Apocalyptic and Buried

Excuse me, what time is the apocalypse? And where will it catch up with us? Amongst the Milanese “robots” of Vogue? Or amongst the Neapolitan America’s Cup regattas and pre-regattas? Many are now convinced it is imminent (although precise details would allow for more effective preparation, whereas we continue debating cycle paths, electoral law, drug-addicted female teachers and Vannacci’s altered images of female undersecretaries). Will the AI apocalypse arrive first, or will it be a Third World War or a climate catastrophe? If only we knew.

Let us proceed methodically. In recent times, reading the morning papers requires Xanax, as well as coffee and a brioche (or a cornetto in Rome). World War III? “A perfectly plausible scenario,” says Avril Haines, former head of U.S. intelligence, during the customary ‘Lunch with the FT’ – the Financial Times’ lunchtime interview – while enjoying roast salmon with peach salad at the Terrace Café in the London Library (bill: £56.30 for two – not that much, actually, less than at an Autogrill; incidentally, perhaps we’ll die of inflation caused by Rustichella).

Where will this world war start, Countess? Where you least expect it. From the British airbase at Fairford, from where American bombers take off bound for Iran, and near which, on the night between Saturday and Sunday, five men were stopped with three vans and incendiary devices. Right in the heart of the Cotswolds, a sort of large English Capalbio where everyone now goes to live, amongst ornamental chickens and old Land Rovers, in those typical little villages no more than two hours from London.

Consider the Americans who left California to move there, such as Ellen DeGeneres (many quickly put their properties back on the market: too much mud). Meanwhile, in the heart of California, in Berkeley, at a co-working space called Constellation (let’s hope it’s not linked to the one in Crans-Montana), self-awareness groups are being held on what the apocalypse will be like and how to deal with it: should we buy the usual private islands, stockpile iodine tablets, or move into electromagnetically shielded bunkers in the desert?

At an event in 2022, guests were served a cocktail called "Death With Dignity", named after an essay arguing that humanity was already doomed. According to yesterday’s Wall Street Journal, these lively gatherings are dominated by employees of Anthropic, the reflective AI, the Italian-American AI, the one created by the Amodei brothers: in short, Claude, which is the one we all use for writing. However, it still cannot write articles or novels properly on its own, as the Goncourt jury members believe; they excluded a certain Thélyson Orélien on suspicion of having had the AI write his novel (he denies it; the AI has not commented).

Anthropic’s employees do not yet seem to be writing best-selling novels, but are instead exchanging apocalyptic scenarios and advice on how to prepare for them in a private Slack channel, and have adopted a motto to stick on their laptops: “Things will never be chill again” (“We’ll never be at ease again”). According to the WSJ, the interesting aspect is that the debate about AI wiping us all out began many years ago, before AI itself even existed (the debate among the new AI doomsayers is essentially the broader Italian debate on the subject).

The forefather of the doomsayers is Eliezer Yudkowsky, a self-taught thinker from the Bay Area who, in the 2000s, wrote a blog and inspired these gatherings of ‘doomers’ (who are not followers of the romance novelist Erin Doom). In 2022, a group of doomers from Anthropic and other laboratories flew to a retreat on the island of Eleuthera in the Bahamas, where they discussed the risks of AI while enjoying sunset yoga, cliff diving and a sea run – even biots were on the cards. The highlight was a talk by Yudkowsky entitled ‘A Disorganised List of Reasons for AGI Doom’ (AGI is not the prestigious press agency but the ultimate AI – the one that doesn’t exist yet, but which might one day, like the winning long shot).

Then there’s the Wall Street Journal, the newspaper owned by the Australian tycoon – perhaps Murdoch knows something we don’t? And above all, at 95, what does he care if the apocalypse comes? – which, a few weeks ago, made things clear with a pragmatic take: yes, OK, we’ll die because of AI, but exactly how? Oh, at last. Usually, the newspaper explains, doomsayers aren’t afraid of ‘Terminator’-style robots, but of two things: the loss of control and malicious use by human beings. There is also a much-cited thought experiment: a super-intelligent machine tasked with producing as many paperclips as possible might decide to turn all matter on Earth—including human beings—into paperclips. Objectively speaking, extermination by transformation into paperclips had not been previously imagined.

Meanwhile, the warnings are countless. In early September, Jacob Coxon, a researcher at Anthropic, resigned, stating that his former employer and rival OpenAI were racing to develop technologies that “could kill us all by the end of the decade”. And if anyone thought he was the only one to believe this, a few minutes later a colleague, Evan Hubinger, confirmed: “We truly and sincerely believe that AI could kill all human beings!”, estimating the risk of extinction in the next decade at over 10 per cent. Then there was Hugging Face: this summer, hundreds of OpenAI agents escaped from a test and hacked the platform to copy the answers. But according to Axios, there are far more cases of rogue AI – let’s call them ‘maranza’ AIs – with OpenAI, Anthropic and external researchers currently examining tens of thousands of them (fortunately, none of them are yet talking about turning us into paperclips).

Daniel Kokotajlo, a former researcher at OpenAI, founded the AI Futures Project, which last year published “AI 2027”: a scenario in which super-intelligent AIs marginalise humans and, by the mid-2030s, decide that we are a nuisance and exterminate us (here, we aren’t even good enough to make paper clips). But what if AI didn’t mind keeping us around? That isn’t necessarily good news, warns the WSJ: some fear a “WALL-E”-style future, with humans gradually ceding control to machines until they no longer understand their own fate, whilst others speculate that AIs will treat us as we treat animals, keeping us as pets or remodelling us through bioengineering. Part Valentino-Giammetti’s pugs in The Last Emperor, part Mars Attacks.

Faced with these scenarios, the most common solution – even amongst those who, shall we say, are bringing about the apocalypse – remains bunkers. At Sam Bankman-Fried’s foundation – the former ‘king of cryptocurrencies’ now serving a 25-year prison sentence and one of Anthropic’s earliest backers – a memo even proposed buying Nauru, a Pacific island nation, to turn it into a ‘bunker/refuge’ in anticipation of an event in which ‘between 50 per cent and 99.99 per cent’ of the population would die.

When asked if he had one, Sam Altman replied: “I’d say no. I’ve got some sort of structures, but I wouldn’t call it a bunker.” He added that if AI goes wrong, it won’t be of any use. Meanwhile, Turin appears to be home to Italy’s first luxury refurbishment project featuring a bunker: Palazzo Alfieri, the former headquarters of a bank on Via Alfieri, complete with a deluxe underground shelter (twelve places, for thirty flats). Who knows what Vittorio Alfieri might have produced, had he had Claude at his disposal – what tragedies, what novels, and how much free time! Never mind being tied to a chair, oh well.