The Ministry of Science and Technology Has Taken Action! The AIGC User Manual for Researchers Is Here, and the Academic Community Is Beginning to Guard Against AI Gunmen
The Supervision Department of the Ministry of Science and Technology issued the "Guidelines for Responsible Research Conduct (2023)", which clearly drew red lines for hot issues of concern to the society, such as artificial intelligence and the release of major results.
Falsification of experimental data, falsification of experimental images, improper thesis signatures, plagiarism in textbook writing... On the afternoon of January 16, 11 students from Huazhong Agricultural University filed a 125-page complaint against Professor Huang of the school for academic misconduct. Public opinion quickly fermented, and related content such as "Gambling on future to defend academic innocence" and "Everyone is a victim of academic fraud" sparked heated discussions on major platforms.
Some netizens associated this with the previous fraudulent paper on Alzheimer's disease. This Nature paper, which has been cited more than 2,300 times, misled global Alzheimer's disease research for 16 years.
If you pursue the truth without distraction, you will not be able to achieve anything without mastering the art. Over the past 100 years, the noble belief of academic integrity and benefiting mankind has been a beacon guiding students on the road of scientific research. However, nowadays, it is mixed with more and more temptations of fame and fortune, resulting in academic fraud. Especially with the rapid development of technologies such as AI and large models, emerging technologies have become the "accomplices" of scientific research cheating, which is not only unfair to researchers who insist on studying, but also false research data may cause serious consequences.
Therefore, while strictly investigating academic fraud, it is also very important to regulate the application of technologies such as AI in scientific research.
Draw clear boundaries and use generative artificial intelligence reasonably in accordance with regulations
Academic paper fraud has a long history. In addition to conventional methods such as plagiarism and fabrication of data, it is also common to pay writers, use "paper factories" to write for you, and fabricate papers.
Nowadays, generative artificial intelligence technology represented by ChatGPT (AIGC) has emerged, providing a new helper for people to write papers. From topic selection to manuscript polishing, from statistical analysis to chart production... Its powerful functions cover almost every aspect of the academic paper writing process, and can indeed help scientific researchers quickly complete literature retrieval, data processing, translation polishing and other tasks.
But everything has two sides. If researchers regard the texts generated by AIGC as their own creations, it will largely create a worthless "academic bubble". In addition, over-reliance on AIGC to produce unreliable research results will greatly reduce the credibility of scientific research.
December 21, 2023The Supervision Department of the Ministry of Science and Technology issued the "Guidelines on Responsible Research Conduct (2023)" (hereinafter referred to as the "Guidelines》),Clear red lines have been drawn for hot issues of public concern such as artificial intelligence and the release of major achievements.
The Guidelines point out in the research topic selection and implementation section that the application materials of scientific researchers' research projects should be true, accurate and objective. They should not use the same or similar research content to submit repeated applications, and they should not list others as members of the research team without their consent. They should not plagiarize, buy or sell, or write application materials for others.Generative artificial intelligence may not be used to directly generate application materials.
We should follow relevant laws, regulations and academic norms, and use generative artificial intelligence reasonably to process text, data or academic images in accordance with regulations to prevent risks such as forgery and tampering of data.
In terms of literature citation, the Guidelines also clearly stipulate that content generated by generative AI, especially those involving key content such as facts and opinions, should be clearly marked and its generation process should be explained to ensure authenticity and accuracy and respect for the intellectual property rights of others. Content that other authors have marked as AI-generated content should generally not be cited as original literature, and if it is necessary to cite it, it should be explained.Unverified references generated by generative AI may not be used directly.
The "Guidelines" clarify the boundaries of the reasonable use of generative AI in terms of research implementation, data management, authorship and publication of results, and literature citations. They are of great significance for protecting the credibility and ethical principles of scientific research, and help prevent the irresponsible use of generative AI and related technologies.
Well-known domestic and foreign journals regulate AI-generated papers
Academic paper fraud is a global problem. The emergence of generative AI has lowered the threshold for fraud to a certain extent. Chatbots such as ChatGPT have the ability to "tell lies with a straight face and justify themselves" and provide convenience for these people. However, as the boundaries of AIGC use become clearer, in addition to clarifying laws and regulations, more and more academic institutions at home and abroad have reached a consensus and begun to face up to and regulate the use of AI.
Many domestic journals have issued statements restricting the use of AI by contributors in the process of writing papers. For example, the Journal of Jinan University, Journal of Literature and Data, China Science and Technology Journal Research, Think Tank Theory and Practice, and Library and Information Work have all issued statements saying thatIf the main content of the paper is generated using AI tools, once discovered, it will be treated as academic misconduct.
Foreign academic journals have also standardized the application of AI in paper writing. According to incomplete statistics, Nature, Cell, The Lancet, JAMATop journals such as the Journal of the American Medical Association have issued statements saying,Artificial intelligence does not qualify for authorship, and researchers using AI should indicate this in their manuscripts.
Using AI to defeat AI and find out the "gunmen" of the paper
Promoting the development of AI for good requires effective technical means. It is worth noting that at the current level of technology, AI-generated papers are similar in form to original papers, and traditional text similarity comparison tools may not be able to accurately detect them. Therefore, both at home and abroad, we are exploring the development of detection tools specifically for AIGC.
"The core idea of the AI-generated content identifier is to first build a training dataset that contains real content and AI-generated content, and then train a classifier to distinguish between these two types of content," explained Tang Jian, a well-known Chinese scholar at MILA Research Institute, in an interview with the media.
Specifically, the AI language model works by predicting and generating one word at a time. After generating a word, the watermark algorithm randomly divides the language model's vocabulary into a "green list" and a "red list," and then prompts the model to select words on the "green list." The more words that are listed on the "green list" in an article, the more likely the text is machine-generated. Texts written by humans tend to contain more random combinations of words.
In simple terms,AIGC detection technology is "using AI to defeat AI".Relying on massive amounts of text and data samples, we can identify the differences between humans and AIGC tools in terms of average sentence length, vocabulary diversity, and text length, thereby identifying the "gunmen" of AI papers.
However, there are some technical difficulties behind this. For example, the language model is trained with human-created text. The larger the number of parameters, the closer it is to human creation and the more difficult it is to distinguish. In addition, detection is also limited by the length of the text. The accuracy of detection can only be guaranteed if the text is long enough.
Reasonable use of AI in scientific research, don’t throw the baby out with the bathwater
AI for Science The successful application of AI has, to a certain extent, "rescued" researchers from the processing of text and data, allowing them to focus more on the research itself, which can improve efficiency and speed up the progress of scientific research. In addition, AI is also not inferior in completing the exploration of some in-depth and complex scientific research content.
For example, the prediction model, which has received much attention in the field of medical research in recent years, has left some new researchers scratching their heads. However, AI can gradually establish ideas and methods to help researchers quickly build a qualified model.
For example, in terms of experimental design and optimization, AI can generate detailed experimental plans by simply providing information such as the purpose, methods and materials of the experiment, helping scientific researchers to evaluate and optimize existing plans and reduce experimental costs.
In conclusion,We should not be afraid of generative artificial intelligence, but should regulate and guide it so that it can serve scientific research along the right track.For scientific researchers, it is necessary to make good use of AI technology and leverage its advantages to efficiently and accurately complete scientific research and paper writing. However, all opinions and data must be reviewed by the author to avoid some false and fabricated conclusions of generative AI in order to maintain the dignity of science.
In fact, this is exactly the original intention of AI for Science, which is to use AI to reshape and innovate traditional scientific research paradigms.Today, AI for Science has been upgraded in many fields such as biomedicine, material chemistry, mathematics, physics, etc., and has created a number of important results with practical application value. However, many researchers are still on the sidelines. On the one hand, there may be a lack of low-threshold AI tools in their fields, and on the other hand, they may not have found the integration point between AI and their research fields.
The exploration journey of AI for Science has just begun. Only when scientific research and AI go hand in hand can we create more universal tools and methods.