AI in Teaching and Learning

Map the learning before making the AI decision.

• Map the learning before making the AI decision.
  • Start with the competencies.
    What should students know, practice, and be able to do by the end of the course?
  • Map the learning activity.
    What will students do to develop and demonstrate those competencies?
  • Use the ICAP Framework as a filter.
    Look at what students will actually do during the activity:
    • Passive: Receive information without doing anything further with it.
    • Active: Select, manipulate, identify, or rehearse information.
    • Constructive: Generate something beyond what was provided, such as an explanation, interpretation, comparison, question, or justification.
    • Interactive: Build knowledge through substantive dialogue in which contributions respond to and develop one another.
  • Keep AI out when it makes student thinking more passive, less visible, or unnecessary.
  • Bring AI in when it helps students engage more deeply, such as by requiring them to:
    • compare multiple responses;
    • identify errors or assumptions;
    • explain and justify decisions;
    • revise work using feedback;
    • test an argument or solution;
    • defend their reasoning; or
    • build upon ideas through dialogue.

• Clarify AI’s role in each activity.

  • No AI: Students must demonstrate the competency independently.
  • AI as a learning partner or scaffold: AI supports practice, feedback, questioning, or revision, while students remain responsible for generating and evaluating ideas.
  • AI as an offload tool: AI handles a bounded task that is not itself a learning goal, such as reformatting information or producing an initial transcription.
  • Make student thinking visible.
    Ask students to explain, annotate, compare, reflect, revise, verify, or defend how they reached their conclusions.

  Why ICAP matters here?

  ICAP helps faculty examine the quality of student engagement, rather than assuming that using a new technology automatically produces active learning. An AI activity may support learning when it prompts students to generate, evaluate, connect, or defend ideas. It may bypass learning when students simply request and accept an answer.

  Additional Resources: 

• ICAP Framework

• Using Backward Design to Plan Your Course (From Ohio State University) 

   

   

Students need more than a list of AI rules. They need faculty guidance that helps them understand how, when, and why AI may or may not support their learning. This means creating a course environment where students can practice responsible use, reflect on their decisions, and continue building their own judgment.

• Clarity: What AI use is allowed, limited, required, or not appropriate for this course and each assignment?
• Purpose: Why does this expectation matter for the learning students are expected to do?
• Modeling: How do people in this discipline question, verify, document, revise, or choose not to use AI?
• Practice: Where can students safely experiment with AI, compare responses, identify limitations, critique outputs, or revise their work?
• Reflection: What did AI contribute, what did the student contribute, and what decisions still required human judgment?
• Verification and accountability: How should students check AI-generated claims, sources, calculations, interpretations, or recommendations? Who remains responsible for the final work?
• Feedback: How will students know whether both their disciplinary thinking and their AI-related decision-making are improving?

• Access and alternatives: Do all students have equitable access to the required tools, appropriate technical support, and a reasonable alternative when privacy, accessibility, cost, or other concerns make AI use difficult?

  Faculty do not need to have every answer about AI. Students benefit when instructors communicate their expectations clearly, explain their reasoning, model thoughtful use, and create space to revisit those expectations as the course and technology evolve.

Additional Resources: 

• Cornell's AI Attribution Guidelines
• Communicating Course Policies and Talking with Students - UC Berkely 

Not every part of a learning task serves the same purpose. Some moments help students build foundational knowledge, some provide supported practice, and others ask students to demonstrate judgment, synthesis, or transfer.

Faculty can decide where AI belongs by identifying the thinking students need to practice at each stage.

Before consulting AI:
When foundational knowledge or an independent first attempt matters, students may need to retrieve, explain, analyze, solve, or form an initial position on their own. Actively recalling information strengthens long-term retention, while an initial attempt gives students something of their own to compare, question, and revise.

With AI support:
Students can use AI to generate alternatives, test counterarguments, seek feedback, role-play perspectives, compare approaches, identify gaps, or revise their work. These uses are most productive when students must actively evaluate and build upon the response rather than simply accept it.

Independently of AI:
Students should still have opportunities to demonstrate the knowledge and capabilities the course expects them to possess themselves. Depending on the discipline, this may include foundational knowledge, disciplinary judgment, ethical reasoning, interpretation, synthesis, oral explanation, final decisions, or reflection.

Appropriately offloaded to AI:
AI may handle parts of a task that are not central to the learning goal, such as reformatting material, reorganizing notes, producing practice examples, clarifying instructions, or completing routine transformations. What can be offloaded will vary by course: a task that is incidental in one discipline may represent essential practice in another.

A useful question is:

If AI completes this part, what thinking will the student no longer practice?

If the answer names an important course competency, keep that thinking with the student or redesign the activity so the student must examine, explain, verify, or defend the AI-supported work.

When AI is available, the final product may not be enough to show what students understand. Build in moments where students make their thinking visible through multi-step assignments.

• Process: Ask students to show drafts, checkpoints, notes, or decision points.
• Explanation: Ask students to explain why they made specific choices.
• Reflection: Ask what AI helped with, what they questioned, and what they still had to decide.
• Revision: Ask students to show how feedback shaped their next version.
• Transfer: Ask students to apply their learning to a new case, problem, or context.

For more ideas for assignment design using AI, check out the AI Pedagogy Strategy Deck

Additional Resources: 

Project Zero Thinking Routines Toolbox - Strategy Collection | Harvard Graduate School of Education

A collection of short, adaptable routines that help students observe closely, organize ideas, explain reasoning, consider different perspectives, and reflect on how their understanding is changing. Routines such as Claim–Support–Question and Connect–Extend–Challenge can be used before, during, or after AI-supported work.

Metacognitive Strategies - Practical Guide | Cornell University Center for Teaching Innovation

Provides simple activities that help students examine their own thinking, recognize gaps or errors, and identify what they still need to learn. These strategies can be incorporated through planning prompts, learning reflections, exam wrappers, and brief explanations of decisions.

This is a decision you make for each activity, not one policy for the whole course. A single course can use all three roles at once, and students won't find that confusing as long as you name the role and explain it. What confuses students is one blanket rule applied to activities that actually need different things.

Start with what the activity is supposed to build

Before you assign a role, ask what this activity is supposed to build in the student. Does AI help with that, or get in the way?

The answer depends on where the value sits. Some activities are valuable because of the process: working through a problem, making a first attempt, exercising judgment. Others are valuable because of the product: a polished piece of work where the thinking already happened somewhere else.

Research on "desirable difficulty" (Bjork, 1994) backs this up: struggle and effort are often what make learning stick. If AI removes that struggle at the wrong point, it doesn't make the activity more efficient. It removes the reason you assigned it in the first place.

Ground the three roles in an existing framework

The AI Assessment Scale (Perkins, Furze, Roe & MacVaugh, 2024) gives you more precision than a simple three-way split. It defines five levels of AI involvement: No AI, AI-Assisted Idea Generation/Structuring, AI-Assisted Editing, AI Task Completion with Human Evaluation, and Full AI Integration. It doesn't rank one level as better than another. "No AI" counts as just as deliberate a choice as full AI use, as long as you chose it on purpose.

Your three categories map onto this scale:

• Off-limits maps to No AI. Use this for moments where the thinking itself is the point: a first attempt at a problem, an exam, a final argument, an ethical judgment call. You want students working on their own here. This is also where the ICAP framework (Chi & Wylie, 2014) says active engagement matters most.
• Practice tool maps to AI-Assisted Idea Generation or Editing. AI gives students more chances to practice: generating practice problems, giving draft feedback, offering a counterargument to test against. Research on retrieval practice (Roediger & Karpicke, 2006) and feedback (Hattie & Timperley, 2007) shows this kind of repetition builds durable learning. The student still does the thinking. AI just gives them more reps.
• Partner maps to Task Completion with Human Evaluation or Full Integration. Here AI does more of the generating, and the student's job is to direct, evaluate, and decide what to keep, comparing AI's output against their own first attempt. This only works if the student stays the one making decisions instead of just accepting what AI gives them.

The role can shift within one assignment

This connects to the earlier question about before, with, and without AI. One project might be off-limits during the first attempt, a practice tool during drafting, and a partner during a later brainstorming stage. Same assignment, three roles, each tied to what that stage is asking students to do.

Name the role and explain why, for that activity specifically

Once you decide on a role, say so clearly, and explain why for that specific activity rather than pointing back to one general AI statement in your syllabus. Students follow expectations better when they understand the reason behind them, not just the rule (Hattie & Timperley, 2007). "This is off-limits because the whole point is your first attempt on your own" lands very differently than the same rule stated without explanation.

Rigor doesn't come from the topic or the length of an assignment. It comes from the cognitive work students have to do: applying a concept, explaining a choice, revising based on new information, judging between options, transferring an idea to a new context. Research on effortful learning backs this up (Bjork, 1994; Bransford & Schwartz, 1999): tasks that require real generative work tend to produce more durable learning than tasks that mainly reward polish.

Start with a blunt test: can AI already produce a convincing version of the final product you're assigning? If yes, that's a sign the assignment is testing something AI can already do, not something the student needs to build. From there:

• Identify the specific disciplinary thinking this assignment should develop.
• Add requirements tied to your course's own data, cases, and class discussions, things a generic AI tool can't easily generate.
• Ask students to explain their decisions, trade-offs, and evidence, not just hand in a finished answer.
• Be upfront about where AI is allowed, limited, or off the table for this specific assignment, and explain why, so the boundary itself teaches something.

For a structured way to build all of this into an actual assignment sheet, the TILT (Transparency in Learning and Teaching) framework is worth adopting. It has you state the purpose, task, and criteria for success explicitly, and it has research showing it improves student confidence and success, especially for first-generation and underrepresented students (Winkelmes et al., 2016). Ohio State has a worked example of applying TILT to an AI-integrated assignment: Using the Transparent Assignment Template. The full template and more examples live at tilthighered.com.

Another useful model is the "two lane" approach (Liu & Bridgeman, 2023): decide which parts of an assignment belong in an AI-free lane, work that has to show unaided thinking, and which parts belong in an AI-integrated lane, where AI collaboration is expected and visible. UMass Amherst's Center for Teaching and Learning walks through this with concrete examples: How Do I (Re)design Assignments and Assessments in an AI-Impacted World?

When AI can produce a polished final product, that final product stops telling you much. What you need is evidence of the reasoning that got a student there. That means building checkpoints into the assignment instead of relying on one submission at the end.

Formative assessment research shows students learn more when they get information they can act on before the final grade, not just after it (Black & Wiliam, 1998), and that they learn better when they're prompted to monitor and adjust their own thinking along the way (Schraw, 1998). In practice:

• Break a big assignment into stages: outline, draft, revision, final, each submitted separately.
• Ask students to explain what changed between versions and why.
• Use a short decision log where students note assumptions, trade-offs, and choices as they go.
• Ask what feedback they used, rejected, or revised around.
• Grade the reasoning and revision, not just the final product.

UC Berkeley's Center for Teaching & Learning has a clear practical guide on this shift, moving from grading final products to grading checkpoints like draft plus reflection plus revision: Redesigning Assignments and Assessments. The University of Iowa's Center for Teaching also has concrete tactics, including having students submit an annotated portfolio of drafts or a version history alongside the final work: Where We Are Now: Designing Assignments in the Age of AI.

This is really a question about sequencing. Retrieval practice research shows that actively recalling and attempting something strengthens long-term learning (Roediger & Karpicke, 2006), and feedback research shows it's most useful when students interpret it and act on it themselves (Hattie & Timperley, 2007). AI can multiply how much practice and feedback a student gets, but only if the student is still doing the retrieving, comparing, and deciding.

A workable pattern:

• Use AI to generate practice questions, examples, or counterarguments, more reps than you could hand-build yourself.
• Require students to attempt the task before they consult AI at all.
• Have students compare AI's feedback against their own reasoning instead of just accepting it.
• Ask them to explain what they accepted, rejected, or changed based on that feedback.
• Keep the final judgment and decision with the student, always.

FIU's Center for the Advancement of Teaching has a good worked example of exactly this pattern in a biology course: AI generates a bank of low-stakes practice quiz questions with immediate feedback, and the actual exam is a case study students still have to reason through themselves, with a short reflection on how AI shaped their thinking. See Design Strategies for Assessing Learning with AI.

Go over the AI Pedagogy Strategy Deck for more ideas as well! 

The ICAP framework (Chi & Wylie, 2014) is the clearest research base here. It sorts student engagement into four modes: passive, active, constructive, and interactive, and finds that learning improves as students move from passive toward interactive. The practical takeaway: don't just ask whether students are using AI in an activity, ask what mode that use puts them in.

Ways to design toward the interactive end:

• Ask students to critique AI's reasoning instead of treating it as an authority.
• Have students compare their own first attempt to an AI-generated response.
• Use AI to generate cases, scenarios, or role-play partners for students to work through.
• Ask students to explain where AI's output is strong, weak, generic, or misleading.
• Build peer discussion around where student thinking and AI output diverge.
• Keep students in the role of evaluator or decision-maker, never just the recipient of an answer.

The UNH Teaching & Learning Resource Hub has a short practitioner guide on Applying the ICAP Framework to Improve Classroom Learning. Stanford's Center for Teaching and Learning also has discipline-specific examples of critique-and-compare activities in AI Teaching Strategies.

Knowledge doesn't sit in a vacuum. It's shaped by the context and activity in which it gets used (Brown, Collins, & Duguid, 1989), and transfer research shows students need chances to apply what they know flexibly across contexts, not just reproduce it in a familiar format (Bransford & Schwartz, 1999). An authentic assessment asks students to do something close to what a practitioner in the field actually does, not summarize a general idea back to you.

In practice:

• Use real or realistic cases, data, audiences, or professional constraints.
• Ask students to apply your course's specific frameworks to a specific context, not restate them generally.
• Require students to explain judgment calls, trade-offs, limitations, and ethical considerations.
• Build in transfer: ask students to apply the same concept somewhere new.
• Consider oral defenses, scenario extensions, portfolios, or reflective memos as alternatives to a single written product.

Students are more likely to act on expectations when they understand the goal, the criteria, and the next step in their learning process (Hattie & Timperley, 2007). In the context of AI, clarity should not only tell students what is allowed; it should help them understand why certain uses support learning and why others may bypass it. AI literacy also involves learning to evaluate, communicate with, and use AI responsibly, which means expectations should help students build judgment, not just follow rules (Long & Magerko, 2020; UNESCO, 2024). 

• Name the level of AI use: Is AI prohibited, limited, allowed with disclosure, encouraged, or required? 

• Make it assignment-specific: A course-wide policy helps, but students need clear guidance for each major task. 

• Explain the "Why”: Tell students why AI is allowed in some parts of the work and restricted in others. 

• Give examples: Show what acceptable, questionable, and unacceptable AI use looks like in your course. 

• Clarify responsibility: Students are responsible for the accuracy, ethics, reasoning, citations, and final decisions in their work. 

• Ask for transparency: When AI use is allowed, ask students to briefly explain how they used it and what choices remained on their own. 

• Revisit expectations: Restate AI expectations before major assignments, not only in the syllabus. 

A strong AI policy does more than prevent misuse. It teaches students how to make better learning decisions when AI is available. 

Students are more likely to use expectations well when they understand the standards of the task, what quality looks like, and how to regulate their own work against those expectations (Sadler, 1989; Nicol & Macfarlane-Dick, 2006). In AI-related guidance, the goal is not just to say “allowed” or “not allowed,” but to help students understand the learning reason behind the boundary. 

• Use plain language. Avoid vague terms like “responsible use” unless you define what that means in your course. 

• Separate course policy from assignment guidance. Students need to know the general rule and the specific rule for each task. 

• Show examples. Give 2–3 examples of acceptable, questionable, and unacceptable AI use. 

• Explain the why. Connect each expectation to the learning goal: What thinking are students meant to practice here? 

• Use quick checklists. Before submission, ask students to confirm what AI use was permitted, what they used, and what remained their own. 

• Revisit expectations at the moment of use. Remind students before major assignments, not only on the first day of class. 

If students cannot explain the AI expectation back to you in their own words, the policy may not be clear enough yet. 

AI disclosure should not feel like a confession; it should function as a learning record. Research on self-regulated learning emphasizes that students learn when they monitor their process, evaluate their choices, and understand the standards they are working toward (Nicol & Macfarlane-Dick, 2006). AI literacy also involves knowing how to evaluate and use AI responsibly, not simply whether a tool was used (Long & Magerko, 2020). 

• Ask what they used AI for. Brainstorming, outlining, feedback, coding, editing, summarizing, generating examples, etc. 

• Ask where AI entered the process. Before drafting, during revision, after feedback, or near final submission? 

• Ask what they accepted, changed, or rejected. This reveals judgment, not just tool use. 

• Ask what remained their own. What decisions, interpretations, evidence, or reasoning did they produce? 

• Keep it proportional. A short note is often enough; not every assignment needs a full AI-use log. 

• Normalize transparency. Frame disclosure as part of responsible learning, not as an admission of wrongdoing. 

The most useful AI disclosure does not just answer “Did you use AI?” It helps students explain how they used judgments when AI was available. 

For Academic Integrity regarding AI at Bentley refer to the Guidelines for Responsible Use of AI 

Academic integrity is strongest when it is treated as a learning culture, not only a misconduct system. Bretag and Mahmud (2016) emphasize access, responsibility, detail, and support as core parts of exemplary academic integrity policy. In AI-era assessment, detection alone is not enough, especially because AI detectors can be unreliable and may misclassify some student writing, including non-native English writing (Liang et al., 2023). Instead, build integrity into the assignment design. 

• Use an AI-use level for each assignment. Tell students whether AI is prohibited, limited, allowed with disclosure, or expected — and explain why. 

• Require an “attempt before AI” moment. Ask students to submit an initial idea, outline, problem attempt, or response before using AI. 

• Build in process checkpoints. Use drafts, plans, decision logs, annotated notes, or short progress submissions so learning is visible over time. 

• Ask for an AI-use note when AI is allowed. Students briefly explain what they used AI for, what they changed, what they rejected, and what remained their own. 

• Anchor work to course context. Require students to use course readings, frameworks, cases, datasets, class discussions, or local examples that generic AI output cannot easily replace. 

• Include a reflection or revision memo. Ask students to explain how their thinking changed, what feedback shaped their work, and why they made final decisions. 

• Use short explanation checks. Add a brief oral defense, in-class walkthrough, or follow-up question where students explain their reasoning. 

• Assess judgment, not just polish. Include criteria for reasoning, evidence use, decision-making, revision, and transfer — not only clarity or completeness. 

• Respond to concerns through conversation first. If something seems off, ask the student to explain their process, sources, decisions, and reasoning before treating it as misconduct 

The goal is not to catch students using AI. The goal is to design learning environments where students understand what responsible work means and can show the thinking behind it. 

Responsible AI use is easier to teach when academic integrity is framed as a shared learning practice, not only a compliance issue. Research on academic integrity emphasizes clear expectations, shared responsibility, detailed guidance, and support for students before misconduct occurs (Bretag et al., 2011). AI literacy research also suggests that students need opportunities to evaluate, collaborate with, and use AI responsibly — not just be told whether it is allowed or banned (Long & Magerko, 2020). 

• Talk about AI before the first high-stakes assignment. Do not wait until misuse happens to define responsible use. 

• Name the learning purpose behind your AI boundaries. Students should know why AI is allowed in one task and limited in another. 

• Use low-stakes AI practice. Let students try AI, critique its output, identify limitations, and discuss what responsible use looks like in your field. 

• Model your own decision-making. Show students how you evaluate an AI response, question its reasoning, check its sources, or revise its output. 

• Make integrity part of assignment design. Build in checkpoints, reflections, process notes, AI-use disclosures, or revision memos. 

• Normalize transparency. Treat AI disclosure as part of learning how to work responsibly, not as an automatic sign of wrongdoing. 

• Create room for judgment. Ask students what they accepted, rejected, changed, or questioned when using AI. 

• Use concerns as teachable moments when appropriate. When expectations are unclear or misuse seems rooted in misunderstanding, clarify, reteach, and redesign before escalating.