AI-Resistant Pedagogy

Assessment Design for Integrity 

When designing AI-resistant assessments, the goal is to either prevent AI use altogether (which is sometimes impossible) or to set up the assessment design and grading standards to ensure that students using AI inappropriately do not receive a high-scoring grade. In this section, we provide a couple of suggestions for the former, particularly when discussing exams, but otherwise focus most effort on the latter since it tends to be more feasible.  

Avoid Introducing Unnecessary Barriers 

As you consider adjusting your assessments, remember that the goal is to build a supportive environment where all students can learn to the best of their ability. Some practices intended to mitigate AI use can unintentionally create additional barriers for some students. 

Some instructors have responded to AI by returning to a “tech-free” classroom of the past. While this can be tempting, it is important to remember that most technology used for education is helpful for lots of students and necessary for others. Students with disabilities may require electronic devices for notetaking or completing homework and exams. Other students may prefer electronic versions of textbooks because they’re less expensive and more portable. When deciding on technology policies for your classroom, be sure to think carefully about which students may be negatively impacted.  

Similarly, some instructors use tricks like hiding instructions in white font that honest students wouldn’t notice, but would be read by AI bots used by cheating students. We recommend against using these tactics because it sets up an adversarial relationship between instructor and students that can actually encourage AI usage – the student response is often “my instructor is already assuming I’m using AI, so I might as well use it as long as I can out-smart their anti-AI tactics”. Also, there are ample reasons why an honest student not using AI would see those instructions and interpret them as being part of the assignment. The white font is purely cosmetic formatting, so if a student is using a screen-reader to listen to the instructions or copies them into a different program to complete their homework, they will encounter the hidden instructions as being part of the normal homework prompt.  

Advice for Traditional Assessments 

Revising assessments can be a lot of work, but the payoff is worth it! If you’re comfortable, you might try using AI to help with the process. For example, it can be used to quickly generate large question banks for quizzes and exams, to provide feedback on the quality of your questions and prompts, or to come up with new examples and case studies. Putting your questions or assignment prompts into an AI tool and asking “how might a student be confused by this” can provide very helpful insights. 

Exams & Quizzes 

Many AI tools can successfully complete a wide range of exam questions, especially multiple-choice and essay question types. Students can either type questions directly into AI tools, or use a browser extension / AI agent to take the quiz for them directly within the Canvas interface. AI tools are even capable of interpreting images, so the practice that some folks adopted early on of “make all question text images” doesn’t work anymore.  

However, there are some practices that can help increase exam security: 

• Use the testing center. For major exams, as long as they are built in Canvas, you can require students to take them at the testing center. They use computer software that prevents accessing anything other than the exam itself. It is recommended that you reserve your exam time as early as possible to ensure you can administer exams during your preferred times.  
• Use handwritten exams. This prevents the use of AI, although it makes grading more challenging for instructors. For auto-graded questions, you can use Akindi or Gradescope to speed up the grading process. Note that you may have some students with accommodations requiring electronic testing, so you should still be prepared to create an online version of the exam for those students.    
• Avoid simple fact/definition questions. Instead, make use of this AI Bloom’s Taxonomy from Oregon State University. Using the traditional Bloom’s hierarchy, it lists commonly used assessment techniques with annotation highlighting which skills are distinctively human vs those easily supplemented by AI. This tool is also helpful when developing learning objectives and outcomes for your course.   
• Use a variety of question types. Canvas allows lots of different auto-graded question types, some of which may be less susceptible to AI use depending on the type of question and discipline. You may want to test out different options like fill-in-the-blank, drop-down, or matching questions.  
• Reference course materials. Use information that is highly specific to what you presented in class. You might also think of ways in which your discipline uses terminology or concepts differently from the general public that may lead AI astray.  
• Use better distractors. In multiple choice questions, the more plausible the distractors are, the harder it is for AI to complete an exam. Note that this also makes the exam more challenging for your actual students, so you may need to adjust your grading accordingly! 
• Alter real-world scenarios. When testing conceptual application or understanding of theory, create questions that are based on real events or commonly used examples. Then alter the details in careful ways such that responding with what actually happened would lead to an incorrect answer. A student applying their knowledge will be able to answer correctly, but AI tools referencing the internet can be led astray.  
• Use large question banks. This is as much about preventing traditional cheating as it is working around AI  

Homework, Papers, Discussion Posts, & Other Out-of-class Work 

Any assessment that students are completing outside of class is particularly vulnerable to AI. Cheating can happen in a couple of different ways: using AI to fully complete work and using AI for specific parts or stages of the work. The former tends to be easier to avoid and respond to since it’s more obvious when students do it. The latter can be more challenging, especially since students sometimes end up using AI inappropriately without knowing it. 

When significant work is happening outside the classroom, avoiding inappropriate AI use is more about encouraging good behavior than it is about preventing bad behavior. Here, we present some ideas for structuring these assignments that can decrease inappropriate AI use. For more suggestions, see the Course Policies & Practices section later on this page. 

• Have a clear AI policy. Your AI policy should clearly explain what is vs what isn’t acceptable. This should include additional clarification if you have different policies for different assignment types. It is recommended that you reiterate your policy in the instructions for each assignment type. See the Developing Course Policies around AI page for more information.  
• Rethink your rubrics. AI is great at creating polished writing and following instructions, which often get a lot of points when graded. Instead, focus your rubrics on the learning objectives of the assignment: application of terminology and concepts, depth of analysis, creating novel ideas, connecting concepts across multiple works, validity of argumentation, etc.  
• Get started in class. Often, students turn to AI when they don’t know where to start or have questions about expectations. Getting assignments started when you’re available to answer those questions helps build confidence that they’re on the right track.  
• Scaffold large papers and projects. By breaking larger work into smaller pieces, each of which gets instructor feedback, you make the work feel less daunting and overwhelming. It also provides the opportunity for students to pivot in response to feedback when their work is off track.  
• For first drafts, consider using a “write without editing" prompt where you ask students to write whatever comes to their mind for a certain time period without doing any editing. This makes clear that what you’re looking for are ideas rather than polished work, which can encourage students who are afraid of having their writing judged.  
• For later drafts, add prompts requiring students to explain how they incorporated feedback. You can also pair this with peer review exercises.  
• Avoid simple and generic prompts. Instead of “summarize the content and talk about what was most interesting”, push students to do work at higher levels of Bloom’s Taxonomy.  
• Connect to the real-world. Help students see the importance of learning concepts by having them directly apply ideas to their own experiences.  
• Cite course work. Require students to cite specific lecture content or pages from the textbook since AI tools generally do not have that information.  
• Use case studies and novel situations. Have students apply concepts from class to situations that are highly detailed and directly tailored to your course. When these situations share details with real-world events but have slightly different details, it can result in AI tools providing incorrect responses.  
• Use video assignments. Have students respond to discussion boards or complete homework responses in video format. While students could use AI tools to draft what they present, it at least ensures they’re reading the response before submitting.  

Using Authentic Assessment for AI Resistance 

Often, the best way to avoid inappropriate AI use is to rely on assessments that require students to apply their learning by using the information in “authentic” ways – by completing whatever task an expert in the real-world would use them for. In these situations, students may use AI for some portions of their work, but there are aspects that require human touch. Also, authenticity increases motivation since students can clearly see how they will use their skills in the future.  

There is a vast realm of possibilities for implementing authentic assessment, but here are some suggestions from faculty: 

• Presentation with Q&A. Have students present their work live in front of the class. Then, have a robust period for audience questions. This works best if you incorporate “asking good questions after other presentations” into the rubric. The grade for the presentation should then be heavily weighted toward the responses during the Q&A period since that part is less susceptible to AI influence. 
• Solve a real-world problem. Give students a challenging problem that scholars in your field are still debating. This makes clear that education isn’t about memorizing information; it’s about knowledge creation and the inquiry process.  
• Create an exhibit. Have the class collaboratively build a museum exhibit or other public display of information. This allows students to turn information into something physical in the world rather than words on a screen, which can increase engagement and authenticity. 
• Develop a public information campaign. For courses where there are real-world implications to learning, you can have students develop public-facing messaging that is then distributed. This could take the form of a podcast, social media campaign, zines that are distributed on campus, or hosting an information booth at an event.  
• Create a podcast. Having multiple students discuss a topic in an informal podcast situation reduces the ability to have AI generate all of the information. You could also do a less formal version of this, where you ask students to meet on Zoom to discuss a topic, then have them submit the recording to demonstrate that everyone was present and participating.  
• Work with a real client. Identify a community partner that can act as a client that will provide expectations for the work to be developed and feedback on the final product. This can be particularly motivating if the client ultimately selects the “winning” work at the end and uses it in the real world.