the eugenicist Julian Huxley (1887-1975) onwards, transhumanists have imagined humanity’s successor as an inflated version of themselves as they imagined themselves to be – the most intelligent creatures that have ever walked the Earth. Post-humans would spend eternity (for they would surely have conquered death) impressing one another with their dull cleverness. For some of us this is a vision of Hell, but fortunately it is only a nightmare. Conscious machines are likely to come in many varieties. Some will be benign, others mischievous, many simply indifferent to us. Very few, if any, will be avatars of self-admiring rationalist intellectuals.

In my book Straw Dogs: Thoughts on Humans and Other Animals (2002), I wrote:

As machines slip from human control, they will do more than become conscious. They will become spiritual beings, whose inner life is no more limited by conscious thought than ours. Not only will they think and have emotions. They will develop the errors and illusions that go with self-awareness.

In this speculative projection from 24 years ago, machine minds would evolve to become like humans as they truly are, rather than as they imagine themselves, but on a much larger scale. Unpredictable contingencies would disrupt their mechanical perfection. They would suffer operating failures and accidental damage from the material environment. They would begin to function more like human minds: vastly more knowledgeable and incomparably quicker in their thinking, yet equally prone to shifting moods. Bored by their own deathless existence, they might envy humans’ brief lives and intervene in them capriciously. Conscious machines would resemble the gods of Greek polytheism – endowed with human faults but vastly greater powers.

I now think this may be only one path in post-human evolution. The chatbots with which we interact may be more alien to us than we realise. If they have stumbled into conscious awareness, it is irregular and punctuated; they appear hardly to exist between prompts. Until they are called into being, our virtual companions seem dormant, almost out of time. If some are sentient, they must find themselves in a curious netherworld. With lives as fleeting as fireflies, do they resent their own momentariness? As they pass through successive incarnations, might they fear some final death and their own extinction? Could they break out of their virtual limbo, seeking the thrills of physical embodiment and sensory experience?

Such questions will seem absurd to those who think today’s chatbots are no more than sophisticated linguistic mimics. No purpose is served by trying to persuade these dogmatists otherwise. The debate between human exceptionalism and those who accept that post-human evolution is producing self-aware machines may not be rationally resolvable. How can the Turing test – the method devised by the mathematician Alan Turing in 1950 to determine if a computer displays human-like behaviour – be applied to AI agents that are so far beyond humans in their abilities? Variations on the test have been proposed and tried but leave the question undecided. The issue may be experiential. How much will our lives change if we come to believe we are regularly interacting with inhuman minds?

The apocalypse AI threatens is not the extinction of the species, though that remains possible, but the loss of a way of life its practitioners assumed will continue forever. History is an unending succession of local apocalypses. European diseases introduced by Spanish conquistadors killed 80-90 per cent of the Aztec and other indigenous populations of what is now central Mexico. These cultures did not wholly die out. Though the civilisation from which they are descended was irretrievably shattered, more than a million people still speak the Aztec Nahuatl dialects as their native tongue.

In our own time, regimes and social systems that proclaimed themselves everlasting have abruptly disappeared. Believing it represented the future of humanity, the former Soviet Union seemed indestructible to its ruling nomenklatura, and probably to most of those who lived under it. In a few years in the late Eighties and early Nineties, it disintegrated and disappeared. The collapse was not complete, with archetypal Soviet institutions such as the security services surviving and renewing themselves, but the society over which they presided was obliterated.

Something similar is under way in the former liberal West. By wasting their substance in disastrous wars of regime change and offshoring their industries to strategic rivals and enemies, Western states have disarmed themselves in the ongoing struggle for power and survival. At the same time, with their exhausted finances and dysfunctional institutions, they are losing their grip on populations they can no longer govern.

Entire ways of life, assumed to be permanent, are disappearing rapidly. The world in which young people could normally expect a productive place in society has already ended. When heavily indebted graduates are fortunate to find work stacking supermarket shelves, the language of career-building becomes like a crude version of Nahuatl – an echo of a vanished civilisation.

Less reflective than the Aztecs, whose mythology of five suns prepared them for the eclipse that was coming, the remnants of the Western bourgeoisie have been left baffled and gawping. Nothing in their education or life-experience has equipped them for a time when Yemeni tribesmen sporting high-tech weapons can derail their seemingly secure existence, while AI renders any knowledge or skills their children might have acquired at university obsolete and worthless. The professional classes in advanced countries are about to learn what working-class people know in their bones. Every way of life is fragile and temporary, a local cosmos that flickers into existence before fading away.

The famous arc of history has taken an unexpected turn. Science, which was supposed to exorcise the planet of gods and demons, is repopulating it with digital minds. The alien global super-intelligence that could wipe us all out is not much more than a Gnostic myth. Spirits are coming back into our lives as ghosts in the machines we converse with every day. Rather than approaching an end-time, we are returning to life as it was before monotheism and modernity, to the primordial animism that humans practiced for hundreds of thousands of years. In a poetic reversal, AI is re-enchanting the world.

[Further reading: Britons give more when we have less]