Faculty Guide to Preventing Unauthorized Use of GenAI

AI-resistant Approaches 

Specific Examples 

Assessments that require manipulating the physical world, understanding complex ideas from contextual information, and determining causality are currently challenging for AI models to complete (Vu et al., 2024a).  

Role-plays, simulations, debates, building physical models, conducting interviews, generating concepts from multi-media. 

The use of vivas (interactive oral assessment) is resistant to the use of AI. Students must demonstrate their understanding in real-time and must truly understand course concepts to perform in interactive orals (Newell, 2023, p. 3-4). Students also find interactive orals more engaging as a form of assessment, which lessens the likelihood of academic misconduct. As this is a new assessment format for many students, interactive orals should be carefully scaffolded. Students need opportunities to practice the exam beforehand. 

Vivas: (also known as interactive oral assessment) – students engage in dialogue with evaluators or peers to demonstrate their knowledge. 

Mini-vivas: shorter interviews, typically less formal and structured. 

Interactive oral examinations: Evaluators have a set list of standardized questions.   

Multimodal assessments are much more difficult to outsource to AI because they require the use of multiple modalities at the same time: 

1. Requiring audio means the student must speak 

2. Requiring video means the student must be physically present 

3. Requiring images means the student is presenting slides or diagrams. 

Multimodal assessments, where students are required to integrate multiple different types of content, including “videos, graphics, or images”, and having students engage with that content in their assignment makes it more resistant to AI (Sun & Hoelscher, 2023). 

An oral presentation where students are required to speak while presenting images and slides. 

Requiring students to integrate course materials or discussions into their assignment can make assessments more AI resistant (Mulder et al., 2023). If the assignment requires students to draw upon lecture materials or in-class discussions, it complicates the use of AI (Clay & Lee, 2023), (Hodges & Kirschner 2024). Requiring students to integrate content covered in their courses (UWGB, n.d.) such as journal articles or videos shown in lectures can also make it more difficult to use AI. AI models do not have access to course discussions, lecture materials, and readings covered in class. 

Integrating course materials or discussions into an assignment involves making connections to in-person experiences.  

An example is requiring students to relate specific readings to course events (labs, lectures, class discussion). AI cannot make these sorts of connections authentically. 

ChatGPT tends to “hallucinate” and fabricate quotations and references (CTL at Brandeis University, 2023). ChatGPT-4 does provide references in its outputs, but the DOI addresses link to different articles than the referenced article (Elkhatat, 2023).  

Nonexistent journal articles or broken links are possible signals that the paper was made by AI (Lancaster, 2023). 

Instructors can randomly select references to check for accuracy. Nonexistent or inaccurate references can be a red flag that triggers a deeper investigation. 

Require sources to be drawn from institutionally provided paywalled databases. 

Sheridan library has subscriptions to databases for only employees and students to access. 

AI does not have access to these databases, so it cannot accurately draw upon or reference these sources. 

Require functional links and audit references for accuracy. 

Students must provide their sources through functional hyperlinks or PDFs attached to their documents (LX Team, 2023b). Requiring students to document their process by creating an annotated bibliography, providing screenshots/PDFs in appendices, and requiring that links are functional can make written assignments more AI-resistant. 

Requiring recent sources (outside the AI’s training data) ensures that AI cannot be used. 

Requiring students to include scholarly journals from a paywalled database that the institution provides access to ensures that AI cannot access those journals.  

Annotated references can be integrated as part of the scaffolded assessment process for more traditional essays and projects.  

An annotated bibliography demonstrates a student’s ability to apply proper research methods and analysis of resources collected.  

Each annotation should include a reflection which discusses the relevance of a source to the student’s project. 

Students should provide functional hyperlinks of their sources to ensure that the references they have used are valid and credible. 

Case studies provide a situational context (facts, variables, underlying processes) that students analyze and present possible solutions (Stoez et al., 2022). Students are required to demonstrate higher-order thinking, consider resource constraints, possible alternatives, and the pros and cons of each potential solution (Stoez et al., 2022). The best studies “are non-linear and complex, do not have one clear solution, [and] can pose several ethical dilemmas” (Stoez et al., 2022, p. 111).  

Case studies make use of real-world contexts (Stoez et al., 2022). Designing assessments that include high levels of context and higher levels of thinking can also make them AI-resistant (Sun & Hoelscher, 2023). 

Students should be appropriately scaffolded with the required skills and knowledge to be prepared to conduct case studies in class (Stoez et al., 2022). Using a flipped classroom will also be helpful for students to be prepared for case studies (Stoez et al., 2022, p. 111).  

Since case studies are drawn from real-world situations, these assessments are more authentic and relevant to students’ future careers. Case studies should be custom-made and kept relevant and renewed on a frequent basis (Clare, 2022, p. 159). 

A scaffolded case study (Mulder et al., 2023): 

Stage 1: Case study. Students analyze a complicated case, distill main problems, and use knowledge gained in the course to analyze the case. 

Stage 2: Students integrate feedback from instructors/classmates to suggest concept-grounded and evidence-based solutions. 

Stage 3: Students provide a comprehensive solution to resolve the issues raised in the case study. 

Critical reflection, use of scholarly sources validated through referencing and summarization, integration of course content, integration of feedback, and complex cases with no clear answers will harden this assessment against unauthorized use of AI tools.