The Yomiuri Shimbun reported on the 3rd that the Japanese government has decided to suspend the introduction of artificial intelligence (AI) errors related to child abuse decisions as more than 60% of them have been recorded.
According to reports, Japan’s Children’s and Families Agency postponed the introduction of AI, which was distributed to child counseling centers across the country suffering from chronic shortage of manpower and tried to use it for judgment of abuse and temporary protection. The system was developed by the Japanese government with an investment of about 1 billion yen from 2021. About 5,000 abuse records were learned in the AI, and 91 items, such as the presence or area of the wound and the attitude of the guardian, were entered to quantify the possibility of abuse as 0 to 100. With the prototype almost complete, the verification was conducted in a way that allows the government to determine the risk of 100 past abuse cases with the cooperation of 10 local governments.
“As a result of checking the precision of AI judgment by the executives of each child counseling center, 62 out of 100 cases were judged to be ‘significantly low,'” Yomiuri said. Even if there were child testimonies such as “I was subjected to more than kneading by my mother” and “I grabbed my clothes and hit my head on the floor,” AI marked the possibility of abuse as 2 to 3 points, which was very low.
“It is difficult for AI to accurately judge abuse that has different patterns in each case,” Yomiuri said, adding, “AI learning requires a vast amount of records, but 5,000 cases used for AI learning are also small.” He then pointed out, “The most psychological abuse should be judged by the child’s words and facial expressions, but it is difficult to judge mechanically.”
Previously, the AI system developed independently by local governments made a wrong judgment. In Mie Prefecture, which pioneered AI in 2020, a 4-year-old girl was assaulted and killed by her parents in 2023. Before the incident, AI lowered the need to 39% as it checked whether to isolate the child from his parents and temporarily protect him.
“AI is not a magic wand that can do anything,” an official at the National Institute of Information Studies told Yomiuri. “We need to check the feasibility of AI and design it carefully before development.”
EJ SONG
US ASIA JOURNAL
