Summary of Key Points
Denmark has introduced measures such as oral defenses to combat AI cheating in high schools, but the issue goes beyond simply preventing cheating. Generative AI has challenged the fundamental logic of traditional educational assessment: “The work submitted by students equals their true abilities.” Nowadays, AI makes it easy to produce high-quality assignments, and its use often falls into a gray area between “assisted learning” and “direct plagiarism,” rendering the traditional method of judging ability based on final outcomes ineffective. Countries are shifting from focusing on detecting AI to verifying students’ actual capabilities by implementing controlled tasks, oral defenses, and recording the learning process. However, these reforms face challenges related to cost, fairness, and privacy protection.
1. AI Is More Than Just a Cheating Tool; It Undermines Traditional Assessment
Traditional education evaluates students based on their final products (papers, assignments), assuming that writing a good paper requires time for research and organization, making cheating (plagiarism or ghostwriting) less likely to occur. However, AI has changed this:
- Dramatic Cost Reduction: Students can use AI to generate topics, outlines, arguments, and even adjust the style of their work in minutes, completing what used to take days.
- Expanding Gray Area: Not all AI use is cheating; some students use it to clarify concepts or correct grammar, but the line between legitimate use and fraud is difficult to draw.
- Data Tells the Story: A UK survey shows that 95% of undergraduates use AI to assist with assignments, and in China, over 85% of students use it for learning, with some directly copying content. When teachers receive a good paper, their first question is no longer “How capable is the student?” but “How much of this was written by AI? Does the student really understand it?”
2. Trying to Catch AI with Detection Tools Is Unfeasible
Initially, schools tried using AI detection tools to combat cheating, similar to using plagiarism checkers, but this is a losing battle:
- AI Generates Unique Content: Plagiarism detectors look for similarities to existing text, while AI creates entirely new text that is hard to identify.
- Improved After Modifications: Even minor edits can significantly reduce detection accuracy from 39% to 22%.
- Risk of Misjudgment: These tools tend to catch students with weaker digital literacy skills, allowing those who know how to use AI to evade detection, raising concerns about fairness.
Denmark’s approach is more clever: instead of focusing on identifying which parts were written by AI, teachers ask students to explain their papers in person. While AI can produce results, it cannot answer questions like “Why did you choose this source?” or “Would the conclusion hold under different conditions?”
3. The New Trend: Verifying Abilities
Countries are rethinking their assessment systems to ensure that students can demonstrate their skills through various means:
- Maintaining a Baseline Without AI: For example, Denmark requires more assignments to be completed on campus, allowing teachers to monitor the learning process and ensure students can complete basic tasks independently.
- Adding Ability Verification: Students must defend their papers, with teachers asking random questions (e.g., “Is the source of this data reliable?”).
- Process-Oriented Assessment: Recording the entire learning journey—drafts, revision history, and interactions with AI—makes the final product less of a isolated document.
- Balanced Use of AI: AI is not banned, but its use must be disclosed (e.g., the International Baccalaureate requires citing AI sources). The focus shifts to whether students can use AI to solve problems effectively.
4. Domestic Schools in China: Moving Beyond Prohibitions
Chinese schools are also taking action, but many still rely on banning AI or using detection tools. True reform should revolve around the purpose of assessment:
- Upgrading Rules: Universities like Tsinghua require students to report their use of AI, with decisions based on the task’s requirements (e.g., AI can be used for research papers but not for basic writing assignments).
- Practical Learning: In Tsinghua’s journalism class, students conduct interviews and write observation reports; while AI can help organize materials, it cannot replace hands-on experiences like talking to people in real-life settings.
- Differentiated Rules: Different subjects and grades have different requirements (e.g., elementary school students cannot use AI for writing practice, while college students can use it for projects, but they must explain how).
5. Reform Is Not Easy: Challenges Include Cost, Fairness, and Privacy
Reforms are challenging to implement:
- High Human Costs: Grading 100 papers and organizing 100 defenses takes significantly more time and resources.
- Fairness Issues: Introverted students or those with language barriers may be at a disadvantage in defenses. Schools with limited resources (large classes, few teachers) struggle to conduct effective assessments, exacerbating educational inequality.
- Privacy Concerns: Screen monitoring and process recording raise privacy concerns—how long can data be stored? Who can access it? What if there are misjudgments?
Education technology companies need to evolve from providing AI generation rate checks to tools that help teachers collect evidence, such as tracking assignment revisions and generating questions for defenses.
Conclusion
Denmark’s approach of oral defenses is not the ultimate solution, but it highlights a shift in educational assessment: in the AI era, we should no longer focus on whether results were generated by AI. Instead, we need to ensure that students have truly learned something. AI can assist with writing, but the ability to ask questions, verify facts, and take responsibility must come from the students themselves. The essence of education is to verify that learning has actually taken place.