▲ On the e-commerce platform, businesses guide customers to add WeChat to communicate the details of papers. Information map of Beijing News. Thesis writing has always been a hidden disease in the black production of academic cheating. With the advent of the era of artificial intelligence, AI has become a powerful production tool.
Recently, according to the undercover investigation of the Beijing News, in the paper generation industry chain, the black ash production team has begun to mass fabricate papers through AI tools, and receive orders for production and sales on various platforms.
What is worth noting in the report is that some members of the ghostwriting team told the reporter that the emergence of AI ghostwriting tools has lowered the threshold of ghostwriting, and even the phenomenon that "vocational high school students use AI tools to help doctoral students write papers" has emerged.
At AIGC (AI generated content) At the beginning of tools, the industry generally believed that AI would improve the efficiency of content production, but perhaps surprisingly, AI tools have not been widely used in serious academic research and production for the time being, but it has become the first tool to improve the output of the black ash industry.
It can be found from social media search that some "AI Writing Thesis" tutorials and small advertisements have appeared as early as about two years ago, but in the past six months, there have been media reports about "AI Thesis" in universities and academic journals, and even foreign media believe that academic papers are suffering from a large-scale attack from "AI Thesis".
In terms of technical principles, there are two main forms of AI tools cheating on papers. The first is similar to the paper generation, which is mainly used for some comprehensive and analytical papers that do not need research data. AI generates full text based on database and prompt words. This is also the case of falsification of papers mainly involved in the report.
The second is more subtle, mainly involving the use of AI tools to fabricate logically forged data sets, so as to generate logical experimental results and test data according to the forged data. This kind of data fraud has higher requirements for users and is more difficult to investigate and find, and it also has the greatest damage to academic and social public interests.
At present, we still haven't established more effective technical prevention measures for the above two fraud cases. In recent years, many academic journals at home and abroad have been found to have obvious "AI traces" after their papers were published.
In the past, our technical inspection of academic paper fraud was mainly "duplicate checking", which was essentially a search and comparison of texts based on big data.
This kind of technical inspection is actually a post inspection of individual "plagiarism" behavior with technical tools. To make an inappropriate analogy, technical duplication checking is to use hot weapons against cold weapons, and use big data to monitor individual plagiarists. Therefore, it can effectively prevent the proliferation of paper fraud.
However, with the rapid development of artificial intelligence, the traditional duplicate checking technology has gradually become ineffective. At present, it has become a contest between the "big data retrieval system" and the "big model". With the help of the big model tool, black ash producers and paper counterfeiters have grasped the technical advantages.
Therefore, the investigation and punishment of such fraudulent papers has returned to the early stage, and highly depends on the personal judgment of the reviewers. For example, some university teachers have previously concluded in an interview with the media that such articles are "relatively water" and "less innovative".
This individual judgment can play a role because the content generated by the big model is still "data induction" rather than "logical reasoning", because if the reviewer carefully reads and observes the text, it can still accurately identify those articles suspected of AI fraud.
However, in the face of the potential massive falsification of papers and the profit driven gray industry chain, individual judgment cannot become a long-term effective means of prevention and control. In order to fight against the "AI generation" of black ash production, regulators need to upgrade technical tools faster and establish more stringent industrial norms. That is, on the one hand, use "AI to detect AI", and on the other hand, establish clear standards for "how to use AI tools".
At present, domestic and foreign research and development have begun to focus on "AIGC content" Test tools for. On some content platforms, suspected“ AI generated content ”Exceptions are marked.
From the logic of data detection and recognition, the characteristics of this kind of AI forgery content are relatively more obvious. Through the detection of text structure, words and language models, it should be possible to identify and detect AI creation traces.
Therefore, in addition to the traditional plagiarism checking system, at least for now, the detection tools for "AI content" can also be applied to the detection process of academic papers as soon as possible, and the content suspected of AI fraud can be marked to assist the reviewers in judgment.
In addition to technical means of prevention, the academic community can also act quickly to set a clear boundary for the application of AI tools against AI paper fraud. For example, AI can be used to assist in the production of illustrations, but no text or data generation is allowed. Or, the use of AI tools must be carefully explained in the pre research method section of the paper, and the text should also be clearly marked. If there is no marking and is marked by the system, the writer should be punished to a certain extent.
Although some black ash products can make illegal profits by taking advantage of the regulatory time difference in the short term, with the application of AI in monitoring and early warning, and the establishment and adjustment of the industry's common norms, the loopholes of AI fraud will eventually be quickly filled. In the long run, AI will still be a friend of innovation rather than an accomplice of fraud.
Writer/Marvin (media person)
Editor/Chi Daohua
