Summary of Key Points
As the graduation season approaches, "using AI to write papers and reducing the percentage of AI-generated content" has become the new norm for college students dealing with their final assignments. Universities have incorporated AI detection into the paper review process, but existing systems face issues such as misjudgments, inconsistent standards, high costs, and lack of transparency in algorithms. Students are coming up with various creative ways to lower the AI ratio (using AI to generate ideas, selecting less common topics for research, taking advantage of free detection services, or even hiring professionals to do it on their behalf), yet they face financial burdens and privacy risks. Experts believe that AI detection is a last resort due to an imbalance in the teacher-student ratio and suggest improvements such as implementing tiered management systems, traceable processes, and appeal channels. Students, on the other hand, hope for more free detection opportunities to ensure that the technology is more reliable.
The Students' "AI-Paper Battle": A Comprehensive Approach
Many students have adopted a three-step process: using AI tools like DouBao to create a preliminary draft, then checking the AI content ratio with detection platforms, and finally revising the paper with AI to reduce the percentage of AI-generated text. For example, Cheng Chunfei from Zhengzhou University used AI to organize her thoughts and find relevant literature before rewriting the paper in her own words. Xian Junxuan from Renmin University of China chose less common topics like "rural banks" for his research because they had a lower chance of being flagged as AI-generated. Some students also make use of free detection services offered by platforms like PaperPass, repeatedly editing their papers until they meet the requirements. Others have learned from online resources (such as REDnote) to use specific phrases that can make AI-generated text less recognizable.
The Pitfalls of AI Detection Systems: Misjudgments, Confusing Standards, and Lack of Clarity
The problems with AI detection systems are causing significant headaches for students:
- Severe misjudgments: Wang Tian from Huazhong University of Science and Technology found that his own paper had an AI ratio of 36% (exceeding the school's 16% limit), while a friend's paper generated by AI had an almost zero AI ratio. Articles submitted to core journals often have high AI ratios, and even fixed templates like acknowledgments and survey instructions are flagged.
- Inconsistent results: Free detection services often show higher AI ratios than university systems, and the areas marked as problematic can vary greatly between platforms, meaning that revisions may not be effective on the university system.
- Lack of transparency in algorithms: Students do not understand how the systems determine what is AI-generated; experts point out that well-written papers with repetitive conjunctions and vague arguments can also appear AI-generated, leading to misjudgments.
The Universities' Reluctance: AI Detection as a "Lesser Evil"
Why do universities use AI detection? Experts explain that it is a necessary measure due to the imbalance in the teacher-student ratio, as it helps quickly identify potentially AI-generated content and saves teachers time. However, it is only a temporary solution due to technical limitations. They recommend:
- Tiered management: Clearly defining which parts of the paper can be handled by AI (e.g., literature research and idea organization).
- Traceability: Requiring students to provide explanations for using AI and making the detection processes transparent.
- Appeal mechanisms: Providing a way for students to request re-examinations if their papers are misclassified.
The Financial and Privacy Challenges
Reducing the AI ratio comes at a cost for students: reputable platforms (such as CNKI and VIP) charge per word (2-10 yuan per thousand words, up to 40-50 yuan for 20,000 words), leaving many students with limited access to free services. Online "AI reduction services" can be even more expensive and of poor quality; Xian Junxuan's roommate spent over 100 yuan on such a service, only to receive poorly written content that lacked professional coherence and required further revisions. Privacy is also a concern, as some platforms may store students' papers for other purposes (e.g., selling them), while reputable platforms have better data security measures.
Students' Requests: More Free Detection Opportunities and Better Technology
Students acknowledge the usefulness of AI detection but want improvements:
- More free detections: Increasing the number of free attempts from two to four or six to reduce financial burdens and anxiety.
- Consistent standards: Ensuring that results are consistent across platforms to provide a clear direction for revisions.
- Improved technology: Reducing misjudgments and avoiding mistaking human-written content for AI-generated text.
In summary, while AI has made paper writing more convenient, the flaws in detection systems and the associated gray market create difficulties for students. In the future, universities, technology providers, and students need to work together to find a balance between the use of AI and the quality of academic papers.