Ex-DeepMind Executive Warns That the AI Arms Race Might Lead to Catastrophe

Discussions surrounding an artificial intelligence arms race are ubiquitous—even appearing in this publication. But could this perspective be inherently perilous?

This is the premise put forth by Verity Harding. From 2016 to 2020, Harding briefed global leaders, including Barack Obama and Emmanuel Macron, on AI advancements. As the head of global public policy at Google DeepMind, she focused on navigating ethical dilemmas and foreseeable dangers. In a recent interview with WIRED, she reflected that at that time, AI research was characterized by international cooperation. However, over time, the narrative shifted towards competition—between organizations like Anthropic and OpenAI, and among superpowers like the US and China. The notion of an AI arms race became prevalent.

In a recent anthology edited by Harding, Reframing the AI Arms Race, she collaborates with prominent voices from politics and academia, including historian Lawrence Freedman and Japanese politician Taro Kono, to assert that the language we use to depict AI significantly influences policy and international relations.

Harding argues that framing AI as a dangerous weapon threatens to undermine the international collaboration necessary for ensuring the technology’s safety and equitable distribution of its advantages. For smaller nations that rely on this technology, adopting the arms race narrative inadvertently relegates them to the interests of either superpower, potentially to their detriment.

She observes that the nationalist rhetoric surrounding AI from the Trump administration, alongside attempts to establish export controls on domestic models, exemplifies the detrimental framing of an arms race—and signals the emergence of a dire scenario.

WIRED met with Harding in early June to explore the origins of the arms race narrative, its implications for geopolitics, and how lesser nations might assert influence in the AI development arena.

The following exchange has been edited for brevity and clarity.

WIRED: What draws people to war-related metaphors in discussions about AI?

VERITY HARDING: I think it’s an appealing framing. It feels straightforward, but upon closer examination, it can narrow our thinking.

During my time at DeepMind, the goal was to help policymakers grasp the technology and its potential. It stemmed from the understanding that while the technology holds promise, it also presents challenges that need to be addressed collaboratively on an international scale. Over time, I observed a shift towards viewing it as a larger civilization struggle: the West versus China.

What factors contributed to that change?

One factor was a deeply held belief that the technology posed risks—or could inappropriately be wielded—and thus should be controlled by democracies.

Another factor was an anti-regulation perspective that found it advantageous to cast China as a threat: “If you impose regulations on us, you’re enabling China to succeed.”

Was there a specific moment that acted as a catalyst?

The launch of ChatGPT [in November 2022] significantly heightened interest in AI. However, it coincided with other critical events.

ChatGPT emerged amidst a global pandemic, when concerns arose about a world that was becoming more divided, and during the conflict in Ukraine, where discussions about AI and its geopolitical implications—especially concerning weaponry—became pressing.

It rapidly became commonly accepted that AI represents a new arms race, drawing parallels to the latest arms race of note, the Cold War; discussions likened it to a nuclear weapon.