Networks / Cyber

AI For Five Eyes? New bill pushes AI collaboration with UK, Australia, Canada, New Zealand

“If we want to fight as a system... you have to start sharing technology now,” the former head of the Pentagon’s Joint AI Center told Breaking Defense. “We can’t build the system on the eve of battle.”

231121_five_eyes_cyber_GFX

The flags of the Five Eyes countries, but cyber-style. (Graphic by Breaking Defense, images via Pexels and ODNI)

WASHINGTON — Two key House lawmakers have introduced a bill [PDF] requiring the Pentagon to collaborate more closely with America’s closest allies on artificial intelligence.

Rep. Mike Gallagher (R-Wis.), chairman of the House Armed Services subcommittee on cybersecurity and information technology, and the ranking Democrat on the subcommittee Ro Khanna (D-Calif.) introduced the bill on Tuesday. It would direct the Secretary of Defense and the Director of National Security to jointly form an interagency working group “to develop and coordinate an artificial intelligence initiative among the Five Eyes countries” — Australia, Britain, Canada, New Zealand and the United States.

Formed by the five allies in 1946, the Five Eyes intelligence-sharing apparatus also coordinates on sensitive technical matters ranging from space policy to spectrum management. Meanwhile, in 2021, the three biggest members of the Five Eyes, Australia, the UK and US, signed an even more intimate AUKUS agreement focused on nuclear submarines but with “Track 2” provisions to jointly explore “advanced technology” such as AI.

The proposed bill mandates no further action beyond a report to Congress and provides no new funding. It also explicitly states that “any knowledge or technical data produced by a Five Eyes country under any cooperative project … shall be controlled by that country under the export control laws and regulations of that country.” Export controls — especially the International Traffic in Arms Regulations, which can apply even to software — have been the bane of inter-allied technical cooperation for decades.

Those self-imposed limitations are likely to ease the bill’s progress through a divided and dysfunctional Congress, where it may well end up wrapped into the must-pass policy bill, the annual National Defense Authorization Act.

“The goal was to make this operational and not set it up for immediate failure,” an aide to Gallagher told Breaking Defense, adding that the congressman’s staff had consulted with both the Pentagon and potential co-sponsors in the Senate. “This bill is a stepping stone for more extensive reforms.”

While this initial “stepping stone” would just create a working group, it expects that group to think big on topics like reforming export controls and advancing AI ethics and innovation.

In particular, the working group must “identify (including by experimenting, testing, and evaluating) potential solutions to advance and accelerate the interoperability of artificial intelligence systems used for intelligence sharing, battlespace awareness, and other covered operational uses.” In other words, the objective is not just to talk about sharing the latest technology, but to work towards sharing data across the allies’ increasingly computerized command-and-control systems — what the Pentagon refers to as “Combined Joint All-Domain Command & Control.”

But the Five Eyes already collaborate closely, so what would the Gallagher-Khanna bill add? One expert who spoke to Breaking Defense was deeply skeptical.

“I’m not sure if the bill moves the ball forward as much as sideways,” said James Lewis, a former diplomat and information security expert at CSIS. “Here are already [Five] Eyes working groups on emerging technology, so this seems like Congressional micromanagement. With no new money or authorities? Nice idea, needs work.”

But other experts, including the former head of the Pentagon’s Joint AI Center, were far more enthusiastic.

“On its own, it doesn’t really drive a lot of activity,” acknowledged retired Marine Corps Lt. Gen Mike Groen. “However, I think it does really start to open windows for Five Eyes cooperation in AI.

“That’s exciting,” Groen told Breaking Defense. “There’s a lot of stuff this bill could enable: mechanisms for interoperability, mechanisms for sharing of data … [from] a warfighting perspective, this becomes really critical.”

The working group, if created, “could be a really good opportunity to make these things real, so we don’t fight as five separate allies, [but] as an integrated system of capabilities,” he said.

“If we want to fight as a system… you have to start sharing technology now,” he told Breaking Defense. “We can’t build the system on the eve of battle.”

Bryan Clark, a former Navy submariner and Pentagon strategist at the Hudson Institute, agreed that if the Five Eyes are expected to fight together, they “need to ensure their AI-enabled systems are compatible, if not interoperable with one another.”

If allied countries are running incompatible algorithms or built divergent assumptions into models for such areas as electronic warfare, for example, their AI-assisted operations could end up going in different, mutually unhelpful directions or even directly interfering with each other.

“The bill does a good job setting up the construct for considering how AI will be employed militarily by the Five Eyes countries [and] highlighting some of the challenges,” Clark told Breaking Defense in an email. The bill’s repeated references to “testing and evaluation” are particularly important in this context, because machine learning algorithms are notoriously opaque, even unpredictable — witness ChatGPT’s tendency to “hallucinate” wrong answers for no obvious reason — so the allies need a common, robust approach to ensuring they actually work and that each others’ AIs can be trusted.

Clark said he does have concerns about cooperation and the bill itself, however. It’s important for the Five Eyes team to all pull together — but without yoking the fastest members to the slowest. Rapid US progress “could leave behind Five Eyes countries that are not as far along in implementing AI-enabled systems,” he said. “The bill would help to level the field between allies, but could slow US AI advancement compared to the PRC.”

Keeping up — or catching up — with the People’s Republic of China is crucial, agreed Arthur Herman, who heads Hudson’s quantum technology initiative.

“We are far behind China in thinking and planning strategically about AI, particularly in integrating commercial apps and IP systematically into national security-related uses,” Herman told Breaking Defense in an email. “Having a Five Eyes AI Working Group alone won’t catch us up, but having a working group that is also working to integrate quantum technology into AI planning and research and development — since all five countries have strong and well-developed QIS [Quantum Information Science] sectors — would allow us to steal a march on the Chinese in this regard.” (While the bill doesn’t specifically call out quantum technology, it’s closely intertwined with AI).

Groen agreed that the allies can do better against China together — if they can step up their coordination game.

“When you think about like China, [as] an authoritarian government … they have an integrated system of artificial intelligence for warfighting from end to end, because they own all the pieces” under a policy known as “civil-military fusion,” he said. “We can give the Chinese a run for their money, we have all kinds of innovation and great capabilities — we just deploy them in disparate ways and not through an integrated system.”