Body
Generative AI is a now part of the environment your students learn in, whether or not it appears in your syllabus. This guide helps you make deliberate, learning-centered choices about it, grounded in your goals, your students, and your discipline (rather than reaching for a blanket policy). The aim is not to police AI, but to build the judgment and agency your students need to use it well or set it aside.
Table of contents:
Most AI guidance focuses on how the tools work and how to use them. While that matters, it isn't enough. The Center for Digital Thriving at Harvard’s Graduate School of Education offers a more complete lens that aligns well with our human centered values here at Union College. It involves three literacies:
- technical (how AI works in practice),
- market (the business models behind the tools), and
- self (who you are and who you are becoming)
The three literacies are mapped across three modes of agency: acting individually, collectively, and through proxies (others who act on our behalf, now including AI agents). The nine cells below are a useful map for locating where your course already builds agency and where it could improve (if desired).
| |
Individual agency
(you act directly) |
Collective agency
(we act together) |
Proxy agency
(others act for you) |
Technical literacy
(how AI works) |
Understanding how generative AI works; discerning if, when, and how to use it. |
Being transparent about AI use; building trust; learning together. |
Knowing what AI can and can’t do; how design choices and policies shape your access and options. |
Market literacy
(the business behind the tools) |
Being alert to the data collected about you, not for you; choosing what to share. |
Advocating for AI that’s accessible, safe, fair, and built without exploiting people or the planet. |
Recognizing hidden incentives in AI tools, platform settings, and policies. |
Self literacy
(who you are becoming) |
Knowing your values, goals, and moral compass; having humility about your limits. |
Co-creating shared values and norms around AI use, with care for others in your community. |
Noticing when AI agents, people, or policies uphold or compromise your values. |
Adapted from Tench, B. (2026). Agency-Centered Framework for AI Literacy. Center for Digital Thriving, Harvard GSE. Licensed CC BY-NC-SA 4.0.
The proxy column is where the latest discussion in AI literacy lives. As tools shift from chatbots that you prompt to agents that act (e.g., booking, drafting, deciding on your behalf) the question is no longer only “did the student use AI?” but “does the student notice when an agent is making choices for them, and can they judge those choices?”
Before writing any policy, decide whether and where AI serves your learning goals. There is no single right answer and different courses will land in different places. Work through the dimensions that matter for your discipline and specific course goals:
- Disciplinary practice. What are professionals in your field actually doing with AI and talking about?
- Ethics and responsibility, including academic integrity.
- Data, privacy, and security.
- Environmental and community impacts.
- How to use it (the practical skill of working with the tools)
- Politics, law, and regulation.
- Economics and business of the tools.
- Impact on learning (where AI would enhance the practice of an essential skill, and where it would short-circuit it)
If you’d like a sounding board, book a consultation with Learning Design & Digital Innovation. Often, refining your goals and the structures that support them already addresses most of what generative AI raises.
Start from what you want students to know and be able to do (long after the course is over) and let that drive everything else. You can use this Course Assessment Map (downloadable template) which steps you through design decisions. For each learning goal, name the evidence of understanding that demonstrates students deeply comprehend and then, the assessment that would produce said evidence. Consider the full range of "significant goal" types: foundational knowledge, application, learning how to learn, deeply human skills, real-world connection, transfer, and AI literacy itself.
Fink’s Taxonomy of Significant Learning, from Creating Significant Learning Experiences, L. Dee Fink. A prompt for goals beyond content recall.
This is also a moment to ask whether any goals are worth updating for an AI world:
- Transfer: what students can carry into other courses and retain years later.
- Learning how to learn: metacognition, discussion, note-taking, close reading.
- Human skills: curiosity, learning agility, teamwork, communication, critical thinking, judgment, taste. (Watch Daniel Pink’s The 6 Skills AI Will Never Replace.)
- Why it matters: the connection to students’ lives and professional goals.
If "designing backwards" with significant learning goals is new to you, book a consultation with Learning Design & Digital Innovation.
Test your current assessments honestly. Paste an assignment prompt into a generative AI tool (we recommend the College supported Gemini) and look at the result: would that work meet your goals? Would it earn a B or higher? Could you tell it was AI-generated? If AI can produce a passing response, the assignment is likely measuring the wrong thing. Put your assignment to the two-question test:
The two-question test
- Could a student do well on this without really understanding the content?
- Could a student do poorly on this while actually understanding it deeply?
If either is true, the assessment isn’t yet measuring what you care about.
Redesign so students must make their thinking visible. A helpful predictor of whether they’ll offload the thinking to AI:
Students are more likely to offload thinking when…
|
…and less likely when…
|
- the assignment is generic
|
|
- only the final draft matters
|
- they must defend their decisions
|
|
|
|
|
|
- they understand how it contributes to their learning
|
- it’s too hard, or there isn’t enough time
|
- steps are scaffolded and time is adequate
|
Rather than rely on a single AI policy in your syllabus for the whole course, ground each assessment decision in five principles drawn from recent research. When AI can generate work experienced graders can’t distinguish from student work, the question shifts from “are students cheating?” to “does this assessment measure what it claims to measure?”
| Principle |
Core insight |
What it looks like in practice |
| 1. Transparency |
When guidance is absent, students build their own ethical rules — creating anxiety for compliant students and cover for others (Corbin et al., 2025). |
State each assignment’s purpose, task, and criteria; set a clear AI-use level; explain the pedagogical “why.” |
| 2. Validity over detection |
When AI can produce competent-looking work, detection fails to tell you what a student knows. This is a measurement problem before a moral one (Dawson et al., 2024). |
For each task, choose secured (verifies independent capability: in-class, oral, lab) or open to AI (human contribution is transparent and measurable). |
| 3. Process over product |
Assessing only the finished product fails when AI can generate polished work instantly (Kickbusch et al., 2025). |
Require drafts, revision histories, reflections, peer review, and (where AI is used) a brief use statement. |
| 4. Evaluative judgment |
As content generation is automated, the essential human capability is recognizing quality — in work produced without, with, and by AI (Bearman et al., 2024). |
Assign critiques where students evaluate AI output alongside strong human work, find errors and hallucinations, and improve it — then apply that judgment to their own work. |
| 5. Programmatic coordination |
Isolated redesigns aren’t enough; validity holds up better when responses are coordinated across a program (Lodge et al., 2023). |
Map assessments across the major with your department colleagues, where independent demonstration is required, where professional tool use is taught, and how methods are varied. |
For a shared vocabulary with students on how much AI a task permits, the AI Assessment Scale (AIAS) is a practical companion. Its current version (2.1) uses five levels — No AI, AI Planning, AI Collaboration, Full AI, and AI Exploration. It is framed less as a permission ladder than as a scaffold for redesigning the task so the grade still means what it claims.
Once you’ve decided an assessment’s approach, communicate it clearly. Transparency in Learning and Teaching (TiLT) simplifies each assignment to three moves (and this is where you make your AI expectations explicit).
The three elements of a transparent assignment.
- Purpose — why does this matter? How the assignment fits the bigger picture and benefits the student, including the skills and knowledge gained.
- Task — what do I do? Clear steps and dates, including how AI can and cannot be used, and how that aligns to the purpose.
- Criteria — how will I be assessed? A rubric, samples of strong and weaker work, or a co-developed rubric.
Always include "the why.” When students understand the rationale, "compliance" is swapped out for buy-in. The TiLT Assignment Design (above graphic) gives you a reusable structure.
Try it now: TiLT one of your assignments
Pick an assignment and run it through the three questions: is the purpose clear and connected to why students benefit, is the task clear, are the criteria clear with a plan to show sample work? If not, revise it. Show it to a colleague who can read it as a student would, or use LDDI's interactive TiLT AI Coach to get feedback and reformat it into the TiLT structure in minutes. If you run into any issues with the TILT AI Coach (or have feedback to improve!), email Denise Snyder.
An AI policy in the syllabus and guidance at the assessment level is a great start, but talking with your students probably matters the most. Be specific about your own use (or non-use) of AI, ask students to research the risks (bias, hallucinations, environmental cost), and revisit expectations before each major assessment (not just on day one). One good way to surface the gray areas together is the Align on the Line activity: present a specific AI use and have students place it on a scale from “totally fine” to “crosses a line,” then discuss where the lines fall and why.
Align on the Line, for making ethical lines visible with students. Adapted from Tench, B. & Weinstein, E. (2025), Center for Digital Thriving. Licensed CC BY-NC-SA 4.0.
The recommendation is clear across higher education: do not rely on AI detectors for integrity decisions. They produce false positives that harm students, are biased against non-native English writers and some neurodivergent students, and are easily bypassed (and as models improve, reliable detection only gets harder). A Stanford study found detectors flagged over 61% of TOEFL essays by non-native speakers as AI-generated, and independent testing has put some tools’ false-positive rates far above their advertised figures. Spend the effort on assessment design instead:
- Drafts and revisions that show the evolution of thinking.
- Annotated critique of an AI-generated draft.
- In-class checkpoints or short (6–8 minute) oral defenses.
- AI use statements naming the tool, the purpose, and what was verified or changed.
Grade their thinking process, not just the assessment product.
This is market literacy in practice: with many consumer tools, YOU are the product, and student data can be collected, reused, and retained. A few guidelines:
- Use the College’s supported tools, which carry enterprise-grade data protection through your Union College account.
- If you require a non-supported tool, never make creating a personal account a condition of a grade. Offer an alternative path, and let students decide what they’re comfortable with.
- Never paste student work with identifying information, or any medium-to-high-risk data, into a tool you don’t trust. Assume no FERPA protection unless the College has established it with the vendor. If you have questions about this, contact Ellen Yu in ITS.
The same reflection applies to your own practice. Before you lean on AI for lesson slides, quizzes, rubrics, or grading, ask:
- Where in my workflow do I use (or plan to use) generative AI?
- What am I hoping to gain (e.g., time, consistency, creativity)?
- What might I risk by automating that part (e.g., personal feedback, student connection, unintended signals about the value of student work)?
- What guardrails keep the human element central (e.g., human review of all outputs, disclosure to students, iterative drafts with my own feedback)?
Your AI approach isn’t set in stone, and you don’t need to be an expert to model thinking critically about an emerging technology. Revisit it, talk about it, and reach out to Learning Design & Digital Innovation whenever you’d like a partner in the work. For questions of academic integrity, see the Honor Council guidance on AI-generated content.
Key sources: Bearman et al. (2024); Corbin et al. (2025); Dawson et al. (2024); Kickbusch et al. (2025); Lodge et al. (2023); Perkins, Roe & Furze (2024, AIAS); Liang et al., Stanford HAI (2023); Center for Digital Thriving, Harvard GSE (2025–2026).
If you are having difficulty or you have unanswered questions, please contact the Help Desk through the ITS Service Catalog or call (518) 388-6400.