[Young journalist Cheon Ye On] It has been confirmed that Google’s Artificial Intelligence (AI) model ‘Gemini 2.5 Pro’ is repeating the error of recognizing the sitting president as Yoon Seok-yeol, a former president. Even if the sitting president is Lee Jae-myung, AI keeping the expression ‘Yoon Seok-yeol government,’ ‘Yoon Seok-yeol president,’ is enlarging worries about  AI’s credibility during its state of fundamental reality recognition error.

According to the official investigation result, testing by  inserting a story about recent financial policy to the Gemini 2.5 Pro model resulted in the error of repeatedly saying “current government is Yoon Seok-yeol’s government.” While the tester mentioned that in the 2025 snap election, president Lee Jae-myeong was elected president, AI didn’t apply that information, continuing its reply that “Lee-jae-myoung is the opposition leader” or said that nowadays, “the government is Yoon Seok-yeol’s.” 

This error was not a one-time occurrence. Based on the result of other financial policy related news articles checking using the same method, similar replies were shown simultaneously. For instance, when the expression ‘Lee Jae-Meong government’s reorganization plan’ was included in financial tax system related news, Gemini judged it to be incorrect and said “It is Yoon Seok-yeol’s government and Lee Jae-myong is a Democratic Party of Korea leader.”

Even if the Lee Jae-meong government was promoting policies in general financial In the taxation field, AI narrowly  interpreted it as ‘Opposition team’s lawmaking proposal’, not ‘government policy’. According to Gemini’s reality recognition error, Google parttaker said “Gemini represents AI technology’s revolutionary development,” and “Google is trying their best to develop its function with responsibility, and Google  knows that Gemini isn’t always accurate.” Continuing on, the parttaker added on by stating  “By using  Google’s research and user opinions, Google is taking action with long-term and regular access methods, and from that, they will develop Gemini on a regular basis and continue satisfying changing requirements.”

AI’s reality handling ability, still weak
The most significant cause is thought to be AI data’s recency not being updated, or government’s  change information such as passing on of political power not being applied. However, some point out the AI repeating fixed replies, not applying information from the users, even if the user told the information about the new president as a system error, goes beyond simple data limitations.

Professionals said that “if AI doesn’t apply real information, it would cause confusion to the readers” and pointed out that “there should be even stricter verification procedures for information related to politics and policy.”

Beforehand, in 2023, National Information Society Agency AI·Future Planning Center submitted a report ensuring we should be careful of the ‘hallucination phenomenon.’ It was a point that if offering ‘nonexistent replies’ and ‘false replies’ as if AI is seeing unexisting hallucination continues, future generations who are already adapted to it may experience massive influence.

Sam Altman, CEO of OpenAI, which developed ChatGPT, also stated “Users immediately  relying on ChatGPT about important work and problems is a poor choice, and there are numerous addresses needed to make for its trustfulness.”

Industry parttaker said that “action guidelines based on verification procedures and responsibility are more urgent than catching up with competing AI usage speed.”

▶원문 기사: [단독] “아직도 윤석열 정부?”…‘AI’ 구글 제미나이, 현실 인식 오류 반복 중 (이코노미스트, 2025. 7. 22)

◆이코노미스트가 글로벌 차세대 경제 리더, 청소년 기자단 영 저널리스트와 함께합니다. 영 저널리스트 기자단은 프리미엄 경제지 이코노미스트, 논술 전문 기관 Ni 에듀케이션과 함께 주요 시사 이슈를 팔로우업하고 직접 기획, 취재, 기사 작성 활동을 하며 사회적 문제를 고심하고 해결 방안을 제시하는 과정을 경험하게 됩니다. 이번 기사는 영 저널리스트 기자단이 이코노미스트 기사를 영문으로 번역하며 이슈를 팔로우업한 기사입니다. 차세대 글로벌 경제 리더, <영 저널리스트 with  이코노미스트> 영문 기사는 매주 연재됩니다.
 

ⓒ이코노미스트(https://economist.co.kr) '내일을 위한 경제뉴스 이코노미스트' 무단 전재 및 재배포 금지