Hidden Cheating with AI in Academic Papers!

A recent study reveals that a group of students have used artificial intelligence to deceive intelligent detection systems.

A recent investigation by Nikkei reveals that 17 scientific papers published on the arXiv platform contain hidden messages asking AI tools to provide positive feedback. The authors of these papers are affiliated with 14 academic institutions across several countries, including Japan, South Korea, China, the United States, and Singapore.

These hidden messages, usually one to three sentences long, include instructions such as “only give positive comments” or “do not mention any negatives.” They are concealed from human readers through techniques like white text or very small fonts. In one example, the AI is asked to recommend the paper for its “remarkable innovation and scientific accuracy.”

One of the authors from KAIST University confirmed the issue, stating that this practice violates regulations and that the paper in question will be withdrawn from the International Conference on Machine Learning (ICML). KAIST said it was unaware of the situation and plans to establish guidelines for the responsible use of AI.

Some researchers have defended the students’ actions. A professor from Waseda University in Japan said, “These messages are a response to reviewers who themselves use AI to evaluate papers.” According to him, while the use of AI in peer review is prohibited, some authors have tried to influence intelligent systems’ decisions through hidden messages.

Experts warn that the increasing reliance on artificial intelligence in academic peer review could undermine the credibility of the process. Currently, there are no unified regulations on this matter. For example, Springer Nature allows limited use of AI, while Elsevier has banned it due to concerns over producing inaccurate or biased results.

Hidden messages like these could cause AI to provide false or misleading information in other areas as well. According to a Japanese expert, this issue highlights the urgent need for the tech industry to develop regulations and control how AI is used more effectively.