R3 3.13 November 21, 2025 AI-Aware Course Design: What Not To Do
Six pitfalls to stay away from as we scramble to cope with generative AI.
If you’ve been reading this newsletter for a while, you know that I’ve made a lot of recommendations for how to design around AI – things like how to lead productive conversations among faculty, align AI practices to particular disciplines and so on.
This time, I’m going to change tack and share approaches that I’m less enthused about. Here’s a short list of the ones with the most potential to lead our efforts astray.
Red-, yellow-, and green-light systems for orienting students to AI policy in a course
One of the approaches I’ve seen floated for a while is a simple framework mapped onto one of the most familiar “stop–go” systems out there: the traffic light. Of course, this takes lots of forms depending on who’s using it, but the versions I’ve seen tend to look like this:
Red light means no AI is ever acceptable in the course—at least on the student side. This approach may especially appeal to faculty who want to limit or avoid interactions with AI.
Yellow light is “use with caution”: students are supposed to read instructions carefully and cautiously incorporate AI accordingly.
Green light is “anything goes.”
On the face of it, this seems so intuitive, and it is a simple, at-a-glance indicator of how AI is likely to be addressed in the course. That simplicity does matter, especially for institutions taking on the admirable task of trying to create coherent, consistent systems for AI policy across a campus.
But it also gives me the impression of being faculty-oriented and sanction-based - there to get across a fixed rule as a function of faculty preference and not much else. To be sure, behind the scenes faculty (or whoever is in charge of the policy) may be putting a lot of thought and care into assigning a label. But as it faces students, that label doesn’t do much to illuminate or even acknowledge that care for student learning, student well-being or other drivers of the policy.
More seriously, the big issue I have with it is this: What on earth does the green light actually correspond to? While “anything goes” is the intuitive interpretation of a green light, surely even the most AI-friendly, AI-centered instruction does not mean that students can use AI in any way they can think of.
I can’t imagine a faculty member who would be OK with a true “anything goes” approach. Does this mean, for example, that students can go into an online discussion board and simply have AI whip up lots of responses with no active involvement from them? Does it mean that students can have an AI agent log into their course, search for assignments, and simply do the work without any intervention on their part?
If we can’t really make sense of a true green-light policy, then it’s probably not a framework we should be using. And if we end up having to hold an in-depth conversation about what “green light” does and doesn’t mean, there go all the advantages of having a simple, at-a-glance, intuitive system.
Relying on automated AI detectors
This point won’t surprise anyone who has been watching the emergence of AI policies in higher ed. Since the dawn of the latest wave of generative AI tools, faculty developers and instructional designers have rightly criticized automated AI detectors, pointing to their coin-flip levels of accuracy and, concerningly, the way that errors are unevenly applied—potentially flagging non-native English speakers, for example.
As someone who has kept a close eye on what generative tools can do, especially when it comes to serving up human-like text, I’ve never wavered from my initial impression: we will probably never have the ability to say with any real certainty whether student work was generated entirely by an AI tool, by the student, or by some combination of both.
And yet I continue to hear from people intimately involved in crafting campus responses to generative AI that faculty keep requesting support in putting such detection tools into practice. I worry—and here I’m thinking of the long history of hype and hard-selling by tech companies—that there will be too many semi-scrupulous or flat-out dishonest companies willing to oversell what their tools can do to flag unsanctioned uses of AI.
Relying on syllabus policies to buffer the risks of unsupervised assessments like online quizzes
In today’s landscape, it really does make sense to treat online and face-to-face instruction as more distinct modalities than perhaps we have in the past. I’ve long argued that the mind learns the same way regardless of whether digital interfaces are in the middle. However, I’ve had to revise this stance a bit —not so much on cognitive grounds as on practical ones.
The practical fact is that the kinds of assessments we typically use in online courses now present a potentially irresistible temptation for students to simply have ChatGPT (or its peers) do the work.
I could go on a long digression here about how our online course designs haven’t kept pace with potential innovation for a long time, and how we now find ourselves seriously behind as we look for new ways to engage students in fully online courses, especially asynchronous ones. I do believe that stagnation in online pedagogy—wrought by years of institutional underinvestment and lack of incentives for the hard work of true innovation—has left us with stale approaches in too many cases. That argument is one for another time.
Even so, we simply cannot attach high or even medium stakes to something like an online quiz that students complete outside our supervision and expect them to choose the harder path of doing the work themselves, just because of some sharply worded, harshly punitive policy in the syllabus. Especially given the lack of tools we have to reliably detect unsanctioned uses, too many students are simply going to take the risk.
Encouraging students to “just ask ChatGPT” when they have questions about the material
Partly this is because I’ve grown averse to advising anyone to “ask” generative AI tools anything at all - because asking a simple question to find a simple fact is a task best handled by a search engine like Google. ChatGPT and similar tools can answer questions—and, goodness knows, they will try if you let them—but where they really shine is not in giving answers but in executing instructions.
That nuance is something we should be conveying to our students whenever we can: not to dip into their AI tools for quick one-offs, but to develop and refine exchanges in which they remain in charge, with their own goals at the forefront.
Furthermore, asking ChatGPT to expound on subjects where there’s a great deal of scholarly discussion and debate – or even fiery controversy—is likely to bring its worst qualities to the forefront. That includes its well-worn propensity to hallucinate (or, as I prefer, to be flat-out wrong), and to give simplistic treatment to complex issues, ones where students really need to bring their own critical reasoning to bear.
This is not to say that a GPT can never be helpful as a study aid or tutor. I’ve been impressed by research findings showing improvements to student learning as a function of interacting with custom chatbots; over the long run, I think this might turn out to be one of the more promising applications of AI to learning. It also tracks with my own experience over the past semester offering my Introduction to Psychology students a course-specific GPT they can use for study, general guidance, or exam prep. It’s not required, nor is it intended to replace interacting with me or any other substantive aspect of the course, and so far, quite a few students report that they’ve had positive experiences using it on an optional basis.
The difference between thoughtfully configured, course-aligned AI tutoring and plugging random questions into a chat window couldn’t be more clear. In the positive scenario, the expert – i.e., faculty member – has set guardrails around the way it interacts with students (for example, not offering answers to homework or quiz questions), instructed it to steer students toward substantive engagement with the material, and provided specific, properly vetted sources to draw on.
Failing to align our approaches to generative AI with other policies and practices in a course
Here I worry that some faculty—quite reasonably—might respond to institutional calls to address AI by inserting a policy or a brief reference into their syllabi, but leaving it decoupled from the rest of the course.
As I see it, especially when it comes to discouraging unsanctioned, dishonest, or counterproductive student uses of AI, a holistic approach is absolutely necessary. So, for example, if we’re worried that students will rely on these tools to whip up mediocre-but-passable prose for a writing assignment, or to generate answers to a quiz, we should be thinking about why students do that in the first place.
That raises questions of flexibility and how late or missing work is handled. I’ve been a vocal advocate for moving away from the midnight-deadline, no-late-work-accepted-even-one-minute-past, points-deducted-per-day model. In my own courses, I ask students to communicate with me if a deadline is going to be missed, not so much to police their excuses but to create an opportunity to surface remaining questions or barriers that might be getting in the way. And as I often put it to my students when they do ask for extensions: “Take the time you need to do your best work.”
There are, of course, limits and practical parameters around this kind of flexibility, and faculty absolutely must adapt the idea to the unique setting and circumstances of their teaching. The point isn’t to have or not have particular policies or sanctions in place, but to look at the whole structure of a course when figuring out where AI fits into the overall picture and adjust for any discrepancies accordingly.
Similarly, if we haven’t adequately incorporated transparency into a course, students may fall into inappropriate uses of AI because they simply don’t see the point of what they’re doing. Take the classic “post one, reply to two” discussion boards (another example of an online teaching technique that has gone thoroughly stale). If all students are doing is posting some stilted text and walking away, having checked a requirement off the list, it’s no wonder they conclude that AI could do the job just fine. But if we put in the time to show where such an activity fits into the overall goals and flow of the course—and if we do our best to ensure that it truly is a vibrant exchange—we’re at least in the game as far as student engagement.
Believing that just because we’ve excluded AI from a course, we’re promoting critical thinking
These days, hardly any conversation about AI and pedagogy goes by without some mention—often in passing—of “critical thinking.” This isn’t, in itself, a bad thing. Critical thinking, and higher-order thinking of all kinds, is arguably the highest and best aspiration for any serious teacher and for any impactful learning experience.
But I’m disturbed by the speed with which questionable research findings claiming to link AI to declines in critical thinking make it into feeds and even go viral. Faculty who see such headlines—“critical thinking hampered by AI, study shows”—might in good faith assume that AI damages critical thinking capability by its mere presence and reshape their courses accordingly.
But as of today, remarkably little evidence has established a solid link between AI use and across-the-board decreases in critical thinking. And I don’t think this is just a matter of a lack of research. Cognitive psychologists who study the subject have cautioned that there may not be a single “critical thinking” ability located in the mind or brain. It stands to reason, then, that it’s unlikely any technology, AI or otherwise, would have the power to selectively affect that one thing. Critical reasoning, as we see it in my field, is something that develops in a specific context, or even within a specific discipline.
It’s a wonderful thing, of course, for faculty to work on strengthening critical reasoning in the unique way it shows up in their field —and many of the most engaged and effective faculty do just that. But I would argue that in higher education overall, we have a lot more to do in this realm. Strengthening critical thinking is not only contextual; it’s also hard—much harder than it might appear at first glance. Students need far more practice, over a much longer period, than they typically get.
So I worry that we’ll all be lulled into a false sense of security if we think we’ve struck a blow for critical thinking simply by implementing some kind of red-light policy.
I’m sure that semesters to come will bring many more positive and negative lines of advice on this singularly difficult issue. As the research literature in this area builds up, we’ll get a better sense of what actually happens when we come at the problem in one way or another. In the meantime, though, I think we should also be having these conversations about what not to do—and why.