3. Redesigning Assessments and Assignments
The AI-Resistant Approach: Presence Required
Rongjun Sun and Maria Gigante
What Is an AI-Resistant Approach? (By Rongjun Sun)
An AI-resistant approach to teaching is not primarily about catching students using AI or designing impossible-to-cheat assignments. It’s about shifting learning toward experiences that are genuinely difficult for AI to replicate – things that require lived experience, judgment, collaboration, embodiment, and authentic engagement. These approaches are called human-centered learning. The goal isn’t to pretend AI doesn’t exist; it’s to ensure students are still developing the skills we actually care about. Rather than asking, “How do I stop students from using AI?” ask instead: “What learning experiences require students to think, decide, create, collaborate, observe, and reflect in ways AI cannot fully replace?”
The AI-resistant approach stems from the concern that the availability of AI may enable some students to shortcut the learning experience. There are concerns that the use of AI may weaken cognitive abilities and stifle creativity. There has been discussion on “cognitive off-loading” in the academic circle (Gerlich, 2025; Heid, 2026), which refers to using devices to reduce demand on the brain. There are reports showing a negative impact of the frequent use of AI tools on critical thinking abilities, and that younger users rely on them as substitutes, not supplements, for their tasks. A study conducted by MIT suggested that students who used ChatGPT essay writers exhibited “cognitive debt,” a condition in which relying on such programs replaces the cognitive processes needed for independent thinking (Kosmyna et al., 2025).
Correspondingly, the AI-resistant approach refers to strategies that aim to minimize the opportunity for students to adopt AI to complete their work or assignments. The focus is on the design of the assignment to make it as AI-proof as possible. According to AI, this type of assignments tends to “require students to demonstrate their own critical thinking, personal reflection, and engagement with course materials.” Therefore, one needs to think about the purpose or expected outcome of the assignment. This approach seems more appropriate for work, among others, that requires demonstrating the understanding of concepts or theories or generating original products that involve the application of such concepts.
Core Strategies
Shift from “Tell Me” To “Show Me”
This approach moves away from testing and reporting toward interactive, hands-on experiences that require students to put learning into practice – with their own judgment, in their own body.
Replace Single Deliverables with Portfolios
Instead of: Submit one screenplay.
Require a full process portfolio:
- Concept proposal
- Research notes
- Peer feedback
- Revision memo
- Final screenplay
- Reflection
This surfaces how students think – how they applied critical analysis, incorporated feedback, and developed their work overtime.
Design AI-Resistant Assignments
Favor work that is grounded in personal experience, local context, or live process:
- Assignments based on personal experience
- Projects requiring process documentation
- Public accountability (presentations, critique, peer teaching, screenings)
- Situated, specific, local problems
- Collaboration (team projects, client work, community partnerships)
There are at least two general strategies to consider for this approach: encouraging student engagement, and making assignments as specific or personal as possible. The encouragement strategy is about making it clear to students that the purpose of the assignment is not about getting the right or a complete answer (so, this strategy does not apply to all assignments) but engaging in original or critical thinking. Students’ performance will not be assessed by what facts they find but how they think about those facts and how these facts are related to what is covered in class. Being incomplete or even partially biased will not be penalized as long as the product is based on their own thinking. This strategy is likely to encourage students to engage in their own thinking and to reduce incentives in AI outsourcing. Below are three particular approaches for the second strategy – being specific.
The principle of being specific can be applied in different ways. One is to localize the context of the assignment. This is basically to confine students’ knowledge space to particular materials that they can draw upon for the assignment. The focus is on assessing their skill in critical thinking rather than the breath of their knowledge. For example, instead of asking students to write a reflection on a chapter, such as socialization, the instructor may ask students to focus on the three examples discussed in class and ask them to identify how these examples are related to a particular concept covered in this chapter.
Another way is to have the assignment built on materials produced inside the class. For example, in a Research Methods class, students are asked to conduct a survey on campus on their use of social media. In the end, they are asked to do data-entry and to use software to analyze the data collected by themselves.
A third way of being specific is to have personal elements in the assignment.
Or in my Sociology of Aging class, I asked students to write their term paper based on an interview they conducted instead of writing their thoughts on a general topic.
In the end, they are required to draw connections between what they learned from the interview and concepts of theories covered in class. They are not required to cover all the bases. Instead, they can just focus on one or two concepts.
Traditional vs. AI-Resistant Assignments
Looking across disciplines is one of the most useful frameworks for AI-resistant curriculum design. Many fields were grappling with authenticity, practice, and meaningful assessment long before AI arrived. The best examples assess what students can do – not just what they can generate.
Example 1: Film & Media Arts
Traditional assignment: Analyze the visual style of a filmmaker.
This is now easily AI-generated.
AI-Resistant version (same learning outcome):
- Watch a film or scene
- Analyze the cinematic choices and techniques
- Recreate a scene using those techniques
- Document the process
- Present what succeeded and what failed
The learning shifts from reporting information to applying knowledge.
Example 2: History
Traditional: Write a paper on immigration in America.
AI-Resistant version: Students visit a local archive, historical society, cemetery, museum, or community organization and construct a narrative using primary sources.
Deliverables:
- Photographs of sources
- Source analysis notes
- Historical interpretation
- Reflection on gaps and biases in the archive
The learning becomes historical investigation rather than information compilation.
Example 3: Business – Market Analysis
Traditional: Write a market analysis for a hypothetical company.
AI-Resistant version: Students select a real, local small business – a family restaurant, independent retailer, or neighborhood service – and conduct an actual market analysis by interviewing the owner, observing customer behavior, and reviewing publicly available data.
Deliverables:
- Interview notes or recording
- Competitive landscape map of the immediate area
- SWOT analysis grounded in observed, specific details
- Recommendations memo addressed directly to the business owner
- Reflection on what surprised them or challenged their assumptions
The work becomes a real consulting engagement rather than a simulated one. A generic AI response cannot interview a specific owner, observe a specific storefront, or account for the particular dynamics of a specific neighborhood.
Example 4: Sociology
At the end of the semester in Sociology of Aging, students are required to conduct an in-depth interview of an older adult. Specific instructions are given, which require information about the date, time, place, and length of the interview, and the relationship with the interviewee (whose name is required to be concealed). The final report covers the following five sections: 1) background of the person interviewed; 2) lifestyle, which includes daily routines (physical and social) and their social network; 3) changes and adaptive strategies, which include changes in one’s health status and social life, and compensatory strategies adopted; 4) self-perception/assessment, such as their views on aging, including joys, fears and concerns related to aging, and their definition of successful aging; and 5) reflection, including major findings and how they are related to concepts or theories discussed in class, any inconsistencies or surprises, and an evaluation of the interview process.
Engineering / IT: Systems Troubleshooting
Traditional: Describe the steps you would take to diagnose a network failure.
AI-Resistant version: Students are given a deliberately broken system – a misconfigured virtual machine, a network simulation with injected faults, or a physical lab environment with real hardware – and must diagnose and resolve the failure in real time, documenting each step as they go.
Deliverables:
- Live troubleshooting log (timestamped, handwritten or typed during the session)
- Root cause analysis written immediately after
- Short presentation explaining what they tried, what failed, and why the fix worked
- Reflection on what they would do differently
This cannot be effectively outsourced to AI because the student must respond to a specific, unpredictable system state. The mess is the point.
Challenges to Keep in Mind (By Maria Gigante)
- Reading student handwriting
- Time investment for process-based assessment
- Adaptation for large lecture courses
- Adaptation for fully online courses
Applying This to Large Classes and Online Courses
The honest answer is that AI-resistant design works best in smaller, in-person settings – and that tension is worth naming directly rather than papering over. That said, it is far from impossible to apply these principles at scale or online; it simply requires deliberate adaptation.
In large lecture courses, the key is to push AI-resistant work into smaller structural units if possible. Lab sections, recitations, and discussion groups become the venues for process-based and applied work, even when the lecture itself remains traditional. Portfolio requirements, peer review systems, and locally-situated research prompts can all be deployed at scale with thoughtful assignment design and, where necessary, structured peer assessment rubrics that distribute the grading load. Staged deliverables — a proposal due week three, a draft due week seven, a reflection due at the end — also make it much harder to drop in an AI-generated product at the last minute, because each submission creates a paper trail of developing thought.
In online courses, the absence of a shared physical space is a real constraint, but it can be partially offset by anchoring assignments in the student’s own local environment. A business student in Cleveland and one in rural Georgia are not interchangeable — and an assignment that requires them to investigate something specific to their own community, interview someone in their own network, or document a process they personally completed creates inherent variation that AI cannot easily flatten. Synchronous check-ins, recorded process videos, and oral defense components conducted over video can also serve the same accountability function that in-person presentations do.
Where AI-resistant approaches are genuinely difficult to implement — very large online courses with minimal instructor contact, fully asynchronous formats with no live components, or high-enrollment courses with no lab or discussion infrastructure — the honest recommendation may be to combine modest AI-resistant elements with explicit AI literacy instruction: teaching students how to use AI critically and reflectively, rather than assuming the design alone will deter misuse.
For example, an introductory online psychology course of 500 students may have little opportunity for individualized oral exams, in-class writing, or faculty feedback. Instead of relying on take-home essays that students can easily outsource to AI, the instructor might redesign the course so that students use AI as one source among many. Students could be required to:
- submit the prompt(s) they used with an AI tool,
- evaluate the accuracy and limitations of the AI’s response by comparing it with peer-reviewed course readings,
- identify at least two errors, omissions, or biases in the AI-generated content,
- explain how their final submission differs from the AI output and why those revisions improve the work.
In this example, AI is neither prohibited nor ignored. Instead, students are assessed on higher-order skills: critical evaluation, disciplinary judgment, and reflective decision-making – things that AI cannot demonstrate on their behalf. While this approach does not eliminate opportunities for misuse, it shifts the educational emphasis from detecting AI use to cultivating the discernment required to use AI responsibly within the discipline.
Summary
An AI-resistant curriculum doesn’t fight technology — it refocuses teaching on what makes learning irreducibly human. When students investigate real archives, recreate film techniques, document their own creative process, or defend their thinking in public, they are doing something AI cannot do for them. The goal is not to close doors on new tools, but to design learning experiences so rich in context, process, and presence that the most meaningful work can only come from the student themselves. This is the foundation of durable, future-ready education.
