AI Camera Introduced to Replace Cameraman
Camera Follows Sunshim as He Runs
Shiny 'Bald Head' Mistaken for Ball
Office workers are also struggling. This is because of the easy demands at work like, "Hey, let's create something with AI too."
Does having AI guarantee a big hit? Not at all.
In fact, the world of AI has been a continuous series of failures. Companies that have hit the jackpot with AI are only about 0.1% of cases. 99.9% of companies and services have failed.
However, one of the fastest ways to success is studying failures. The same applies to AI. Examining 99.9% of failures is a shortcut to the 0.1% success.
The 'AI Wrong Answer Notebook' series explores failure cases related to AI products, services, companies, and individuals.
The AI camera, replacing the cameraman, followed the referee's head instead of the soccer ball. It 'mistook' the shiny bald head for the soccer ball. ChatGPT
The whistle blew and the soccer match began. It was a Scottish Championship (second division) match between Inverness Caledonian Thistle and Ayr United held in Scotland in October 2020. Due to social distancing measures amid COVID-19, the game was held without spectators.
Another highlight of the match was the ‘artificial intelligence (AI) camera.’ With limited spectator attendance due to the pandemic and decreased gate revenue, the Inverness club introduced it decisively. It was an AI-based automatic tracking camera system, so there was no longer a need for cameramen to frantically follow the movement of the soccer ball. The AI recognized the ball in real time and adjusted the camera’s direction accordingly.
The ‘AI camera’ became a new symbol for the club and was even mentioned as an innovation case in sports management. The club heavily promoted it before the match. It was a clever strategy to comply with social distancing and also reduce labor costs.
Fans watched the game through the screen. The match proceeded fiercely as usual.
However, sometimes the broadcast screen was strange. The camera did not keep the soccer ball in the center of the screen but kept showing irrelevant places. This was especially true during counterattack situations. At the tense moment when a long pass sent the ball into the opponent’s penalty box, the camera did not capture that but showed something else.
When the Bald Linesman Runs, the Camera Follows
Actual broadcast screen of the Scottish Championship (2nd division) match between Inverness Caledonian Thistle and Ayr United. YouTube capture
Viewers soon realized what the camera was trying to show repeatedly. The camera seemed more interested in the movement of the linesman (the referee who enforces rules related to the lines) rather than the soccer ball. When the linesman ran, the camera’s “head” followed exactly. Once viewers understood this, their irritation and anger quickly turned into laughter.
The linesman was bald. The AI camera mistook his shiny bald head for the soccer ball. When the linesman ran, the AI judged that the soccer ball was moving and the camera followed the movement of the head. Both the broadcast team and the club realized a serious problem had occurred. But by then, it was impossible to send cameramen to work. Nor could they stop the broadcast. The commentators had no choice but to repeatedly apologize, saying, “We apologize to our viewers.”
Social networking services (SNS) were also in an uproar. Like soccer fans watching live on TV, people on SNS could not hide their laughter. Humorous criticisms such as “From the next game, bald referees should wear hats” appeared.
Xi Jinping is ‘Mr. Shithole’
The National Security Advisor's Office Facebook page, which reports in Burmese on the meeting between Aung San Suu Kyi, Myanmar's State Counsellor, and President Xi Jinping, automatically translated President Xi's name as "Dunggudeongi" (a Korean slang term for "cesspit"). Screenshot from Myanmar State Counsellor's Office Facebook page.
The AI’s debut on the Scottish soccer stage was not glorious but rather embarrassing. AI’s ‘funny yet sad mistakes’ are not limited to sports. On the international political stage, where every word can affect national interests, AI can also be a source of instability.
In January 2020, Chinese President Xi Jinping visited Myanmar. It was to expand China’s core national project, the Belt and Road Initiative (land and maritime Silk Road), and to strengthen China-Myanmar relations. It was the first visit by a Chinese president to Myanmar in 19 years, making it a major topic in the region.
However, a commotion arose from an unexpected place. Some SNS posts claimed that “Xi Jinping’s name translates to ‘shithole’ in Burmese.” Some users even uploaded screenshots of the translation results. It was absurd but true. Facebook’s automatic translator really translated President Xi’s name as ‘shithole.’
Facebook’s second largest revenue market after the U.S. is none other than China. A large-scale boycott was inevitable, and the Chinese authorities could impose a severe crackdown. As the problem spread, Facebook had to issue an official apology. They said, “There was a problem in the process of translating President Xi’s name into English in our Burmese translation service,” and “This issue does not represent how our products and services operate, and we sincerely apologize for the offense caused.”
Finding the Key to Success in the Graveyard of Failures
The above cases show that the introduction and application of AI systems can proceed differently than expected. They suggest that numerous considerations and tests are necessary in AI system development and application.
The Inverness club’s AI camera incident especially exposed the fundamental problems AI has, such as the limitations of AI image recognition technology and the lack of training data. The AI camera developer collected thousands of photos of soccer balls with various patterns and input them into the AI. The AI learned virtually every soccer ball photo existing on Earth. However, no one anticipated that a bald head could be mistaken for a soccer ball.
Facebook’s case also demonstrated how important the quality and diversity of training data are. Minority languages like Burmese have limited data. It reminded us again that improving AI service quality in data-scarce environments is a challenging task.
Despite some ‘AI bubble’ skepticism, AI is solidifying its status as the ‘AI revolution’ following the agricultural, industrial, and computer revolutions. The AI revolution has just begun. Considering the productivity innovations AI revolution will bring, some believe it has not even taken its first steps yet.
However, failure has always accompanied the process of innovation. Addressing failure does not signal the end of the AI revolution. Rather, failure is a necessary process to succeed with AI. Even the brilliant achievements of today’s AI are filled with numerous trials and errors and setbacks. The key to success can be found precisely in this graveyard of failures.
Next Series Preview
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
![[AI Error Note] ① "That's Not a Soccer Ball, It's the Referee's Head"](https://cphoto.asiae.co.kr/listimglink/1/2024112209534320595_1732236823.jpg)
![Clutching a Stolen Dior Bag, Saying "I Hate Being Poor but Real"... The Grotesque Con of a "Human Knockoff" [Slate]](https://cwcontent.asiae.co.kr/asiaresize/183/2026021902243444107_1771435474.jpg)
