Anyone distressed by the proliferation of “Will AI kill us all?” headlines over the past two weeks would do well to remember the principle articulated by technology journalist Ian Betteridge in 2009. Betteridge’s Law asserts that when a headline takes the form of a question, the answer is invariably no. The law codifies a time-honoured understanding in media that the headline-question format signals a startling assertion unsupported by sufficient evidence.
So how do you approach a story about the imminent extinction of humanity? The answer, it seems, is to tuck it below the fold, somewhere between the Ed Sheeran controversy and the chances of rain tomorrow, then follow up with a flurry of think pieces (such as this one).
The sudden explosion of coverage of the story has introduced millions of people to a highly disputed hypothesis that has been bouncing around for years in the occasionally overlapping worlds of Silicon Valley and academic philosophy. The difficulty for the rest of us lies in disentangling the strands of hyperbole, assertion and blatant self-interest to arrive at some sort of informed picture of what the hell is actually going on. In theory, that’s what the media is supposed to help with.
Scepticism about the predictions of disaster is not new. Yann LeCun, then Meta’s chief AI scientist, accused the bosses of the leading AI labs of attempting a regulatory capture of the industry. That set a pattern that has continued, of different companies adopting very different positions depending on where they sit in the AI pecking order, and of the loudest doomsayers often being the biggest players in the field. Meredith Whittaker, now president of Signal, memorably described these warnings as ghost stories that double as advertisements for a product only a few firms happen to sell.
READ MORE
The current media attention began when Jacob Coxon, a researcher who had worked at both OpenAI and Anthropic, quit the latter with a viral post accusing both firms of gambling with our lives. His colleague, Anthropic’s alignment lead Evan Hubinger, cheerfully agreed that the company earnestly believed its technology could kill everyone, putting the odds at better than one in 10 within a decade. Anthropic’s chief executive Dario Amodei then published an essay urging the industry to “pace the frontier”, arguing that AI systems could soon start improving themselves faster than anyone can supervise.
Should we believe them? Their critics come from all sides. To some on the left the doom talk is a deliberate distraction that pulls attention from the surveillance, bias and data misuse the technology already enables. To others on the libertarian right it is a racket, fearmongering to build a regulatory moat around the handful of firms big enough to clear it. A third group simply doubts the machines are even capable of what is being suggested.
Despite this, the theory that it is all just a scam has to strain to explain some of what is happening now. Some of the most insistent warnings come from people with no model to sell, including two of the field’s founding figures. The uncomfortable possibility is that the incentive to hype and the sincerity of the fear are not mutually exclusive, and that a warning can be both self-serving and true.
All of which gives the media coverage a peculiar flavour. The emotional register wobbles between apocalypse and farce. Washington last week hosted a “pro-human assembly” at which the unlikely pairing of Bernie Sanders and Steve Bannon, presumably bonded by their shared support for the continuation of the species, fretted about the machines going rogue. Meanwhile, practitioners trade in an absurd shorthand, p-doom, for their personal odds on annihilation. The most cited doom scenario, a document called AI 2027, reads like fan fiction, complete with thinly veiled firms named OpenBrain and DeepCent and a few lucky survivors in submarines being hunted down by drones.
In Adam McKay’s star-studded and unsubtle 2021 satire Don’t Look Up, a comet hurtles toward Earth and nobody in power can be bothered to care. McKay’s target was our wilful blindness to climate catastrophe, though the film’s accelerated hysteria seemed out of sync with the actual slow-motion car crash of procrastination that characterises the climate debate. Yet its depiction of venal politicians, messianic tech oligarchs and religious fantasists squabbling while disaster approaches now seems to be playing out on our screens and pages.
[ Will Anthropic’s $2tn IPO make AI safer or more dangerous?Opens in new window ]
McKay’s main focus is the media. In the film a morning television show, all forced smiles and light banter, can’t process the two astronomers’ warning because catastrophe makes for poor television, and the segment is duly sandwiched between celebrity gossip and the weather. The scientist who learns to soften his message is rewarded with fame and a makeover, while the one who refuses to modulate and simply shouts the truth is turned into a reaction meme and hounded off the air. A plain fact becomes a matter of opinion with two respectable sides, and the comet keeps losing the news cycle to a pop star’s break-up. Nihilism triumphs. Fade to black.
Every one of those reflexes has been on display over this past fortnight. But McKay’s comet was real, its trajectory a settled matter of physics. The AI threat itself is what is in dispute, which is why the coverage cannot simply be dismissed as either complacency or scaremongering. The only thing we can confidently predict is that if the answer to this particular headline-question turns out not to be no, then there will be nobody around to say “I told you so”.
