Pathfinders – Panic attacks

The last ripples seem to have faded away from what in January was widely viewed as an extinction-level asteroid impact on the US AI industry. Frantic stock-market gamblers fled in response to what one prominent Silicon Valley venture capitalist described as AI’s Sputnik moment.

Chinese AI firm DeepSeek certainly threw a big rock in the investor pond by approximating US-backed generative-AI capability at a tenth of the usual cost and, what’s more, using old chips and comparatively little training data to do it. Chipmaker Nvidia, formerly rated as the world’s most valuable company, suddenly didn’t look quite so mission-critical, neither did its bleeding-edge and extortionately priced chips, and neither did the AI companies relying on them.

But a few days later the story had sunk like the proverbial stone, as it became clear that the world hadn’t ended. Was the DeepSeek system even as good as early news reports made it out to be? Never mind that it refused to answer questions about Tiananmen Square or make assessments of Xi Jinping’s qualities as leader, a rather churlish objection given that no Chinese AI was ever going to do that. According to counter-misinformation website NewsGuard, DeepSeek was the second-worst performer in a test of 11 chatbots, with a false-claim rate of 30 percent and an unsatisfactory-answer rate of 53 percent (Reuters, 29 January ).

Even so, the panic was almost palpable. Tech pundits had pretty much written off Chinese AI after OpenAI’s ChatGPT launched in November 2022. Even as late as 2024, most thought the Chinese were, if not decades, then probably years behind in AI, partly thanks to a rigidly enforced US government export ban on new chips to China. But in December, when DeepSeek released their V3 Gen-AI version, it became obvious that Chinese firms were perhaps only weeks behind. Then on 20 January, not accidentally the same day Trump took office, DeepSeek dropped their free R1 chatbot, a new generation ‘reasoning’ module that self-checks for accuracy. In just seven days this became the most downloaded free mobile app in the US, surpassing ChatGPT and sending Nvidia stocks into freefall.

But the AI dinosaurs survived the experience. Elon Musk, taking time out from his frenzied rampage through the US government civil service to ‘delete entire agencies’, still had enough confidence in, or fear of, OpenAI in February to offer to buy it out for nearly $100bn (that’s billion). You don’t front that kind of dosh for an obsolete company. Unfazed, OpenAI boss Sam Altman tartly responded by offering to buy Twitter off Musk for a tenth of the price.

One other group of people who must have been panicking at the DeepSeek R1 release were the Taiwanese, at least those of them not keen on being invaded by the totalitarian regime currently brow-beating them from across the Straits. As the near-monopoly of global chip production in Taipei is the only thing keeping US interests locked in, it is also the only thing keeping China from invading.

Before DeepSeek, serious questions were already being asked about AI, not least the eye-watering expense, and the environmental cost in terms of increasing carbon emissions. A 2024-5 FT TechTonic podcast series explored the possibility that the AI fizz was about to fizzle out because the sector was overhyped and overpriced and there were no obvious game-changing real-world applications for it beyond mundane and low-paid customer service roles, with the result that Silicon Valley venture capitalists were becoming increasingly skittish at the prospect of yet another AI ‘winter’.

Deepseek has impacted a landscape that was already convulsing for other reasons. Trump’s new (dis)order is ripping up the ‘Green new scam’, in turn provoking a tidal wave of carbon-pledge-breaking by governments and fossil-fuel companies who always knew their climate targets were unachievable and are now keen to downgrade or dump them entirely, and get back to the real business of making money. The Net-Zero Banking Alliance has collapsed after exits by Goldman Sachs, Morgan Stanley and four other major US banks. One of the world’s largest climate investor initiatives, Net Zero Asset Managers, has abandoned their net-zero investment plans. And the big AI firms are also ditching their net-zero targets as they find themselves locked in a death-race to develop the app to kill all killer apps, Artificial General Intelligence (tinyurl.com/3z5fbuf7).

Whether AGI is even possible is highly debatable, given there’s no consensus on what intelligence actually is. But AI firms can’t worry about that. What they worry about is being left behind, especially by the Chinese, and hang the carbon consequences. In the words of Hany Farid, digital forensics expert at UC, Berkeley: ‘Somebody asked me the other day if I’m scared of AI, and I said no, but I am scared of capitalism… I think capitalism is going to do what capitalism does, which is burn the place to the ground in order to win’ (FT TechTonic: The Trouble with Deepfakes).

With the speed of development even scaring AI ‘godfathers’ like Geoffrey Hinton, many believe a Douglas Adams’ style thinking machine cannot be far away, with unknowable consequences for humanity’s future. A chorus of demands has arisen for AI regulation, but last month’s summit on AI in Paris ended with US vice-president Vance rejecting ‘excessive regulation’, and even the UK refusing to sign the regulatory declaration.

In the unlikely event AGI ever does become a real thing, one wonders what conclusion it would draw from a global system that threatens to destroy its host planet for the gratification of a tiny minority of super-rich individuals. What course of action might it suggest to reverse this unsustainable, obscene and suicidal trajectory? If it did conclude that humans would be better off abolishing the capitalist system of production and stewarding the Earth’s resources in cooperative common ownership, we can be pretty sure its proprietors would have a panic attack and pull the plug. Unluckily for them, they can’t pull the plug on the working class.

PJS

Next article: Letter – The polarisation of politics ➤