How Do I Address Suspected Student Misuse of GenAI Tools?

As generative AI (GenAI) becomes more accessible, instructors face new challenges upholding academic integrity while providing authentic learning experiences. The potential for student misuse of GenAI tools raises valid concerns around assessment validity, skill development, and ethical implications in academic settings. While AI detection tools may seem to be an appealing option, their limitations necessitate a more holistic approach. How can we address suspected misuse of GenAI in a way that maintains trust, ensures fairness, and prioritizes student growth alongside content mastery? Instructors can adopt strategies for prevention, identification, and measured response - aiming to resolve GenAI misuse through guidance and education rather than punitive measures alone. The goal would be to foster an environment of integrity where students develop vital skills for future success.

Limitations of a Detection-Only-Approach

Research has revealed significant limitations in GenAI detection tools (Trust, 2024), highlighting their unreliability in distinguishing between human and AI-generated texts (Chaka, 2024; Sadasivan et al., 2023; Salem et al., 2023) and UMass IT has deactivated Turnitin’s AI detection feature. AI detection tools raise ethical concerns, including higher false positives for non-native speakers of English and students with learning differences (Chaka, 2024; Liang et al., 2023), potentially leading to unfair treatment. Punitive approaches may harm the classroom environment and even impede learning by undermining trust and academic values. Additionally, more proficient students may find ways to bypass AI detection tools (Weber-Wulff et al., 2023; Perkins et al., 2024), and some students may be confused about what qualifies as an acceptable writing aid versus a prohibited GenAI tool (i.e. students may already be using Grammarly or Microsoft Editing AI assistants to proof-read their writing). Given these challenges, a comprehensive and multi-pronged approach is recommended instead of relying on AI detection tools. This approach should include providing guidelines clearly distinguishing authorized from prohibited GenAI tools, explicitly communicating boundaries and expectations to all students throughout the semester and addressing misuse in fair and unbiased ways. 

Prevention – Identification – Clarification – Follow-Up

Prevention

"Cheating" is defined by the rules we set and simply banning all GenAI usage creates enforcement challenges (Furze, 2024). 

Establish Clear GenAI Usage Policies to Promote Accountability. Provide clear guidelines detailing acceptable and unacceptable uses of GenAI for academic work. Explicitly define what constitutes misuse and outline the corresponding consequences. Incorporate these policies into the course syllabus, assignment instructions, and in class discussions preceding major assessments. Consider having students use the AI Assessment Scale (Furze, 2024; Perkins et al., 2024) to identify and understand the permitted level of GenAI use for all course assignments. Require students to add either a non-GenAI use statement or a GenAI Usage Disclosure statement that specifies which GenAI tools they used and how (i.e., Belluomini, 2024). 

Foster an Open Dialogue. The affordances and constraints of GenAI tools continue to emerge. Both you and your students may have questions about their use. Create a classroom environment where students feel comfortable discussing AI and asking clarifying questions related to AI use. This can be achieved through regular discussions, Q&A sessions, and even debates on AI ethics. If students disclose using tools like Grammarly or Microsoft editing tools, have an open dialogue about the appropriate use of such writing aids in academics (Hulsey, n.d.).

Easy-to-Implement Integrity Strategies. Some straightforward interventions to reduce academic dishonesty include a) discussing integrity expectations early on; b) requiring mastery of an academic integrity quiz; c) allowing one-time withdrawal of concerning work submitted; d) providing reminders on policies mid-term and before major assessments; and e) normalizing help-seeking and providing appropriate supports (Vahid et al. 2023).

Establish Authentic Baselines Through In-Class Work. Regularly incorporate in-class writing or problem-solving activities that provide authentic snapshots of each student's individual capabilities and strengths. Maintain a portfolio of this supervised work to serve as your reference point for their typical writing style, depth of knowledge, and skill level. Significant deviations from these personal baselines on take-home assignments may indicate potential GenAI overuse warranting further discussion. 

Design for Motivation and Effort. Intentionally design assignments and assessments that amplify the value of human effort in the process, not just the end product. Consider the value of the assignment, what you might need to provide in terms of scaffolding, resources and support, and how you can provide some choice and agency to foster students’ personal investment (Bowen & Watson, 2024). For example, emphasize how the assignment is relevant and what skills or knowledge students will gain, let students know what the process for accomplishing the task is and what resources they can use, and provide them with a checklist or rubric and examples so that students know what is expected in terms of parts and quality of the work. Using the Transparent Assignment Design Template can help you articulate these aspects to your students.

Refine Evaluation Criteria. Adjust assessment rubrics to prioritize elements AI tools struggle to replicate authentically, such as direct references to class discussions, integrating quotes from course materials, and conveying a strong personal voice. For submissions that fall short of assignment criteria or display poor writing quality, utilize the rubric to assign grades reflective of the work's actual merit based on established standards (Bali, 2023).

Build Students' AI Literacy Skills. Provide students with opportunities to understand, use, apply, analyze, evaluate, and create with AI (Hibbert et al., 2024). Build foundational, ethical, and rhetorical critical AI literacy development into your curriculum (Moxie, 2024).

Identification

Given AI's increasing accessibility, there is no failsafe way to prevent its use on unmonitored assessments. As Lance Eaton (2024), educational developer at College Unbound, cautions, "We often suspect AI usage but cannot know for certain. We may recognize drastic deviations from a student's typical work, but not know definitively whether AI was involved."

Here are some potential indicators warranting further discussion with students who you suspect of misusing GenAI tools:

• Language level deviating from the student's usual capacities
• Inconsistent writing style/voice in the submitted work
• Lack of personalized insights, reflections, or class concept connections
• Overly generic responses lacking depth on the specifics
• Uneven level of detail, with some portions thorough and others vague
• Flawed logic or abrupt transitions between ideas
• Factual errors or improper citations
• Redundant phrasing or rephrasing of the same points
• Mismatched tone inappropriate for the assignment genre
• Unnaturally flawless formatting and structure 
• Use of bullet points when narrative text is expected
• Out-of-place or inexplicable details

Keep in mind that the presence of these indicators does not necessarily confirm AI misuse; but they may signal the need for an open discussion with the student about their process while reinforcing academic integrity standards.

Note: If you work with TAs for grading, train them to flag assignments exhibiting signs of significant GenAI contribution for your review.

Clarification

Ensure a Systematic and Equitable Approach. When addressing suspected AI misuse, , it's critical to uphold a systematic, equitable process. Any measures taken must be applied consistently across all students, without discrimination or bias influencing your approach. Remember, AI detection tools and the linguistic cues mentioned have inherent flaws - they indicate probability of GenAI use, not definitive proof. Avoiding unfair treatment based on suspicions alone and maintaining fairness is paramount. 

Offer Students a One-Time Withdrawal Option. If you suspect that students have misused GenAI for an assignment, review the relevant GenAI usage policies with the class and consider allowing students to withdraw submissions they have reservations about as a one-time option only (Vahid et al., 2023). Include this withdrawal option in your syllabus. 

Review the UMass Academic Honesty Office Resources and Follow the Process. Access the Academic Honesty Guide for Instructors, the Academic Honesty Policy and Procedures Guide and the GenAI and Academic Honesty Guide. Provide students with the Academic Honesty Guide for Students, which lets students know what will happen if an instructor notifies them that they suspect Academic Dishonesty.

Take a Good-Faith Approach and Engage in Open Dialogue. Contact the student(s) and ask them to meet with you, approaching them without accusation but with the intent to understand their process for completing the submitted work. Make the meeting a learning opportunity, pointing out specific elements of concern in their work; ask them to explain what they did and verify their comprehension of the content.

Follow-Up

Grade the Work Objectively Based on Established Criteria. If student work that raises red flags about disallowed GenAI use does not meet the rubric criteria for a good grade, grade accordingly.

Determine Proportional Consequences: In cases where GenAI misuse is confirmed, carefully evaluate the severity and intentionality to determine a proportional response, reaching an informal resolution with the student or submitting a formal charge. Consider the extent of AI usage - was it employed for minor writing assistance like grammar and spelling, or to generate substantive portions of the work? Explore whether it stemmed from confusion around GenAI usage guidelines or how to properly utilize authorized tools. Additionally, factor in whether this is a repeat offense or first-time lapse in judgment. If a student does not respond to the meeting request and you think you have strong evidence indicating misuse of GenAI, please refer to the Academic Honesty Policy and Procedures Guide. 

Continuous Monitoring. Keep track of students previously flagged for GenAI misuse to ensure compliance with guidelines moving forward. Offer additional support or resources to help them improve their skills.

Monitor Your Teaching Methods. As GenAI technology evolves, continuously reflect on and if necessary, adapt your strategies for integrating GenAI into your curriculum and assessments, detecting GenAI misuse and educating students about academic integrity. 

Uphold an Environment of Trust and Growth. Throughout, maintain transparency, respect privacy, and offer supportive resources to uphold an environment of trust and growth.

• Bowen, J. A., & Watson, C. E. (2024). Teaching with AI: A Practical Guide to a New Era of Human Learning. Johns Hopkins University Press
• Chaka, C. (2024). Accuracy pecking order – How 30 AI detectors stack up in detecting generative artificial intelligence content in university English L1 and English L2 student essays. Journal of Applied Learning & Teaching, 7(1), 33. https://doi.org/10.37074/jalt.2024.7.1.33
• Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou, "GPT Detectors Are Biased Against Non-native English Writers," Patterns 4, no. 7 (July 2023): 100779
• Perkins, M., Furze, L., Roe, J., & MacVAugh, J. (2024) The AI Assessment Scale (AIAS): A Framework for Ethical Integration of Generative AI in Educational Assessment  https://doi.org/10.48550/arXiv.2312.07086
• Perkins, M., Roe, J., Vu, B. H., Postma, D., Hickerson, D., McGaughran, J., & Khuat, H. Q. (2024). GenAI detection tools, adversarial techniques and implications for inclusivity in higher education. https://arxiv.org/abs/2403.19148
• Sadasivan, V. S., Kumar, A., Balasubramanian, S., Wang, W., & Feizi, S. (2023). Can AI-generated text be reliably detected? arXiv:2303.11156v3 [cs.CL]. https://doi.org/10.48550/arXiv.2303.11156
• Salem, L., Fiore, S., Kelly S., & Brock, B. (2023). Evaluating the effectiveness of Turnitin’s AI writing indicator model. Retrieved from Copy of Evaluating the Effectiveness of Turnitin’s AI Writing Indicator Model (temple.edu)
• Trust, T. (2023). AI text detectors (aka AI plagiarism detectors & AI content detectors). Open Access Slide Deck
• Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S. et al. Testing of detection tools for AI-generated text. Int J Educ Integr 19, 26 (2023). https://doi.org/10.1007/s40979-023-00146-z
• Vahid, F., Downey, K., Pang, A., & Gordon, C. (2023). Impact of several low-effort cheating-reduction methods in a CS1 class. In Proceedings of the 54th ACM Technical Symposium on Computer Science Education (SIGCSE 2023) (pp. 1-6). ACM. https://doi.org/10.1145/3545945.3569731