This month, the White House disclosed that China has rejected American proposals to limit the use of artificial intelligence for controlling the launch of nuclear weapons.
Beijing’s quick rejection, occurring in talks in Geneva in May between U.S. and Chinese officials, is a setback for Washington’s efforts to find areas of agreement with China. President Biden raised this issue with Xi Jinping in California last November and announced that AI was one of the areas where the two countries would hold further discussions.
Tarun Chhabra, director of technology at the National Security Council, revealed China’s rejection and said the White House will soon issue a memorandum on the use of AI by, among other federal agencies, the Department of Defense.
“Our position has been publicly clear for a very long time: We don’t think that autonomous systems should be getting near any decision to launch a nuclear weapon,” Chhabra declared. “That’s long-stated U.S. policy.”
“We think all countries around the world should sign up to that,” he added. “We think that makes a lot of sense to do.”
“Nobody wants to see AI-controlled nuclear weapons, right?” asked, Joe Wang, a National Security Council staffer at the Arlington-based Special Competitive Studies Project, which specializes in AI and emerging technologies. “Like, even the craziest dictator can probably agree.”
Whether Xi is the craziest dictator or not, it’s no surprise he is spurning any agreement on AI with America.
“The Chinese Communist Party has advocated nuclear arms control only for other nuclear powers,” Richard Fisher, Jr. of the International Assessment and Strategy Center told me this month. “For China, the [party] seeks nuclear arms superiority.”
“China is taking its nuclear arsenal to the next level by, among other things, marrying nuclear-tipped missiles to powerful artificial intelligence capabilities,” Brandon Weichert, author of “Winning Space: How America Remains a Superpower,” wrote to me this week.
“AI-controlled nukes would be far more devastating than your ordinary ones. The Chinese are behaving foolishly, though, because the chance for unintended consequences is high. Frankly, AI could easily lead to a nightmare scenario where China loses control of its nuclear weapons arsenal to a rabid artificial intelligence system.”
Will reality mirror fiction? Everyone remembers “WarGames,” the 1983 movie in which an American military computer, on its own, simulated an all-out Soviet attack and almost launched a U.S. counterstrike. In the “Terminator” franchise, a super AI network known as Skynet detonates nuclear weapons to eliminate humankind.
What is the real risk when machines make decisions?
The world almost ended in the early hours of Sept. 26, 1983, when Lt. Col. Stanislav Petrov was the duty officer at the Serpukhov-15 early-warning center south of Moscow. Successive alarms indicated that America had launched five Minuteman missiles from Montana toward the Soviet Union. More than 30 reliability checks in Serpukhov-15 confirmed the attack. Procedures required Petrov to authorize a retaliatory launch.
The Soviet officer, however, ignored procedure. “A feeling inside me told me something was wrong,” he said in an interview.
Petrov was right: Sensors aboard satellite Kosmos 1382 had misinterpreted sunlight bouncing off the tops of clouds as incoming American missiles.
His instinct — what Petrov later called “a funny feeling in my gut”— led him to the right decision. But an AI-controlled system in this situation would almost certainly have launched a first strike on America, which would have been followed by a retaliatory launch, which in turn would have been followed by additional Soviet attacks.
AI technology, if it had controlled the Soviet arsenal that day, would have resulted in the deaths of tens or hundreds of millions of people.
Russia, with its “Dead Hand,” or “Perimeter” system, has already put nuclear weapons under autonomous control. China’s refusal to discuss a ban on AI in these circumstances suggests that it, too, will employ AI for this purpose.
As Fisher told me, there is a “new nuclear arms race with China, Russia, North Korea and Iran.” The race will involve not only numbers of warheads, as it did in the Cold War, but also the automation of launch decisions.
The Chinese appear to believe that AI makes their nuclear arsenal more fearsome. If they do adopt this view, they will be right. For instance, decapitation strikes by adversaries would no longer make sense, because killing China’s national leaders could lead to an automated nuclear strike.
Moreover, China, which routinely has made threats to launch nuclear weapons first, could use AI to improve those initial attacks.
There is, unfortunately, more than an arms race going on here. China is preparing for war with the U.S.
“This war will involve strategic surprise,” Weichert wrote. “AI will help China determine when best to strike, where to strike and what weapons to strike with, perhaps nukes.”
Gordon G. Chang is the author of “The Coming Collapse of China” and “China Is Going to War.”