Blog

AI: Space Race or Arms Race?

Posted on 12/10/25 in AI

In her Regius Lecture at the University of Southampton, Verity Harding, author of AI Needs You, set up two competing views of the geopolitics of AI: is it like the space race (competitive but cooperating around agreed rules) or the arms race (zero sum, winner takes all)?

In what must have been an exhausting week for Southampton’s Web Science Institute (disclosure: I used to be summat in the WSI, although I never really knew what), two major and successful events were held in early October: first, a summit on the Turing test at the Royal Society, and then the Regius Lecture with Harding. I will write a few words about the former in another blog.

Harding’s talk was sensible and measured (as befits a former advisor to that Clegg person who, whatever he fails at, never seems to go away). AI, on her account, is potentially valuable, but not perfect. It won’t turn us all into paperclips, keep us as pets, or solve all the world’s problems. AGI is a silly red herring, and what is Godfather Hinton playing at? National ambitions in the area are entirely appropriate, as it will be a leading and strategic industry. Specialisation and cooperation are far more sensible means to this than trying to reduplicate the full stack within national borders. Populist and nationalist rhetoric, however, is shouting down this judicious analysis.

Harding’s book is curiously unsatisfying. It is not, ultimately, about AI at all. It analyses four different institutional and policy solutions to technological dilemmas of the past – the exploration of space, IVF treatment, the early Internet, and the Internet following the September 11^th attacks. They don’t really have a great deal in common with each other, or with AI, and they don’t seem to deliver the advice that Harding wants to give, which is “to empower us to join the conversation about the future of AI”. In fact, all these dilemmas were resolved, in Harding’s view successfully, by experts, boffins, propellor heads, diplomats, politicos and spooks, with minimal impact from you and me.

The talk itself focused on the first of these, the space race, and argued, convincingly, that the rhetoric driving AI development is more like that of another mid-century roughhouse, the arms race. The space race was certainly full of nationalist rhetoric, and it was a genuine race that could only be won by one country and one system, but behind the scenes, as she argues in the first substantive chapter of her book, there were treaties, cooperation, and aspirations for humanity. The plaque that Neil Armstrong left on the moon says “we came in peace for all mankind”, and Harding might have added that they also left behind memorials to two Soviet cosmonauts, messages of goodwill from 73 leaders from around the world, and a golden olive branch. Would that our current leaders were able to conduct themselves at that level of courtesy, dignity and decorum.

They much prefer Strangelove-era arms race rhetoric. Putin’s declaration that whichever country becomes the leader in AI will rule the world was rightly taken by Harding as the classic statement of the challenge. The ravings of a wannabe Bond villain can indeed dictate policy. Leaders line up to bang their shoes on the podium. We not only have to win, but whoever is first will take all the rewards, and everything must be produced in-house, without trade, advice or expertise from elsewhere.

This is, as Harding emphasised, extremely silly, as AI is patently not a zero sum game. Everyone learns from everyone else: the Americans invented generative transformers, the Chinese hack them for a fraction of the price and computing resource, the Indians and others adapt them for local conditions, the Taiwanese make the chips on which they run, we all supply the data for them, and the Europeans arrogate to themselves the role of conscience. Everyone will make money, apart from the data suppliers and the conscience. Everyone will declare success, except the suppliers. Competitive cooperation would enhance quality and accelerate development; antagonistic partition would slow things down and raise costs.

Harding is right about this. The game is not zero sum; the arms race metaphor is inappropriate. But she herself seems to fall into the same rhetorical trap as her subjects: how will failure to adopt the right metaphor damage AI? If we live in a populist ‘arms race’ world, then we miss a trick, but it’s hardly disaster. The conditions for cooperation will always be there, whenever we are sensible enough to take advantage of them.

In other words, just because AI is spoken of as a zero sum game, it won’t become one. And there are hints of this in the history about which Harding has written. Sabres were indeed rattled in the real and dangerous arms race, but behind the scenes there was a lot of communication, methods of verification, the NPT, START and so on. Simply put, it was not possible to live at the intensity implied by the heightened rhetoric of the arms race; reality took control.

But equally, hands-across-the-water isn’t always sustainable either. Politics is always local, and won’t go away. Of course, as Harding explains at length, the space race did produce stuff like the UN Outer Space Treaty, but that was easy to sign when there was no chance of putting people into space. Nowadays everyone is now blithely ignoring it (or exploiting its gaps, e.g. creating space debris), and the Russians even toying with putting a nuclear weapon into orbit.

And we all know who won (America) and who lost (the space race and the arms race ultimately combined to bankrupt Russia’s sclerotic economy). It is easy enough to put olive branches on the moon, but who remembers them? The space race was kind of zero sum, not only globally, but also within the United States; the key position of Senate Majority Leader Lyndon Johnson of Texas for crucial Sputnik-era funding decisions is the reason that the first word spoken on the moon was ‘Houston’. And in that context, let’s also remember Gil Scott-Heron’s magnificent Whitey on the Moon.

Space race/arms race were not the ends of a scale. They met in the middle, driven by the exigencies of an ideologically-divided bipolar world. In neither case did rhetoric alter reality much.

So where does that leave us, with our post-ideological multipolarity? It’s a mixed message: AI is not an arms race, and never can be, so then why would silly rhetoric be so damaging? It will slow things down and lower quality, sure, but a missed trick won’t bring the ceiling down upon us. While much of the lurid imagery surrounding AI stems from the properties of the technology itself (check out the impressively ludicrous title If Anyone Builds It, Everyone Dies), Harding forbore to point out that the political rhetoric is actually a polluting by-product of the post-globalisation world. For our populist leaders, all strategic industries need to be made secure. The longer and more complex the supply chain, the more beholden we are to foreigners. The collapse of American cooperation means that even Europeans are decoupling and reshoring in the name of Macronian strategic autonomy. Trade is a risk, allies are unreliable, win-wins are out.

This isn’t an AI problem, it’s wider, geopolitical and ideological. As the 5AIMS papers which I have co-authored with Dame Wendy Hall pointed out, technology governance can be managed with at least five strategies (Artificial Intelligence Management Strategies – AIMS) – openness and access, rights, commercial and business, paternalistic security, and gloves-off competition. As our book Four Internets explained, the Internet was built thanks to emphasis on the first. The trend nowadays is toward the last two, in all sorts of areas, from managing climate change and the electrification of the global economy, to the Internet, to AI, to pandemic control, to outer space, to defence. This is not good, but as globalisation recedes, it is perhaps inevitable.

The message for Starmer and leaders of the EU, relatively minor players in AI, is that this is the world they have to prepare for, while also preparing for cooperation both with each other, and with the AI superpowers if and when the opportunity presents itself. They need to understand the 5AIMS, the range of options open to them, the parts of the AI stack they can and can’t reproduce, and navigate deftly between them – especially if they aspire to joining the AI big league.

Indeed, maybe they should think about how much industrial policy is necessary at all. Maybe all it means is that Whitey is back on the moon.