Military intelligence? Nuclear weapons led by AI is a terrible idea. 

Those of us who are old enough remember the day in 1983 when the world nearly ended.  

It was just after midnight on Sept. 26. Earlier that month, the Soviets shot down a Korean airliner that had crossed into their airspace. All 269 passengers died, including a U.S. congressman. International tensions were high. 

Lt. Col. Stanislav Petrov, a member of the Soviet Air Defense Forces, was on duty that night, monitoring intelligence from his country’s satellites. A computer sounded an alarm that the U.S. had launched five intercontinental nuclear missiles against the Soviet Union. They would reach their targets in 20 minutes.

Petrov’s duty was to send a warning up his chain of command, which could have resulted in the Soviet Union retaliating with a nuclear attack against America. But Petrov’s intuition told him that something wasn’t right. He distrusted his computer because he believed that an actual U.S. attack would involve more than five missiles. He judged it to be a false alarm.

Twenty minutes came and went without incident. It turned out a Soviet satellite had not detected a missile launch in the U.S. It had seen sunlight reflecting off high-altitude clouds.  

The story carries a lesson that’s critical today for three reasons.  

First, nuclear weapons technology is considerably more complicated now. It includes autonomous weapons, automated threat assessments and shorter decision times. It can be affected by cyberattacks on command systems, false information and the possibility that leaders will trust machines over human judgment.

Second, as I noted last June, the treaty that 191 countries have signed to rid the world of nuclear weapons is failing, and the only remaining treaty limiting........

© The Hill