The Physical Society of Japan Introduced Software to Detect "AI-Written Papers"

 Recently NHK news on-line reported the following;

“AIで書かれた論文” を判定するソフト 日本物理学会が導入

2026年3月25日午後5時00分

生成AI・人工知能

海外の学術誌で、AIで書かれたとみられる粗悪な論文が投稿されるケースが問題になっています。これを受けて国立情報学研究所の研究チームが科学論文がAIで書かれていないかを判定するソフトを開発し、日本物理学会が25日から導入したことが分かりました。

日本物理学会が導入したのは国立情報学研究所の越前功教授の研究チームが開発した科学論文の執筆者が人間かAIかを判定するソフトです。

越前教授によりますと、海外の学術誌などで、AIで執筆されたとみられる不正確な内容が書かれるなどした粗悪な論文が多数投稿されるケースが確認されていて、問題になっています。

開発されたソフトには、人間やAIが書いた10万本以上の論文の特徴的な文章を学習させていて、英語の科学論文では、AIかどうかを90％以上の精度で判定できるということです。

さらに、人間らしい自然な文章に書き換える「ヒューマナイザー」と呼ばれる機能を使った場合でも、見分けることができるとしています。

日本物理学会では25日からこのソフトを導入し、学会が刊行している英語の学術誌に投稿された論文などを対象に、使用するということです。

日本物理学会では、文章の推こうや校正などでのAIの使用は認めていて、まずはソフトを使って、論文にどの程度AIが使われているか実態を把握することにしています。

日本物理学会の播磨尚朝理事は「判定ソフトの導入で、査読の効率化と、不正行為への抑止力になることを期待している」と話していました。

また、ソフトを開発した越前教授は「生成AIを使った学術論文の執筆は非常に脅威になってきている。査読に活用することで健全な学術の進展につなげてほしい」と話していました。

論文でAIを利用する際のルール

論文でのAIの利用はどこまで認められているのでしょうか。

国立情報学研究所などによりますと、日本では学術論文の執筆にAIを利用する際のルールについては、各大学や研究機関、投稿先の雑誌ごとの判断に委ねられているということです。

このうち日本物理学会では、文章の推こうや校正などでAIを補助的に利用している研究者は多いと考えていて、執筆者の責任のもとに、AIを利用すること自体は制限していないとしています。

ただ、

▽論文に掲載するデータや画像について、意図しないねつ造や盗用につながるおそれのあるAIの利用や

▽論文の大部分を生成AIで執筆するなどの補助の範囲を超えている利用は認めておらず、注意を呼びかけているということです。

またAIを利用した場合には、どのように使ったかを論文に明記することも推奨しています。

日本物理学会の林青司理事は「この数年でAIが関与したと思われる論文数が急増しているというデータも報告されているほか、ペーパーミルと呼ばれる粗悪な論文を不正に大量生産する問題も海外で指摘されています。私たちも、チェックしてきましたが、それだけで今の動向に対応できるかは不安もありますので、今回のような判定ソフトに期待しています」と話していました。

Translation

The Physical Society of Japan Introduced Software to Detect "AI-Written Papers"

March 25, 2026, 5:00 PM

Generative AI/Artificial Intelligence

The submission of poorly written papers seemingly written by AI to overseas academic journals had become a problem. In response, a research team at the National Institute of Informatics developed a software to determine whether scientific papers were written by AI, and the Physical Society of Japan (日本物理学会) brought in this on March 25th.

The software introduced by the Physical Society of Japan was developed by a research team led by Professor Isao Echizen (越前功) of the National Institute of Informatics to determine whether a scientific paper was written by a human or an AI.

According to Professor Echizen, numerous poorly written papers, including those containing inaccurate content seemingly written by AI, had been submitted to overseas academic journals, and that was posing a significant problem.

The developed software was trained on characteristic sentence patterns based on over 100,000 papers written by humans and AI, and could determine whether an English scientific paper was written by AI with over 90% accuracy.

Furthermore, the software could tell a text written using a function called a "humanizer" which rewrote text into more natural-sounding human-like language.

The Physical Society of Japan would introduce this software on the 25th and use it on papers submitted to its English-language academic journals.

The Physical Society of Japan allowed the use of AI in text editing and proofreading, and would initially use the software to understand the extent to which AI was being used in papers.

Naotomo Harimam (播磨尚朝), a director of the Physical Society of Japan stated "We hope that the introduction of this adjudication software will improve the efficiency of peer review and act as a deterrent against academic misconduct."

Professor Echizen, who developed the software also stated "The writing of academic papers using generative AI is becoming a serious threat. We hope that its use in peer review will contribute to the healthy advancement of academic research."

Rules for using AI in academic papers

To what extent the use of AI is permitted in academic papers?

According to the National Institute of Informatics and other sources, in Japan, the rules regarding the use of AI in writing academic papers were left to the discretion of induvial  university, research institution, and journal to which the paper was submitted.

The Physical Society of Japan believed that many researchers used AI to assist in tasks such as text revision and proofreading, and stated that it would not restrict the use of AI itself, as long as the author was responsible for its use.

However,

- regarding data and images to be included in academic papers, the usage of AI in ways that could lead to unintended fabrication or plagiarism, or

- the usage went beyond the scope of assistance, such as writing the majority of a paper using generative AI would not be permitted. They were urging caution in these regards.

They also recommended that if AI had been used, how it was used should be clearly stated in the paper.

Seiji Hayashi (林青司), a director of the Physical Society of Japan said "Data also shows that a sharp increase in the number of papers that appear to involve AI in recent years, and in overseas, the problem of paper milling, where low-quality papers are fraudulently mass-produced, has been pointed out. We have been performing checking, but we are concerned whether that alone would be sufficient to deal with the current trends, so we have high hope for adjudication software like this."

 So, the submission of poorly written papers, seemingly written by AI, has become a problem. In response, a research team at the National Institute of Informatics has developed software to determine whether scientific papers were written by AI. Apparently, AI is affecting our daily life in many respects.

Note:

1. The National Institute of Informatics (国立情報学研究所)（NII）is a Japanese research institute located in Chiyoda, Tokyo, Japan. NII was established in April 2000 for the purpose of advancing the study of informatics. This institute also works on creating systems to facilitate the spread of scientific information to the general public. (Wikipedia)

2. A "paper mill" (ペーパーミル) is an unethical organization (paper factory) that undertakes the fabrication and ghostwriting of research papers, mass-producing and selling them. By forging scientifically unfounded papers and submitting them to academic journals, they help researchers advance their careers and pose a serious problem that undermines the credibility of the scientific community. (Google)