Will it take a catastrophe to wake us to Al's dangers?

Incidents relating to artificial intelligence rose again in 2024, but regulation has yet to catch up to the threat posed by this world-changing technology, writes Anthony Cuthbertson

- Anthony Cuthbertson

The closest the world has ever come to armageddon was the result of a computer error. On the night of 26 September 1983, in the depths of the Cold War, 44-year-old Stanislav Petrov was on duty at a Soviet radar station outside of Moscow when he saw five intercontinental ballistic missiles appear on his screen.

The protocol was to alert his higher-ups of the incoming threat, potentially triggering a retaliatory attack against the US that would lead to all-out nuclear war. But Petrov refused to follow the procedure – he didn’t trust the computer. Petrov, who is now remembered as “the man who saved the world”, was right. The recently installed early warning system had mistaken the sun reflecting off clouds for missiles.

The event is the first incident listed in the AI Incident Database, launched in 2018 to index dangerous moments caused by computing and artificial intelligence systems. These incidents are now on the rise, with a record 253 reported in 2024, and even more projected for 2025.

Some incidents are fatal, like the robot that crushed a worker to death at a Volkswagen factory in Germany, or the growing number of self-driving car fatalities. Others are merely problematic, such as Amazon’s sexist recruiting tool that marked down female candidates, or faulty facial recognition technology leading to wrongful arrests.

Yet despite the surging numbers, some fear that the threat posed by AI will not be taken seriously until something truly catastrophic happens. Dr Mario Herger, a technology trend researcher based in Silicon Valley, calls this an “AI Pearl Harbour” – referring to the devastating surprise attack by Japanese aircraft on a US naval base in 1941 that killed 2,400 people and finally drew the United States into the Second World War.