Wimbledon error exposes flaws in AI trust

The Dong-A Ilbo has no "Achilles’ heel." This is not a metaphor. According to the Dong-A Ilbo stylebook, the fibrous tissue at the back of the human ankle must be referred to as the "Achilles tendon." Although the paper has covered many cases of Achilles tendon injuries, this rule had been forgotten until an artificial intelligence (AI) assistant brought it back to mind.

A stylebook is a customized grammar and usage manual for a newsroom. The first thing we did after adopting AI in the newsroom was to train it on our stylebook. We hoped it would serve as a sort of style checker, much like online grammar tools.

Thanks to repeated instruction, the AI quickly learned to mimic Dong-A’s editorial style, significantly speeding up the first round of copy editing. There were even instances where draft articles reviewed by the AI were sent to print under deadline pressure, and yet no critical errors were found during final review. A well-trained AI can be more dependable than a dozen desk editors.

But that does not mean we can blindly rely on AI. While it can recite the rules of a stylebook, it cannot grasp nuance or implicit meaning. As German philosopher Karl Jaspers warned, “machines can never be responsible; responsibility always lies with humans,” and that still rings true today.

Wimbledon, the most prestigious tournament in tennis, failed to heed this warning when it introduced AI line judges. The problem was that human referees were not given the authority to override. Organizers apparently never considered the possibility that AI might have a fatal flaw, an Achilles’ tendon.

As a result, an unprecedented scene unfolded on July 7 during the women’s singles round of 16 between Anastasia Pavlyuchenkova of Russia and Britain’s Sonay Kartal.

With the score tied 4–4 in the first set, Pavlyuchenkova had just earned a game point when Kartal fired a forehand. The ball landed on the grass about 20 centimeters outside the line. But the AI line judge remained silent. In tennis, a call must be made for an “out” ruling to apply. The chair umpire halted the match but had no option other than referring the call to the review center.

As the crowd murmured “out” from the stands, the official decision was “replay,” meaning the point was voided and play would resume. Pavlyuchenkova, who lost her chance to lead 5–4, protested that a game had been stolen from her, but to no avail. Only after she pulled off a come-from-behind victory did the incident settle as a minor mishap.

The All England Club, which hosts Wimbledon, explained that the error occurred because the AI system temporarily lost power due to human error, insisting that the AI’s accuracy was not to blame. Humans took the fall to defend the credibility of AI.

Maybe the real Achilles’ tendon of our AI-driven era is the belief that “AI must never make a mistake.” When I asked the AI if this “Gwanghwamun Sketch” column was well written, it simply said, “You are responsible for what you write.”