Doomsday scenarios and few concrete proposals: AI’s ‘Greta moment’

There are no glowing wonders streaking across the sky, save perhaps for the odd illuminated drone, but the atmosphere of medieval millenarianism is palpable. A highly technological form of millenarianism that seems to promise the extinction of the human race, caused by a supremely powerful and out-of-control artificial intelligence. CNN, interviewing a seemingly repentant former AI researcher, runs the headline ‘Children may not make it to secondary school’. This was broadcast just a few hours after the story of Jacob Coxon, himself a researcher at OpenAI and Anthropic, who was alarmed by the accelerating pace of developments and the lack of awareness amongst the tech giants: following his resignation, publicised in the Wall Street Journal and on X, Coxon has become the standard-bearer for a school of thought and a narrative that are now rife in public debate and on the political agenda. The doomsayers do not climb mountains as penitents wearing hair shirts; instead, they give interviews, write books and inundate us with images of a pitch-black horizon: they never say exactly how or when, but merely paint the absolute catastrophe in the flesh-toned colours of a John Martin painting. Eliezer Yudkowsky, one of the best-known doomsayers in the tech sector, has, ever since 1996 – when he was just sixteen – been rattling off dates and scenarios for an apocalyptic technological singularity, only for them to be repeatedly debunked, retracted or revised. More astutely, thinkers such as Max Tegmark, author of Life 3.0, or Nick Bostrom of Superintelligence believe that the development of a superintelligence represents a supreme existential risk, urging developers and researchers to exercise the utmost caution. A radicalisation, if you will, of Hans Jonas’s principle of responsibility, but with a somewhat mystical and darker tinge.

The idea underlying this apocalyptic scenario is that of recursive self-improvement: in other words, artificial intelligence that has evolved to such a general level that it begins to improve itself, until it reaches the point of no return. Ever since Nobel laureate Geoffrey Hinton estimated the risk of super-AI annihilating the human race at 10 per cent, doomsayers have rushed to update their estimates and, like a bunch of not-too-random parrots, have adopted that very same percentage. Behind the apparent folklore – however apocalyptic it may be – far more concrete stakes are at play. On the one hand, Anthropic and OpenAI have been riding a certain narrative for some time: ‘humanising’ their models, as in the case of the attack on the HuggingFace platform, when expressions such as ‘loss of control’ and ‘runaway’ were used, or by releasing increasingly alarming reports and press releases, such as the most recent one concerning the alleged thwarting of a bioterrorism attack devised using Claude. Behind every apocalyptic scenario, there often lie far more prosaic interests: Anthropic is about to go public, and despite the doomsday aura, it does not appear that Amodei’s company intends to back down from that plan. And indeed, the head of Anthropic has published, in real time, the essay ‘We Must Pace the Frontier’, calling for a slowdown in the development of the models. Surprisingly, Sam Altman and Elon Musk agree with him; Musk has gone from regarding Coxon’s claims as a conspiracy to siding with Amodei in the space of three days. This call for a slowdown does not appear entirely altruistic; it seems, rather, a noble-sounding attempt to prevent the use of open-source and open-weight models, which are becoming increasingly widespread.

Meanwhile, Dean Ball, a former adviser to Trump and now at the helm of OpenAI, is taking steps to engage Congress and push for strict regulatory safeguards. The Democrat Bernie Sanders has taken the same approach; his proposed legislation provides for up to twenty years’ imprisonment for researchers found guilty of circumventing security regulations. On the progressive front, there are interests of a different kind at play. As Politico points out, behind the calls for degrowth and a slowdown in technological progress – such as the movement against data centres – lie organisations that view China favourably. During the American ‘Socialism 2026’ convention, the streamer Hasan Piker openly argued that the political struggle against data centres forms part of the socialist political agenda. The paradox of these apocalyptic or self-serving positions is that they pollute the debate and obscure the real issues posed by AI. But this is a relative paradox, if one considers, as Pedro Domingos points out, that Skynet – the superintelligence in "The Terminator" that seeks to destroy the human race – slipped out of the control of a laboratory obsessed with AI safety. Moreover, the doomsayers have little to offer in concrete terms: their vision of deceleration relies on utopian multilateral international cooperation. It is certainly true that the call to decelerate was also made by the Pope in the encyclical *Magnifica Humanitas*, but socialist China does not appear interested in acting on that invitation: rather, it uses it as soft power to slow down Western technological efforts. In this regard, one should read Breakneck by Dan Wang – which will be published in Italy in a few weeks’ time by Il Mulino under the title A rotta di collo – to understand just how fast China is racing and how much further it intends to accelerate.

Europe, on the other hand, does not need a debate on whether to slow down, given that it appears to be structurally sluggish. Mario Draghi also points this out in a lengthy article of his that appeared in the Financial Times on 11 September, bearing the highly significant title: “Europe’s difficult choices on AI”. Draghi notes how Europe has slowed significantly since the pandemic and how many of its problems – from excessive regulation to the fragmentation of demand – remain unaddressed. In the field of AI, moreover, “AI-driven growth conflicts with the desire for European sovereignty, given that Europe has little control over the AI value chain”. Dependence on American and Chinese models gives rise to unpredictable scenarios, and being left out of the technological race means being left out of history. Draghi admits that “the potential scope for European sovereignty in the field of AI is narrow”. The solution identified by the former Prime Minister is, first and foremost, the creation of virtuous cycles of unified demand aggregation and AI usage, particularly with regard to ‘neoclouds’. The example he cites is that of the European companies ASML, Capgemini and Amadeus, which have committed to purchasing, for several years, the European Compute Units from the equally European firm Mistral. This mechanism, being widespread and participatory, would prevent the consolidation of tech oligarchies. The key to implementing such a project lies in data: Europe possesses massive amounts of data of enormous value, estimated at around 800 billion euros. This raises the politically sensitive issue of the storage and safekeeping of this data. The question of data centres, on which Draghi dwells at length, is the political issue par excellence of our times and one that troubles both Americans and Europeans alike, even though in Europe there is greater scope for expanding such facilities. The data centre marks the return of technology to the local landscape; it is the ‘artificial epidermis’ which, to quote Benjamin Bratton, shatters the perception of complete virtualisation: this leads to NIMBY syndromes, opposition and political friction, as we have clearly seen in Italy, where the project for a mega-data centre in Travagliato, near Brescia, appears to have been definitively scrapped. This is precisely what highlights the real ‘uninvited guest’ in Draghi’s reconstruction plan: political decision-making. Without it, we will head straight towards our own personal apocalypse, cast out of history.