The advent of artificial intelligence (AI) has brought advanced language models, such as GPT-4 (OpenAI) and Copilot (Microsoft), which can generate coherent and detailed texts on almost any topic in just seconds. While this is appealing, it also raises concerns, questioning established teaching practices and the integrity of assessments. AI detectors can play a role in preventing and monitoring cheating, but their reliability is limited. To maximize the pedagogical benefits of AI while minimizing the risks to the integrity of assessments and the validity of diplomas, it is essential to focus on educating students on the responsible use of AI and adapting evaluation methods.
Benefits that go hand in hand with risks
For teachers, the use of AI offers potential benefits, such as:
- providing teaching assistants for students
- helping students understand complex concepts
- offering students immediate feedback
However, Dalalah & Dalalah (2023) and Sweeney (2023) point out that AI also makes cheating more accessible. Students can easily generate written assignments using language models like ChatGPT, posing a direct threat to academic integrity and the process of competency acquisition. Sweeney (2023) even mentions the budding “normalization” of AI use, where students fail to distinguish between authentic work and cheating, driven by the convenience of AI tools and the pressure to achieve high academic performance.
Weber-Wulff et al. (2023) reveal that institutions are under increasing pressure to detect these practices, as the credibility of diplomas is at stake. AI detectors have emerged as one of the main solutions to identify cheating and protect the validity of diplomas and the reputation of institutions. that institutions are under increasing pressure to detect these practices, as the credibility of diplomas is at stake. AI detectors have emerged as one of the main solutions to identify cheating and protect the validity of diplomas and the reputation of institutions.
Among the growing list of AI detectors on the market are:
Each of these tools offers a range of features, from text analysis to advanced detection of language patterns specific to AI-generated texts.
The appeal of AI detectors in higher education
In theory, AI detectors should play a key role in the fight against AI-generated cheating by not only identifying works written by AI but also acting as a deterrent to cheating.
As mentioned by Sadasivan et al. (2024), while they can’t completely eliminate cheating, AI detectors can reduce its occurrence by raising students’ awareness of academic integrity issues.
The efficacy of AI detectors
Although AI detectors seem promising, research shows that their efficacy remains limited. Sadasivan et al. (2024) studied the performance of existing detection tools, revealing that they often fail to identify AI-generated texts, mainly when they have been slightly modified, paraphrased, or translated.
Perkins et al. (2023) observed that AI detectors achieved only 54% accuracy in detecting texts produced by ChatGPT. This means that a large portion of AI-generated content goes undetected, even by advanced systems like Turnitin. One of the most effective methods to deceive AI detectors involves recursive paraphrasing attacks, where the text is repeatedly transformed, making detection extremely difficult (Sadasivan et al., 2024).
In addition, Weber-Wulff et al. (2023) highlight that the existing detection tools suffer from a lack of precision, making them vulnerable to false positives (human-written texts identified as AI-generated) and false negatives (AI-generated texts identified as human-written). Weber-Wulff et al. (2023) estimate that the accuracy of AI detectors rarely exceeds 80%.
For example, literature review sections, which often adopt a formal and general tone, sometimes produce false positives, as observed by Dalalah & Dalalah (2023). This further complicates the use of AI detectors, as they may stigmatize authentic work and undermine students’ trust in detection systems.
As a result, researchers such as Sweeney (2023) suggest not relying solely on AI detectors to examine academic integrity. They underline that human evaluation remains essential for accurately assessing students’ work and determining the legitimacy of suspicious content.
Prevention through guiding rather than detection
AI detectors are neither foolproof nor sufficient. As demonstrated by Weber-Wulff et al. (2023), Perkins et al. (2023), and Sadasivan et al. (2024), these tools have to be used cautiously as a complement to human evaluation and deeper ethical considerations.
Weber-Wulff et al. (2023) recommend focusing more on teaching strategies that promote the ethical use of AI tools rather than depending on unreliable detection systems. This highlights the relevance of implementing teaching strategies that guide the use of AI instead of attempting to detect it at all costs (Weber-Wulff et al. 2023).
Potential solutions
To address the limitations of AI detectors and the growing challenges posed by AI use and its rapid evolution, a proactive approach is necessary rather than simply reacting. Merely monitoring and penalizing the use of AI is no longer sufficient.
A promising strategy involves educating students and revising teaching practices to integrate AI in an ethical and constructive way into the learning process. This would not only help to provide better guidance for the use of these tools but also foster students’ critical thinking skills, which are essential in an increasingly AI-driven world. For instance, dedicated class sessions could be integrated into existing courses, such as methodology or language courses, to address topics such as:
- key AI concepts
- elements that constitute cheating
- appropriate ways to use AI in assignments
Another option would be to offer hands-on workshops on the responsible use of AI, organized at the library. These workshops, although optional, could be promoted by teachers to encourage student participation.
In addition, transparency is equally essential: institutions could require students to declare any use of AI in their assignments. This declaration could include an explanation of how the tool enhanced their reflection or analysis, which could be integrated into the evaluation criteria.
Another approach is to realign the evaluation towards the process rather than the final outcome. For example, assignments could include reflective journals or individual interviews where students would explain their methodology and justify their choices.
AI could be integrated as a collaborative learning tool encouraging creativity and critical thinking. For instance, students could use a generative AI tool to explore an idea, analyze and critique the AI-generated responses, and then compare them to reliable sources.
Diversifying assessment formats by combining written evaluations with additional in-class components (discussions, oral presentations, group work, etc.) would help minimize the reliance on automated tools. A mixed evaluation approach would create a context where personal effort remains central while AI can serve as a complementary resource.
However, while these strategies seem like viable solutions, it is important to note that there is still limited empirical data to evaluate their long-term effectiveness in preserving academic integrity and supporting students’ development of competencies.
Although AI detectors have a role in preventing and monitoring, their limited reliability and shortcomings highlight the need for additional solutions. A proactive approach that includes training students on the responsible use of AI and adapting evaluation methods can provide a balanced response to the challenges posed by AI. By providing guidance on the use of AI, it is possible to maximize its pedagogical benefits while minimizing the risks to the integrity of assessments and diplomas.