https://arxiv.org/pdf/2506.08872
What keeps coming up is the idea that we should have different ways and modes of assessing. For example, an oral explanation of the submission. I kept thinking though that this would require so much time and effort. But then, I thought of the async periods for my BL course. I can use lab time that no one really shows up to anyway for short 10 minute check-ins where students are scheduled to meet with me and explain their code. This way it would not increase the time spent for either of us.
Quoting someone else’s comment on AI in education and study:
Using AI in education is like using a forklift in the gym. The weights do not actually need to be moved from place to place. That is not the work. The work is what happens within you.
To check out:
Artificial Intelligence (AI) is reshaping education by influencing learning, assessment, instructional design, and information literacy. AI tools such as ChatGPT introduce new opportunities and challenges in pedagogy, requiring educators to rethink traditional teaching methods and student engagement strategies.
AI Literacy & Source Evaluation
AI-generated text is probabilistic, not factual, meaning it can generate plausible but false information (i.e. hallucinations) (Coyle, 2023). Teaching AI literacy involves integrating critical thinking, media literacy, and digital competency into coursework. Students and educators must develop AI literacy skills, including:
- Prompt engineering: Asking AI the right questions to refine responses.
- Verification methods: Cross-checking AI-generated claims with credible sources.
- Ethical considerations: Bias in AI models, environmental costs, and labor exploitation in AI dataset curation.
Prompt:
I’d like to know if there is pedagogical research that has explored or developed some kind of framework or structure for teachers to redesign assessments in the age of GenAI. Specifically, I want to focus on post-secondary education. I prefer peer-reviewed sources, but other sources such as educational institutions, EdTech organizations, or professional learning communities are good too.
Assessment Design in the Age of Generative AI
The rise of generative AI tools like ChatGPT and Claude is forcing educators in colleges and universities to rethink how they assess student learning. Traditional take-home essays or problem sets can now be completed (or significantly assisted) by AI, which raises concerns for authentic learning and academic integrity. At the same time, these AI tools can enhance learning if used ethically–for example, by handling rote tasks so that students can focus on higher-order thinking. The challenge for instructors is to redesign assessments in ways that preserve academic integrity and critical thinking while deciding when to resist AI use and when to integrate it as a learning aid.
Frameworks for Deciding: Resist or Embrace AI?
Educators and researchers have started developing frameworks to guide the use of AI in assessments. Rather than a binary “ban it or allow it,” these frameworks encourage nuanced decisions aligned with learning goals:
Figure: The AI Assessment Scale (AIAS) defines six levels of AI use in assignments, from “No AI” (completely prohibited) to “Full AI” integration as a co-pilot. Each level clarifies how much AI assistance is allowed and the expectations for student work (adapted from Furze et al., 2024 ).
-
AI Assessment Scale (AIAS): Perkins et al. (2024) introduced the AIAS as a “practical, simple…tool to allow for the integration of GenAI tools into educational assessment”. This scale “empowers educators to select the appropriate level of GenAI usage in assessments based on the learning outcomes they seek to address”. It outlines levels ranging from No AI (AI is completely banned for that task) through intermediate stages (e.g. AI for idea generation or AI-assisted editing) up to Full AI integration where AI is used as a collaborator. The aim is to embrace AI’s opportunities when it can enhance learning, while recognizing that some assignments are best done without AI assistance for pedagogical reasons. Crucially, the scale provides clarity for students by explicitly communicating how AI can (or cannot) be used at each level of an assignment.
-
PAIGE Framework: Shanto et al. (2023) propose PAIGE (Promoting Assignment Integrity using Generative AI in Education) as a conceptual model for ethically embracing AI. The PAIGE framework “emphasizes the ethical integration of GAI, promotes active student interaction, and cultivates opportunities for peer learning experiences”. In practice, this means designing tasks where students might use AI in a guided way (rather than hiding it), combined with collaborative learning and discussion to ensure academic integrity. By engaging students in using AI responsibly and in teams, the hope is to leverage AI’s benefits while preserving the integrity of the assignment. In short, PAIGE treats AI as an opportunity for innovation in assignments–provided that usage is transparent, ethical, and coupled with human interaction and oversight.
-
“AI In or Out” Decision Guides: Many teaching and learning centers have published practical decision charts for instructors. For example, the University of Pennsylvania’s pedagogy team suggests considering whether to design AI “out” of an assignment or “in”. In their guide, they provide “examples for designing AI out of or into your assignments… to limit or encourage certain uses of AI to help students learn”. The key principle is alignment with learning objectives: “Assignments support the learning goals of your course, and decisions about how students may or may not use AI should be based on these goals.”. In other words, if an assignment’s purpose is to assess a student’s ability to perform a skill unaided (say, writing original analysis or solving a problem from scratch), then AI might be “out.” But if the goal is to teach how to critically use new tools or focus on higher-order skills, then carefully integrating AI (“AI in”) can make sense.
-
Educational Policy Guidelines: Educational authorities are also offering structured approaches. For instance, Australian guidance for schools (which is also relevant to higher ed) suggests choosing the approach based on assessment type. “Diagnostic or high-stakes summative assessments are likely suited to approaches that prohibit or limit the use of generative AI tools… where AI use may prevent the accurate measurement of student understanding.” In contrast, “assessments such as research projects and portfolios are likely suited to approaches that allow some use of generative AI… In some cases, the use of generative AI tools may enhance student assessments by making them more efficient or focused on measuring the intended skills or knowledge.”. In other words, when factual recall or basic writing is not the main point, AI might be leveraged–but when fundamental understanding or fairness is at stake, AI should be restricted. This kind of guidance helps instructors decide on a case-by-case basis whether to resist AI (e.g. require in-class, closed-book work) or integrate it (e.g. allow AI for drafting or brainstorming) for a given assignment.
Preserving Authentic Learning and Higher-Order Thinking
One clear strategy for addressing GenAI is to double down on authentic, higher-order assessments that AI tools struggle with. Research and academic development resources emphasize designing tasks that require analysis, creativity, and personal engagement–qualities that “foster deeper learning and [make] it harder for AI-generated responses to pass undetected.” In practice, this means moving away from assessments that only measure rote knowledge or formulaic essays (which AI can handle) and toward those that demand original critical thinking. A 2025 review in Contemporary Educational Technology recommends “developing assessments that emphasize complex reasoning, real-world problem-solving, and the application of knowledge in unique contexts, all of which require students to engage in higher-order cognitive processes.” Such assessments inherently resist easy AI completion because they are more open-ended and contextual.
Some practical strategies for “AI-resistant” assignment design include:
-
Ask Unique or Personal Questions: Design prompts that “refer to something highly specific to your course or not easily searchable online”, or that relate course material to local or personal contexts. The more recent, specific, or grounded in personal experience the task is, the less likely an AI can produce a credible answer. For example, instead of a generic essay on a theory, ask students to apply that theory to a current event on campus or to their own work experience. Generative AI trained on broad internet data will have trouble with very local, up-to-the-minute, or personal angles.
-
Incorporate In-Class and Verbal Components: It’s harder for students to misuse AI if part of the assessment happens live or orally. In-class exams or written assignments completed under supervision naturally limit AI access. Even for take-home work, you can require students to present or defend their work orally. For instance, have students do a short in-class presentation or a Q&A with the instructor after submitting a paper. “Find opportunities for students to present, discuss their work, and respond to questions… To field questions live requires students to demonstrate their understanding of the topic.” This not only deters cheating but also builds communication skills.
-
Emphasize Process Over Product: When assignments unfold in stages, students must show their thinking at each step, making it more obvious if AI did the work. Instructors are using multi-step submissions–e.g. “submit an outline, list of sources, or first draft before the final product”–and iterative feedback. By “including multiple steps such as outlining, drafting, peer-reviewing… [the design] emphasizes the process over the outcome.” Students might also be asked to turn in notes, brainstorms, or a journal of how they developed their ideas. This not only catches misuse (since an AI-generated final draft with no evidence of process is suspicious) but also fosters metacognition, as students reflect on how they arrived at the result.
-
Add Reflection and Meta-Cognitive Tasks: Even if AI can help generate an essay, it cannot easily reflect on a student’s personal learning. Thus, adding a short reflection component can ensure authentic student voice. For example, “ask students to briefly write about a source or approach they considered but decided not to use and why” or “submit a reflection on how the knowledge or skills gained from the assignment apply to their professional practice.” Such reflections tie the assignment to the student’s own thought process and future goals, preserving authenticity. They also promote critical thinking by having students evaluate their choices and learning.
-
Design Multimodal or Practical Outputs: Presenting knowledge in formats beyond plain text can reduce AI interference. Current AI tools are far less capable of generating multimodal work (like video presentations, data visualizations, physical prototypes, etc.). UBC’s teaching center suggests letting students “demonstrate their learning through videos, podcasts, infographics, etc.”, or annotate media, or engage in case-based problem solving–activities where AI has limited ability to produce a polished final product. For example, a student could be assessed on a lab experiment they conduct and record, accompanied by a reflective analysis; ChatGPT can’t perform or fake the lab experiment. These authentic tasks require students to apply skills in real-world or creative ways that AI can’t fully replicate.
The general consensus is that no assignment is 100% “AI-proof,” especially as the technology evolves. However, by anchoring assignments in personalized, contextualized, and higher-order tasks, educators can “reduce the risk of GenAI doing the work for students”. Not only does this protect academic integrity, it often makes the assignments more engaging and meaningful for students–thereby enhancing authentic learning.
Integrating AI to Enhance Learning (Ethically)
On the other side of the coin, many experts argue that we should teach students to use AI tools responsibly as part of their learning. Instead of treating AI purely as a threat, these approaches aim to integrate AI into assignments in structured ways that actually increase critical thinking and real-world skills. The goal is to harness AI’s capabilities (as research assistant, tutor, data analyzer, etc.) while ensuring the student remains actively involved and reflective about the process.
Here are some structured approaches to incorporating AI into coursework without undermining learning:
-
AI-Critical Analysis Tasks: One powerful model is to have students use a generative AI and then critique or fact-check its output. For example, an assignment might require students to “use AI to generate content and then critically evaluate its accuracy, bias, and citations”, assessing the reliability of what the AI produced. Similarly, students could be given an AI-written essay or solution and asked to analyze its strengths and weaknesses. UPenn’s Center for Teaching suggests asking students to “compare multiple versions of an AI-generated approach to problem-solving” or to “determine and implement strategies for fact-checking AI-generated assertions” as part of an assignment. This turns AI into a case study to sharpen students’ information literacy and critical thinking–they learn to not accept AI output at face value, digging into why an answer might be right or wrong.
-
AI as a Learning Coach: Rather than having students submit AI-generated work, instructors can allow (or even encourage) them to use AI as a personal tutor or editor during the process. For instance, a student might draft an essay, then ask ChatGPT for feedback or help with clarity, and finally revise the essay themselves. In fact, Notre Dame’s learning center notes that AI can “review an essay and provide feedback” on things like argument weakness or style, and students could even be asked to “turn in the transcript of their discussion with the AI as part of the assignment.” This way, the product is still the student’s work, but AI served as a tool to improve it–similar to how a spellchecker or Grammarly might be used (but now for higher-level feedback). Such use of AI can free up time for students to focus on idea development and organization, while also teaching them to critically evaluate the AI’s advice. It’s important, of course, that students remain responsible for the content; instructors often specify that students must verify any AI-suggested changes and remain accountable for the final work’s accuracy and academic integrity.
-
Collaborative Problem-Solving with AI: Another approach is to position AI as a collaborator or brainstorming partner in complex tasks–but not letting it do everything. For example, an assignment could have students “use AI to draft an initial hypothesis,” then the student “gathers and cites evidence to support or refute it,” and finally the student submits both the original AI output and their own analysis. This way, the AI might provide a starting point or a different perspective, but the student must apply higher-order skills to evaluate and build upon it. Likewise, students might “brainstorm with AI… generate many possible positions or project ideas,” then “decide which one to pursue and why,” explaining their choice. These kinds of activities mirror real-world scenarios in many professions, where AI can be a productivity tool but humans must exercise judgment. They also explicitly train students in AI literacy–learning how to get useful input from AI and how to refine AI outputs, rather than using it blindly.
-
Prompt Engineering Exercises: Because the quality of AI output depends heavily on the prompt, some instructors are creating assignments around writing effective prompts. For instance, students might be tasked to “prompt an AI to generate responses on a topic the student knows well, and then evaluate how different prompt styles impact the quality and accuracy of the answers.” This not only demystifies how generative AI works, but also gets students thinking about the material from the AI’s perspective (what information and context does the AI need to answer correctly?). In technical or data-driven courses, students could research how prompting is used in their field, or experiment with improving a prompt to get a more refined solution from an AI. Such exercises treat AI as “a tool to be mastered” rather than an all-knowing oracle–students practice critical thinking in the way they query AI and interpret its output.
-
Optional vs. Required AI Use: Some redesigned assessments give students the option to use AI for certain parts, with proper attribution, while others may require AI use as a learning outcome. For example, an instructor might say: “You may use ChatGPT to help you brainstorm sources for your literature review, but you must cite any AI contribution and still write the review yourself.” This teaches proper citation of AI and reinforces responsibility for the final content. On the more integrative end, an assignment could explicitly ask students to use a generative AI tool and then evaluate its effectiveness–essentially assessing the student’s skill in using the AI. The Victoria (Australia) education guidance gives examples like requiring students “to critically analyse and evaluate the outputs of a generative AI tool, including refining their prompt to produce higher quality outputs and cross-referencing AI outputs with credible sources.” In such cases, part of what’s being assessed is the student’s ability to leverage AI appropriately. This can enhance learning by focusing student effort on analysis and evaluation (instead of, say, spending time on low-level initial drafting that an AI could do).
In all these integrative approaches, the emphasis is on transparency and student accountability. Students should always be clear about when and how they are using AI and give proper credit. In fact, new citation guidelines from APA, MLA, Chicago, etc., have been established for AI-generated content. Many instructors now explicitly require students to cite AI assistance (just as they would cite a source) if it was used, and to explain their reasoning in using it. An example guideline is: if you use AI to help with an assignment, you must include a note describing what tool was used, how, and what parts of the submission were influenced by it. This not only upholds integrity (no secret cheating) but also turns AI use into a learning reflection. Students must think about what the AI contributed and ensure they understand and can justify the result–reinforcing critical engagement rather than passive copying.
Fostering Academic Integrity in an AI-Enabled Environment
Whether choosing to restrict AI or to integrate it, academic integrity remains a central concern. Simply banning AI tools is difficult to enforce (AI-generated text is hard to reliably detect, and outright bans might drive usage underground). Thus, experts encourage a proactive, educational approach to integrity in the age of AI:
-
Set Clear Policies and Rationale: Instructors should communicate early on if and how AI tools may be used in a course or a particular assignment. If AI use is forbidden or limited, explain to students why–e.g. “for this task, I need to see your unaided abilities in order to meet the learning objectives”. If AI use is allowed in specific ways, outline those boundaries clearly (perhaps referencing a framework like the AIAS level or a school policy). Students are more likely to follow the rules if they understand the purpose behind them. Some faculty now include a “Generative AI policy” statement in syllabi or assignment instructions, defining cheating vs. acceptable use. This clarity is part of what the AIAS framework also aims to provide: “greater clarity and transparency for students and educators” so everyone knows what’s acceptable.
-
Build a Culture of Integrity and AI Literacy: Academic integrity in the age of AI isn’t just about enforcement; it’s about educating students on ethics. Institutions are urging instructors to “provide students with clear expectations about values, responsibilities and behaviors” related to both traditional integrity and AI use. This might include class discussions about the ethics of AI in academia, honor code pledges that specifically mention AI, or lessons on how to cite and use AI tools properly. One report emphasizes “the development of AI literacy among students and staff, ensuring that all parties understand how to use AI responsibly and the potential consequences of misuse.” If students appreciate the why of academic honesty (for instance, that misusing AI cheats them of learning and can be unfair to peers), they are more likely to self-regulate. Several universities also offer workshops for students on using AI ethically, turning what could be a temptation into an opportunity to learn about intellectual honesty in a new context.
-
Assessment Design to Discourage Misuse: As detailed in previous sections, smart assessment design is itself an integrity measure. By “redesigning our assignments to include in-class elements, direct connections to local and current issues, and a focus on the process in addition to the product,” educators can make improper AI use far less tempting or effective. For example, if a student knows they will have to do a live Q&A on their paper, they are less likely to submit something they don’t understand (whether from AI or elsewhere). If an assignment requires personal reflection or unique data, a ChatGPT-generated generic essay won’t cut it. This design approach shifts the focus from trying to catch cheating to preventing it through engagement. It aligns with the advice that ensuring integrity “goes beyond simply policing AI use–it requires a comprehensive approach” integrating pedagogical strategies.
-
Honor Statements and Oral Checks: In addition to structural design, instructors can use straightforward integrity checks. Some ask students to sign an academic integrity statement on exams or assignments affirming that their work is their own and any assistance is cited. Others implement brief viva voce (oral) exams or follow-up interviews: after a submission, a teacher might have a 5-minute conversation with the student about their work. If the student wrote code with AI help, can they explain how it works? If they turned in an essay, can they discuss its content extemporaneously? These random or targeted check-ins authenticate student work in a low-stakes way and can deter would-be cheaters when announced as a course practice.
-
Use of Detection Tools (Cautiously): While not a pedagogy per se, it’s worth noting that some institutions are exploring AI-detection software. However, current consensus is that “GenAI detectors are not accurate” enough to rely on, and false accusations could harm trust. The emphasis is therefore on the above educational and design strategies first. Technology may assist (for instance, plagiarism checkers are incorporating some AI-detection features), but no tool can definitively prove a passage was AI-written. Instructors are advised to follow up suspicious work with human judgment–e.g. comparing to a student’s past writing or conducting an oral discussion to verify authorship. Over time, it’s expected that students and instructors will develop mutual understanding that using AI deceitfully is both detectable and detrimental to learning.
Conclusion
In response to generative AI, higher education is evolving assessments along two complementary paths: (1) designing more authentic, thought-provoking tasks that showcase students’ own skills and defy easy automation, and (2) integrating AI as a learning tool under guided conditions to prepare students for a world where AI is ubiquitous. The driving question for every assignment is now, as one workshop framed it, how should we respond to AI’s impact on this assessment–embrace it, adapt to it, or resist it? There is no one-size-fits-all answer; the choice depends on the learning objectives and the value that the task is meant to measure or instill.
What’s clear from emerging research and practice is that simply doing nothing is not an option. Educators are encouraged to proactively adapt their assessment strategies to uphold academic standards. This might mean applying an established framework–for example, using the AIAS or a similar scale to explicitly set the level of AI usage allowed, or following guidelines to decide that a certain exam should be “AI-free” while a project can be “AI-enhanced.” It also means staying true to longstanding principles of good pedagogy: clarity of purpose (why students are doing this task), alignment with learning outcomes, and creating opportunities for students to think critically. Generative AI is indeed a disruptive force, but with thoughtful redesign, assessments can continue to be fair, rigorous, and authentic. In fact, this disruption is prompting a healthy re-examination of what we ask students to do and why–which ultimately can lead to deeper learning and skill development that will serve students in an AI-rich future.
Sources:
-
Perkins, M. et al. (2024). “The Artificial Intelligence Assessment Scale (AIAS): A Framework for Ethical Integration of Generative AI in Educational Assessment.” Journal of University Teaching & Learning Practice, 21(6).
-
Shanto, S. S. et al. (2023). “PAIGE: A generative AI-based framework for promoting assignment integrity in higher education.” STEM Education, 3(4), 288-305.
-
Hoke, R. (2024). “Assignment Design: Is AI In or Out?” (University of Pennsylvania Center for Teaching, Learning & Innovation).
-
Education Victoria (2024). “Generative AI–Promoting academic integrity (Guidance)“.
-
Sadiq, M. et al. (2025). “Ensuring academic integrity in the age of ChatGPT: Rethinking exam design, assessment strategies, and ethical AI policies in higher education.” Contemporary Educational Technology, 17(1).