전 구글 딥마인드 임원의 경고: AI 군비 경쟁은 재앙을 부를 수 있다

Comment Loader Save Story Save this story Comment Loader Save Story Save this story Reports of an artificial intelligence arms race are everywhere— even in this very publication . But what if that framing is fundamentally dangerous? That’s Verity Harding’s conceit. Between 2016 and 2020, Harding spent her days briefing politicians across the globe, from Barack Obama to Emmanuel Macron, on advances in AI. As the head of global public policy at Google DeepMind , Harding was responsible for mapping out ethical conundrums and potential risks. Back then, she told WIRED in a recent interview, AI research “was rooted in international cooperation.” But somewhere along the way, the industry began to be shaped instead by rivalries—between individual labs like Anthropic and OpenAI and between two global superpowers: the US and China. The AI arms race became the metaphor du jour. In a new essay anthology curated by Harding, Reframing the AI Arms Race , she and other figures from across global politics and academia, including historian Lawrence Freedman and Japanese politician Taro Kono, argue that the language used to describe AI sets the tone for policymaking and the terms of engagement between nations. Harding believes that casting AI as a lethal weapon risks closing the door to the kind of international cooperation required to ensure that the technology is safe and its benefits are evenly distributed. For smaller powers that import the technology, meanwhile, conceding to the arms race framing means lining up behind one superpower or another, potentially against their own interests. Harding sees the Trump administration’s nationalist AI rhetoric and its bid to impose export controls on homegrown models as symptoms of the arms race framing—and evidence that a worst-case scenario is taking shape. WIRED met with Harding in early June to discuss where the arms race idea originated, how the narrative is shaping geopolitics, and what smaller countries might do to guarantee they have a say in AI development. The following conversation has been edited for length and clarity. WIRED: Why do you think people are drawn to metaphors of war with respect to AI? VERITY HARDING: I just think it’s a sexy framing. It’s one of those things that feels very clarifying, but if you dig deeper, it restricts your thinking. When I was at DeepMind, the job was to try to help political leaders to understand the technology and what it would be capable of. It was rooted in the idea that the technology was really exciting, but there were also things to be concerned about that would be more appropriately dealt with in a collaborative, international way. What I started to notice [over time] was this notion that it was more of a civilizational battle: the West versus China. What were the forces behind that shift? One was a sincerely-held belief that the technology was dangerous—or would be in the wrong hands—and therefore that democracies should hold the keys. The other was an anti-regulation stream, [for whom] it was beneficial to point to China as a bogeyman: “If you regulate us, you let China win.” Would you point to any particular moment as a trigger? ChatGPT [ launched in November 2022 ] suddenly made a lot of people pay attention to AI. But other things happened at the same time. ChatGPT emerged at the same time as a global pandemic, when people were freaking out about the borderless world becoming bordered again, and the war in Ukraine, when a lot of the discussion about AI and geopolitics—but particularly weaponry—suddenly became very real. It very quickly became accepted wisdom that AI is the new arms race. It was mapped onto the last arms race in living memory, the Cold War; people talked about it as akin to a nuclear weapon. To what extent does the trajectory of any technology boil down to who happens to be in charge of global superpowers at pivotal moments in its development? A huge amount. People traditionally think about tech coming along and changing society, but the reverse is also true. You can really see that happening with AI right now. The political culture in the US is having a huge impact on how AI policy turns out and thus how AI turns out. The geopolitical environment is very tense and febrile at the moment. A lot of that is down to the US taking a more isolationist approach. Tell me why you advocate for an internationalist approach to AI development and against the idea of individual countries turning inward to develop the technology themselves. I think sovereign capacity in Europe and the UK is really, really important. [But isolationism] has just become the only driving force in AI policymaking and obscures other possibilities and other realities. Even the US and China can’t develop everything themselves, which means you’ve got all of these strategic chokepoints: “You can’t have our chips.” “Well, you can’t have our critical minerals, or our scientists—and we won’t buy your product.” It’s unrealistic to suggest that every country can have their own completely sovereign AI stack. It’s easy to see why European powers might worry about their dependence on American technology, though. In the last two weeks, Donald Trump issued an executive order on AI dripping with nationalist rhetoric, and his administration effectively forced Anthropic to withdraw its latest frontier model from the market. Yes, definitely. But the internationalist approach isn’t in opposition to the sovereignty argument. The argument could boil down to: Competition is normal and healthy but doesn’t have to be mutually exclusive with collaboration and cooperation. I’ve been calling for a middle powers coalition. What would that achieve? It’s partly about leverage. It’s partly about scale. Say it was Canada, France, Japan, South Korea, India, and the UK. India has incredible scale and diffusion of technology; the UK has incredible talent and an advanced startup ecosystem; Canada has critical minerals, [and so on]. The point is to not allow the arms race framing to convince you that the entire game of AI is a binary race between two superpowers. By believing it to be true, you make it true, turning yourself into a small chess piece on one side or the other. You are always lesser. Is money the corrupting influence? Does the popularization of the arms race concept have to do with the emergence of a route to commercializing this technology? The amount, the speed, and the way money rushed in—the freneticism—definitely had an impact. But in and of itself isn’t enough to account for what we’re seeing. Do you think the major labs are complicit in shifting the rhetoric from collaboration toward competition? They are. Talking about AI as an arms race accrues power to them, by saying it’s so powerful, new, and unique, that only we have the answer, and only we should be in charge of the solution. Does the ferocity of the competition between individual AI labs and between global powers to master this technology necessarily undermine the objective of safe and responsible AI development—or can you have both? It’s possible to have both. But is it likely? I don’t think the language of a “race” helps anybody to calmly and collegiately plan. So, if we continue down this road, what is our destination? The natural long-term conclusion is excessive government control and centralization of power over these systems—and less safe and beneficial systems, because we couldn’t find ways to collaborate on things like security, food security, or ending disease. Then, a lot of vassal states that just have to pick one superpower or the other. What’s even a small thing for people to work on together, if not totally agree on? The arms race narrative, the competition between labs, the nationalism we’ve seen grow in AI—all of that means that people don’t want to exercise that [cooperation] muscle anymore. You have to keep that muscle working, otherwise it will wither.