Fifteen minutes after midnight, Oyo began to insistently demand the attention of Colonel Stanislav Petrov, the officer at the lonely Serpukhov-15 monitoring station, near the small town of Kurilovo. A Minuteman intercontinental ballistic missile had been launched by the United States, with a warhead that would yield an explosion equivalent to over 1 million tons of TNT. Four more warheads, the Oyo tracking satellites reported, were following just behind it.
To Colonel Petrov, the evidence of his eyes didn’t seem right: “I had a funny feeling in my gut,” he later said in an interview. In training, Petrov had been taught the United States would seem to overwhelm Soviet defences with a single massive nuclear attack. Five missiles would not achieve that end.
Finally, he reported to his superiors: This was a false alert. The decision stopped imminent Soviet retaliation on the night of 27 September 1983, a date which might just have marked the end of modern civilisation. Later enquiries demonstrated the Oyo satellites had been misled by an unusual combination of sunlight bouncing off of high-altitude clouds and the satellites’ unusual highly-elliptical Molniya orbits.
This week, as Chinese President Xi Jinping and US President Joe Biden meet, the most consequential agreement they could make is to keep machine intelligence out of decision-making on the use of nuclear weapons. Fourty-five nations have already signed on to restrictions on the use of military AI, Will Knight reports, and China has brought its own proposals to the table, too. The issue is complicated by United States sanctions, which deny China access to some cutting-edge technology.
An agreement to keep AI out of nuclear weapons control systems, though, will be a solid step forward, even if months of discussions could be needed to agree on how a deal will be implemented and verified. China’s nuclear arsenal has grown, both in size and in sophistication, and will likely triple by 2035. Tensions over the Taiwan Straits, or the Indian Ocean, might at any stage flare into conflict—but arms control negotiations are critical to ensuring both sides avoid marching into the apocalypse.
Planners at the Pentagon, Ross Anderson recently reported, are already experimenting with technologies that give nuclear weapons some degree of artificial-intelligence-driven judgement-making options. The common risks this poses to both the United States and China provide compelling reasons for Xi and Biden to negotiate an agreement.
Also read: US military report on China flags its aggression towards India, stockpiling of nuclear weapons
The Able Archer crisis
In early 1981, North Atlantic Treaty Organisation Forces initiated a series of increasingly aggressive manoeuvres along the Soviet Union’s borders, intended to mount psychological pressure on the enemy. The historian of the crisis, Nate Jones, has noted that United States warships sailed into the far northern and eastern regions of the Soviet Union, areas they had never surveilled before. Air gaps in Soviet defences were relentlessly exposed: On one occasion, a United States aircraft flew over the Kuril Islands, and remained there for 20 minutes.
The relentless pressure on their vulnerabilities alarmed the Soviet leadership to a degree never seen before during the Cold War. The United States National Security Agency’s history later reported that “Soviet concern for border security had escalated into a paranoid intensity.”
Little imagination is needed to see just why concerns were so high. The Soviet Union and the United States were locked in proxy warfare in Afghanistan. According to a now-declassified Central Intelligence Agency estimate, the war was bleeding the Soviet military. Tensions were rising in Soviet satellite states like Yugoslavia and Poland.
From 7 November to 11 November 1983, NATO commanders began Operation Able Archer—an exercise intended to prepare some 40,000 American, Dutch, German, British, and Canadian troops to operate in a future chemical and nuclear battlefield. The exercise, documents now available reveal, envisaged a change of leadership in the Kremlin, leading to a proxy war in the Middle-East, and the defection of East Block states to the West.
After a limited war in Finland, the exercise postulated, the Blue Armies—or NATO—received permission to obliterate East Block cities, code-named Orange—with nuclear weapons. When that didn’t deter the Soviet Union, a larger nuclear crisis broke out.
Fearing that this particular exercise might just be a deception for a real attack—no small concern for a generation of leaders who had seen Nazi Germany almost tear apart their nation—the Soviet Union prepared to unleash armageddon. To make things worse, Valerie Vaughan, Vineet Chopra and Jowell Howell note that a sophisticated computer programme assembled by the Soviet Union in the 1970s was showing the metrics it was presented with were consistent with a looming nuclear attack.
Also read: India doesn’t need to one-up enemies’ nuclear weapons. Its doctrine implies size doesn’t matter
The world got lucky
Lieutenant-General Leonard Peroots, declassified documents reveal, was one of a handful of officers who understood just how serious the Able Archer crisis had become Without NATO ever realising, the exercise had driven the two powers to the edge of a precipice. The United States military, Peroots noted, was confident it would have detected signs of real Soviet military response in time to back down, “but this may be whistling through the graveyard.” Even though Able Archer was conducted against what the West saw as a peacetime context, Peroots wrote, the Soviets saw it differently.
Top leadership in the West, like President Ronald Reagan, had been getting contradictory answers to the seriousness of the crisis, Jones observed. The KGB station chief in London, Oleg Gordievsky—a double agent working for British intelligence—was providing alarming evidence of Soviet paranoia.
For their part, though, CIA analysts assured the president there was no such danger. To their mind, the paranoid reactions Gordievsky was highlighting were just intended to mobilise the domestic military and civilian population.
Forty years after the Able Archer crisis, we confront remarkably similar situations. Levels of trust between the great powers are deeply frayed. Artificial intelligence offers the illusion of more rational, accurate judgment, but holds out substantial risks of its own.
Zachery Kallenborn, a leading expert, asks: “How would a nuclear weapons AI even be trained? Nuclear weapons have only been used twice in Hiroshima and Nagasaki, and serious nuclear crises are (thankfully) infrequent.” Lack of examples hinders human judgment too, he accepts, but people “have the capacity for higher-order reasoning.”
Although autonomous nuclear weapons may not exhibit confirmation biases, the lack of real-world test environments means an autonomous nuclear weapon may also experience biases, which are not discovered until warfighting begins.
“Humans can create theories and identify generalities from limited information or information that is analogous, but not equivalent,” Kallenborn notes. “Machines cannot.”
The real lesson from Able Archer, President Reagan concluded in his memoirs, was that even inadvertent war between the United States and the Soviet Union would have resembled “two spiders in a bottle locked in a suicidal fight until both were dead.” That’s the fate Xi and Biden are charged with helping the world avoid.
Praveen Swami is contributing editor at ThePrint. Views are personal.
(Edited by Theres Sudeep)