A civilian contractor embarked with the 31st Marine Expeditionary Unit, lands a V-BAT drone aboard the amphibious transport dock ship USS Green Bay (LPD-20) in the Philippine Sea, Feb. 2, 2023. (U.S. Marine Corps photo by Sgt. Andrew King)

WASHINGTON –  With both the US and China planning to flood the Pacific with fleets of unmanned mini-subs, surface vessels, and aerial drones, Washington needs to take the lead in creating rules of the road for how uncrewed systems vehicles can avoid collisions — and potential escalations — with systems from Beijing, according to a new report from the DC-based Center for a New American Security.

The findings come from a tabletop wargame run by CNAS and the Hoover Institution. The expert participants’ counterintuitive conclusion: Even if China doesn’t agree to new safety protocols for drones and unmanned vessels — or agrees and then flouts the agreement — it’s still worthwhile for the US and its allies to adopt them unilaterally. That’s the best way to reduce the risk of accidents or, worse, algorithm-driven escalation that no one wants.

The current rules of the road for both air and sea are inadequate, ambiguous, or just plain inapplicable when it comes to unmanned systems, argues CNAS scholar Tom Shugart, who convened the wargame and wrote it up in a forthcoming report shared exclusively with Breaking Defense.

From the UN Convention on the Law of the Sea — which the US abides by even though the Senate never ratified it — to bilateral memoranda between Washington and Beijing, existing agreements either call for precautions an unmanned vessel can’t take by definition, like having lookouts on duty 24-7, or simply omit them altogether, explicitly covering only aircraft and seagoing vessels that have crews.

So Shugart, a retired Navy submarine commander, worked out various options for future “confidence-building measures” (CBMs in arm control-speak), ranging from a draft US statement on new norms to extensive US-China negotiations. Then he worked with Jacquelyn Schneider, director of Hoover’s Wargaming and Crisis Simulation Initiative, to develop the tabletop exercise and convene a group of experts to play it out.

Set six years from now, the wargame examined an all-too-likely crisis in the South China Sea. As it has in the past, the US in the game deployed a carrier strike group to support an ally, here the Philippines, against Chinese incursions in disputed waters and airspace.

The futuristic difference: Both superpowers deployed large forces of unmanned mini-subs, surface vessels, and aerial drones, like those envisioned by the Pentagon’s Replicator initiative. With many systems on both sides run either by remote control (which sharply limits what human operators can perceive even when the link is working) or by AI, and with both naval history and Chinese doctrine emphasizing the decisive advantage of striking first, how do you keep the robots — and the situation — under control?

Even highly trained humans can make deadly mistakes with international consequences. Consider the Chinese navy fighter pilot, Lt. Cdr. Wang Wei, who collided with a US EP-3 spy plane in 2001, killing Wang and forcing the EP-3 to make an emergency landing on Hainan Island, where the crew was interned for weeks. Nor are unmanned systems immune from human bad behavior, with Russian pilots harassing US MQ-9 drones over Syria and forcing one to crash into the Black Sea just last year.

Now imagine what could happen when the humans on both sides are trying to manage a high-stakes situation by remote control, or when both sides are making decision by unpredictable and hallucination-prone AI algorithms. In such delicate and dangerous situations, Shugart argued, the US would benefit from clear, well-publicized rules — even if it was the only country following them.

First of all, “the US likely isn’t giving away much, because that’s how we’d likely program our UxVs [unmanned vehicles] to operate in any case,” Shugart told Breaking Defense in an email. Second, a public declaration by the US and Pacific partners helps set the parameters of the debate on American terms and puts pressure on Beijing to moderate its own behavior, as well as hopefully negotiate down the line.

Finally, but arguably the most important point, is that even if Chinese unmanned systems behave unpredictably or unsafely, at least the US isn’t contributing to the confusion, reducing the risk of accidents. It’s similar to civilian life: Sure, you can drive responsibly as possible and still get hit by a reckless driver, but your overall risk of accidents is still a lot less than if you drive like a maniac yourself.

While Shugart doesn’t say so in his report, this kind of free-world-first approach jibes well with how the Defense and State Departments are pushing international cooperation on responsible military use of AI and automation. Instead of letting Russia and China set the pace, or waiting on the long-stalled UN Group of Government Experts in Geneva to thrash out binding international law, the US is building a coalition of the willing, centered on but extending beyond its longtime allies.

The Pentagon spent years developing its own detailed policies on “responsible AI” — complete with an online RAI toolkit for bureaucrats and tech developers — and then worked with State to distill it into a set of general principles for global use, rolled out 13 months ago in a Political Declaration at an international conference in the Hague. Just last week, over a 150 representatives from over 50 countries met in College Park, just outside DC, to work out how to implement those principles as practical policy.

Figuring out new international norms is hard, Shugart acknowledges in his report, let alone getting rival nations to abide by them. But there are times and places where such norms have held or, even in the breach, at least helped contain the violence. In a fast-moving, increasingly automated world where the risk of accidental clashes is higher than ever, he argues, it’s well worth the effort.

“The process of negotiating confidence building measures for uncrewed systems is likely to be a tough slog, and China may well be of little help on the matter,” Shugart told Breaking Defense. “But given the risks involved, it still may be worth it to try and make some progress, or at least set aside some concrete proposals for a better time.”