There’s a moment in the 1967 film The Graduate that has become renowned. At a party thrown by his parents to celebrate his graduation, Benjamin (Dustin Hoffman) is approached by Mr McGuire, an elderly bore who wants to say “just one word” to him: “plastics”. “Exactly how do you mean?”, asks the hapless Ben. “There’s a great future in plastics,” says McGuire. “Think about it.”

Listening last week to the spending plans of the techlords who run Microsoft, Alphabet, Amazon and Meta leads one to wonder if something analogous might have happened to them on their graduation nights. Except that in their cases, the magic word would have been “AI”.

How else could one explain why four companies that reported combined capital expenditure of $246bn in 2024 – up from $151bn in 2023 – now propose to spend more than $320bn this year on AI, a technology for which no business model currently exists that could conceivably provide a reasonable rate of return on such an investment?

Why are they doing it? Several reasons come to mind, none of them impressive, and all involving the fantasy of getting to artificial general intelligence (AGI). First off, there’s plain old Fomo – fear of missing out – which leads to a conviction that each can’t risk not being in the game. Then there’s terror that there might be an immense “first-mover advantage” for whoever gets to the mythical goal first, with the winner becoming master of the universe. One AGI to rule them all, and all that.

But the most persuasive explanation for the collective madness is the most mundane. It is that tech companies are so obscenely profitable that if they don’t do something risky and exciting with their Everests of spare cash they might – shock, horror! – have to give most of it back to their shareholders. And they would much prefer to cover the countryside with gigantic aluminium sheds – full of Nvidia chips running white-hot, overwhelming electricity grids, lowering water tables and draining reservoirs – than do that.

Until the arrival of China’s DeepSeek-R1, the consensus in the industry could be expressed in Margaret Thatcher’s favourite acronym: “Tina” (There is no alternative). There was no alternative route to AI, it seemed, other than brute-force computing. The Chinese model rather undermines that. It suggests that there are other, less resource-intensive, paths to usable AI. The current lavish spending plans of the US giants suggest that they haven’t yet got the message. Or perhaps it’s just the latest variation on the problem first articulated by Upton Sinclair: that it’s difficult to get someone to understand something if their share options depend on them not understanding it.

Either way, an interesting fault line has opened up in the field of artificial intelligence. Felix Martin, a perceptive Reuters columnist, sees it as a forthcoming civil war. “On one side,” he writes, “are those who strive for artificial general intelligence (AGI), the point where machines match or surpass human capabilities. Let’s call them AI Cavaliers. Facing them are AI Roundheads who are focused on the more mundane goal of solving specific problems as efficiently as possible. Deciding which side to back in this AI civil war will be a defining decision for investors in the world’s hottest technology.”

It’s a colourful metaphor with more than a grain of truth. My favourite Roundhead is Demis Hassabis, the co-founder of DeepMind, a Nobel laureate and now a big shot in Google (which bought DeepMind for a song in 2014). Although he’s always said that his ambition is to get to AGI, his company’s work in London has had a markedly different trajectory from that of the brute-force-computing crowd in Silicon Valley. It’s been focused on specific types of applications, and on training AI models on very carefully curated datasets.

And it’s had some remarkable successes. It built a machine-learning system, for example, that could read retinal scans at very high speed and accurately pick out ones that needed specialised inspection. DeepMind’s AlphaGo model defeated Lee Sedol, the champion Go player, in 2016. More importantly, though, its AlphaFold model solved a major (and formidably difficult) problem in biochemistry – predicting how proteins fold – and then led to a spin-off company that is using the technology for drug discovery. It’s also built a powerful weather forecasting model which seems to outperform traditional physics-based systems, and tons of other interesting stuff. So if there is going to be a civil war in AI, the Roundheads are the side to be on. And there is a great future in AI, at least in London.

What I’ve been reading

The end of programming as we know it?
Very perceptive essay about the future of coding by Tim O’Reilly, one of the sharpest observers of the tech industry, on his website.

Misunderstanding the social media crisis
An insightful piece by the redoubtable Henry Farrell on his Substack, Programmable Mutter. It’s not about radicalised individuals, but about demented publics.

When cats came to my prison
A charming blogpost by an inmate in California for the Prison Journalism Project.