No, AI is not similar to the Manhattan Project

The writer is an FT contributing editor and writes the Chartbook newsletter

To describe the current AI race, the historical analogy of the moment is the nuclear arms race. Trump’s White House has directly tied its Genesis Mission for AI to the precedent of the Manhattan Project — the effort during the second world war to develop the original atomic bombs. This historical association is intended to sound urgent, dramatic and reassuring all at the same time. And that is also why it is unhelpful. As serious as it may sound to invoke Los Alamos, the analogy in fact presents a sanitised image of the drama we find ourselves in.

For those pushing the urgency and national importance of AI, it is obvious why the Manhattan Project is attractive. A spectacular scientific breakthrough owed to the greatest minds of their generation was put at the service of winning the war against fascism. The ethical dilemmas were real, but in figures such as Oppenheimer they can be portrayed as morally serious.

This is not the world that we are in.

For one, the early atomic bombs were comparatively modest explosive devices, suitable for destroying individual cities not entire civilisations, let alone humanity. The sort of apocalyptic risk fitfully debated in Silicon Valley right now did not appear in the nuclear arms race until the advent of the thermonuclear age in the 1950s and the development of the hydrogen bomb.

If you are looking for a narrative from that period, the urtext is Stanley Kubrick’s satire, Dr Strangelove. Tellingly, in that version of nuclear disaster it is rogue agents — not software but actual humans — who take it upon themselves to launch a first strike and thus provoke the Soviet doomsday machine. The analogy is not edifying. Kubrick’s point was to highlight the obvious insanity at the heart of deterrence doctrine. From the cigar-chomping cold war cowboy to the not-so-ex-Nazi ruminating lasciviously about the sex ratios in the atomic bunker, every manner of ideological fixation was mixed into an annihilatory logic of tit-for-tat.

But dark and funny as Dr Strangelove may be, our situation is even more delirious. We have the old framings of autocracy vs democracy and China vs US. We have dark rumours of the antichrist. But this time, the central actors in our drama are not scientists, soldiers or civil servants but businessmen with huge conflicts of interest. When the bosses of OpenAI and Anthropic took it upon themselves to warn of existential risks and to call for “pacing the frontier”, the immediate suspicion was that they were talking their own book. They were seeking to stir up a panic so as to provoke regulation and thus stabilise their precarious oligopoly. “Rocket-makers in clash with Wall Street over nuke safety worries” is not a headline even Kubrick would have dreamt up.

What separates us both from the Manhattan Project and from Dr Strangelove is political economy. Between 1940 and 1996, a Brookings study estimated that the US spent $5.5tn in 1990s dollars on the nuclear arms race. A substantial slice of that went to private contractors, who no doubt did their best to keep the foreign policy “blob” in Washington DC on high alert and the dollars rolling in. But in its overall architecture the Manhattan Project was the paradigm case of technocratic, government-led industrial policy. It inaugurated the era of publicly funded “big university” research, to which the Trump administration and the likes of Elon Musk are so deeply opposed.

What is singular and novel about the current AI threat is that, for all its profound geopolitical implications, it was born out of the private sector, out of the technologically and commercially driven logic of search engines and advertising algos. It is a product of Silicon Valley after the end of the cold war, when R&D budgets had substantially shifted from the public to the private sector. At first it could be funded in-house from the cash-rich tech balance sheets. But in the hyperscaling race for AGI, existential technological risk has been hooked to Wall Street and financialisation.

The world we are in is not that of ascetic bombmakers wrestling with their souls. It is both darker and more mundane. Today’s world-threatening technologies are fuelled by trillionaire greed, junk bonds and jerry-rigged financial engineering. In filmic terms, think Dr Strangelove mashed up with Gordon “Greed-is-Good” Gekko and a dash of The Big Short. If this is the way the world ends, it will be not with a bang or a whimper, but in a whirl of tech jargon, Californian lifestyles and financial engineering.