What does it mean to teach in an AI-aware way — not resistant, not fully integrated, but grounded in pedagogy? In this special live recording from the Perusall Exchange 2026, Eric Mazur sits down with the three authors of the highly anticipated The Norton Guide to AI-Aware Teaching (W.W. Norton, Summer 2026): Annette Vee, Associate Professor of English at the University of Pittsburgh; Marc Watkins, Director of the AI Institute for Teachers at the University of Mississippi; and Derek Bruff, Associate Director of the Center for Teaching Excellence at UVA.

Together, they explore why "AI-aware" is the right framework for this moment, how to make thinking visible when AI can produce a polished final product in seconds, why assessment needs to shift from artifacts to performance, and what one small move every educator can make right now to teach more intentionally in the age of AI.

Eric Mazur: Welcome, everyone, to a very special live recording of the Social Learning Amplified podcast, coming to you from the Perusall Exchange 2026. This year's theme is assessment in the age of AI, and I can't imagine three better guests to help us navigate this landscape. They're the authors of the highly anticipated book The Norton Guide to AI-Aware Teaching, coming from W. W. Norton this July.

Before we dive in, I have an exciting announcement: all three of our guests will be leading a Brussels Engage event starting on July 6th. If you're not familiar with Engage events, they're facilitated communal reading events where you get access to the new book and also engage in the very AI-aware social learning we're discussing today. Be sure to register.

Now let me introduce our panel. Annette Vee is an associate professor of English at the University of Pittsburgh and author of Coding Literacy — she's a leading voice on what AI means for writing and computation. Marc Watkins directs the AI Institute for Teachers at the University of Mississippi and is a pioneer of curious skepticism in the AI classroom. And Derek Bruff is the associate director of the Center for Teaching Excellence at UVA and host of the excellent Intentional Teaching podcast. Annette, Marc, Derek — welcome, and thank you so much for joining me today.

Derek Bruff: Very glad to be here, Eric.

Eric Mazur: Well, good to see you.

Annette Vee: Thank you for inviting us.

Marc Watkins: Yes, very excited to have the conversation.

Eric Mazur: I'm looking forward to it. So let's dive right in. Annette, your book uses the phrase "AI-aware teaching" — not "AI-integrated," and certainly not "AI-resistant." Why was awareness the right word for this cultural moment?

Annette Vee: Well, first of all, AI awareness means paying attention to the landscape — both what our students are currently doing with AI, what AI itself is doing, and what our pedagogical goals are. I think that's actually the only way we can approach it: we can resist within that framework, and we can engage within that framework — it encompasses all of those things. And I think many instructors, at this point, are doing some combination of those things.

So when we were thinking through what a guidebook should look like, we noticed there are things out there that help people resist AI, and other things that help people engage with it. But we really wanted to keep the center on pedagogy — on how we navigate across those things. That, to me, is what AI awareness means in terms of teaching.

Eric Mazur: Marc and Derek, does being aware mean we have to change our goals, or just change our assignments?

Marc Watkins: I think it means we have to question a lot of different things, including our goals, our assignments, and our assessment approach.

One thing that awareness means to me is agency — both for myself as a teacher in the classroom, and for my students. I want them to make intentional decisions about whether or not to use AI, and to be able to advocate for those reasons. And for myself as a teacher, I want to model that same kind of behavior around transparency and disclosure when I'm using AI. Hopefully they'll pick up on those cues too — both in how I design my syllabus and in how I label material as AI-generated or not.

Derek Bruff: Yeah, I think we have to change assessments. I think that's where a lot of this conversation around AI started. People in all professions are using AI in interesting, clever, terrible, ugly, useful, weird ways — it's changing how we communicate, how we understand knowledge. And I think we all have a role to play in helping our students navigate that.

So maybe that means telling students, "We're going to learn how to learn without AI — we're going to do things in a distraction-free, tech-free way, because there's value in that." It could also mean, "I need to prepare you for the profession you're entering, where you'll need to use these tools ethically and effectively" — so we make AI competencies part of our learning objectives. We try not to prescribe which one it should be; that depends on you, your goals, your students, and your context. There's a lot of "it depends" in there. But I think we have to make those choices, and figure out what those high-level choices mean for the course design we do.

Eric Mazur: It's interesting to hear you describe that, Derek. Our students are going to enter a workplace that involves AI — there's no question about it. Maybe we should start by thinking about what AI is going to do to the workplace and society before considering what to do in education, since ultimately we're preparing students for their future careers. So wouldn't it be best to encourage faculty to embrace AI and support the development of, to use your word, AI-aware skills?

Annette Vee: I think we want to be a little careful about just preparing students for the workplace. Higher education does so much more than that. That's certainly part of the landscape we need to take into account — what students are going to do afterward — but there's a difference between workplace training and most college classes, and that's that we need to engage with our students' development. One of the things that really matters there is focusing on critical inquiry — and I'm contrasting that with critical thinking, which is also important. Critical inquiry is about what interesting questions students are asking, what their motivation is, what's driving them.

I think when we look at the reshaping of higher education into something more utilitarian — you just get your degree and move on — that's the kind of thinking that leads students to make bad choices about AI. If they think of their education as simply a hurdle to get through, that's how they end up using AI in negative ways: shortcutting it. So part of approaching higher education in the context of AI is thinking very carefully about those pedagogical goals. They might include workplace preparation, but they're also going to include a lot of really important things about students' development.

Derek Bruff: I want to add to that, because sometimes you hear people say, "Higher ed isn't about preparing for the workplace, so I don't have to pay attention to AI." I don't think that's a good argument. As I said, AI is changing how we communicate, how we encounter information, how we make sense of the world — in our jobs and in other parts of our lives. It's changing how students learn as they use these tools, and there are productive and unproductive ways to use them.

Those are places where our students need our guidance, even setting aside jobs down the road — how do we make sense of information, how do we learn well, what skills and approaches and tools do we need? I think most of us can see ourselves teaching to that.

Eric Mazur: Marc, as a skeptic, what do you have to add to that?

Marc Watkins: I'll try to draw on some concrete details from the book. We crowdsourced many examples from our own teaching and from faculty across the U.S. and internationally. One example I love comes from Maggie Boyd at Boston University. Her team put together an activity called "Pause Before You Prompt," which asks students to think reflectively and critically about why they're going to an AI tool in the first place — not just looking at the output and any bias in it, but asking, "What's drawing me to this? Am I trying to develop my own learning skills with AI, or am I just trying to pass off the work because I feel overwhelmed?" Those are big questions running through students' minds.

As a skeptic, the big thing I keep coming back to with the book is inviting both faculty and students to slow down and consider what this tool is doing — which offers a very frictionless, efficient answer to a lot of the questions we ask it — and to critically engage with that and think about those choices. It goes back to what I said before: it's all about agency, and what we're dealing with in this technology.

Before we go on, I want to remind the audience to use the Q&A box for questions you'd like asked, and the chat for questions to each other or about logistics.

Eric Mazur: As I was listening to what Marc was saying, Derek, I was reminded of something you may be familiar with — if I recall correctly, you were a math major at Harvard, right?

Derek Bruff: Sort of. I taught math there for a couple of years, yeah.

Eric Mazur: Exactly. And you may recall that in the mid-'90s there was a whole back-and-forth over the calculus reform movement and the graphing calculator — should we keep teaching students to plot functions by hand, since that's such a valuable skill, or should we embed graphing calculators and let students build on that automatable skill and learn more about interpreting graphs, derivatives, integrals, and so on? We're at a similar point now with AI, on a much broader scale. The difference is that with AI we can't look forward into the future, but with the calculator, we can look back into the past. I think everyone has now accepted the calculator, and realized it's much better to use that tool to help students develop skills higher than simply plotting a function by hand. Do you think that's a valid parallel?

Derek Bruff: Well, I'll say that some of my math colleagues are still warming up to the idea of calculators in math education, so I don't know that we've quite cleared that hurdle entirely.

Eric Mazur: The fistfights that were breaking out in the mid-'90s...

Derek Bruff: And every time I bring up the calculator comparison, I get somewhat angry pushback, so I'll just expect that in the chat here too. But there are a couple of angles on this comparison. One, which you touched on with the "fistfights," is that this technology entering math education meant math educators had to grapple with what we needed to be teaching students, and whether that needed to change. I think that's true for AI as well — are there fields or bodies of knowledge or skills where we actually need to update what we're teaching, because these AI tools are available to students? I'm not saying that's true across the board, but we have to grapple with it and decide.

The other piece is: what skills and mindsets do students need if they're going to use these tools well? Can they evaluate the output and say, "I think that actually makes sense"? That's something we'd want with graphing calculators too — we'd want students to know what buttons to press, but also to be able to evaluate the output and ask, "Is this reasonable? Is this plausible? Have I done something wrong here?" There are lots of places where the metaphor breaks down, certainly. But thinking about the prerequisite knowledge students need to use these tools well, effectively, and critically — like Annette was saying — is part of that mix. It took math educators a while to figure out how to teach that too, so it's okay that this is hard. We're figuring it out together.

Eric Mazur: And of course, calculus didn't have the same social and economic impact that AI has, as someone in the chat just noted.

Derek Bruff: Absolutely.

Marc Watkins: Yeah — they also didn't try to validate you every time they answered your question.

Derek Bruff: I know.

Eric Mazur: And ask you to carry out another calculation.

Marc Watkins: Yeah, yeah, yeah — "This is the best calculation in the history of the world. Please tell me more."

Derek Bruff: But I also have to share — I took calculus in high school with a really great teacher, Diane Steelman, in South Carolina, and we had graphing calculators. She had a way of helping us understand the concepts through our use of the technology. It wasn't "can we do calculus with a calculator," it was "can we use a calculator to help us understand calculus?" That's a subtle change, but I think it's an important way we can also look at AI.

Eric Mazur: Absolutely, absolutely. I want to shift our conversation a bit toward AI-aware assessment. But before I do, I want to talk about something that's become increasingly important to me — it's actually been a thread through my own pedagogical journey — and that's the idea that now, more than ever, we need to make thinking visible.

It's not the end product, it's not the answer, it's how you get there. In most of life, usually the answer is known or obvious, but the path to it isn't. So we should really train students on process and thinking, rather than just producing an artifact. That was, in a sense, my goal when I introduced peer instruction back in — wow — 35 years ago, in 1991: to make thinking visible.

Now we have AI that can easily produce a polished final product — a derivation, an essay, whatever. So does that mean we should stop assessing the product and start assessing the process of learning instead? Let me rephrase: I'm sure you'll lean toward that, since that's what we're about as educators. The big problem is how do you scale that up? Why do we look at the outcome, the answer? Because it's really easy to do when you have five hundred students.

Marc Watkins: Yeah, the labor involved in all this is massive — even just the labor of responding to it. Part of it is the material conditions of the buildings we teach in; it's very difficult to run active learning in a lecture hall where students can't even turn their chairs more than 180 degrees, in some cases. So you have to work with what you have.

I'm very much about process-based teaching. Annette and I are both writing professors — we work in rhetoric and English — so we're attuned to the fact that when generative AI first appeared, there was no longer going to be the terror of the blank page for students. And a lot of that thinking, that decision-making, used to happen in the rough draft. So we've looked at a lot of strategies for preserving that rough-draft process as much as possible. Marc Marino, for instance, does an "analog sandwich" exercise across the semester with his students.

He has them write in a notebook — not blue books, I'm not a fan of those either — for half the semester. It's very analog, and the readings are all printed out. Then, after the midpoint, he shifts to a digital format, so students have access to whatever tool they want, generative AI or a word processor, and he asks them to critically think about what that shift is doing to their process.

Derek has interviewed colleagues at UVA who've gone back to analog methods too — they don't always have to be pen and paper. You can have students write with a laptop in the classroom, or turn their word processor into deep-focus mode, or turn off their Wi-Fi. It's not all-or-nothing. We can look at different strategies for preserving learning using some analog techniques while also inviting students to explore digital ones. I think that's the pathway I'd like us to consider — we just have to get out of this all-or-nothing idea with AI.

Annette Vee: Yeah — building on what Marc's saying, one thing I want to emphasize is that underneath all this, each of these moments is a deliberate orientation to AI that's chosen and discussed with students. As a teacher, you can choose — though we have somewhat limited agency, as I'm sure many people on this call have noticed; you can say "don't use AI," and students may or may not follow that.

But the deliberate choice, and making sure that choice matters and means something so students get something out of it, is really crucial. I'll throw in another example: the PAIR project coming out of UC Davis, which is wonderful for thinking about intentionality between working with AI and working with peers. It also breaks the process down. So Eric, when you ask, "Should we assess the product or the process?" — really, it should be both. In the process, we help students walk through the different parts and make sure they're making good decisions about AI throughout. PAIR is a peer-review process for papers — it doesn't have to be in composition classes specifically.

It starts with peer review — human peer review, for good reasons, and they have great resources for that — then it moves to AI review, set up through a tool called MyEssayFeedback, with prompts the PAIR team has beautifully designed. Then students reflect and work through the different kinds of feedback, talking back to it, and so on. There's a lot of metacognition happening in that process.

One thing we notice with students is that when they get feedback on their writing — which many of us provide — they often default to a kind of "one and done": here, I'll just give it to AI and get an answer. But if they're going to use AI, we want them to do it in an iterative, critical, inquiry-based way — usually a back-and-forth process where they're pushing back and thinking carefully about what they're integrating. That's a beautiful way AI can support students' development.

The PAIR process does a nice job of that, but there's also Marc Marino's work with the analog sandwich, and a lot of other examples in the book of teachers doing this well.

Eric Mazur: I actually think peer and self-review is becoming really important, and it's one solution to scaling up, because in a sense you bring everyone into the review and feedback process. I tend to think feedback matters more than scores. What I've been doing, since I have quite a bit of writing in my course, is use calibrated peer review as a feedback cycle: students go through a calibration phase, then look at three anonymous peers, then look at their own work — by which time they've seen a huge spectrum of different pieces — and then hand in a revision, which I provide feedback on.

By the way, did any of you see the tool that works with Google Docs called "Revision History"?

Marc Watkins: I do, yeah. I actually haven't loaded —

Eric Mazur: What do you think of it? I was blown away when I saw it, because you can actually see the process of writing revealed. What are your thoughts?

Annette Vee: It's an amazing thing, I have to say — it can be really good. The thing I think we want to be careful of is not using it for surveillance; it shouldn't be a policing technology, but a way of showing how writing is taught. When I've shown that in my classes, one thing I've done is take a document I've edited and written, and show students what it looks like. A lot of times, if students are writing top to bottom and hit any difficulty, they think something's wrong with them — that's the point where they start to turn to AI. They feel that friction, that difficulty of learning, and may turn to AI in negative ways.

But I'm a fairly expert writer at this point, and if you see my editing, it's notes, half-formed sentences, a paragraph I moved over here, a deletion, me talking to myself about how dumb that paragraph is — there's a lot happening there, and I think it's really important, when we talk about process, for students to see that process in experts and in their peers. Eric, you talk about that with social annotation on Perusall, which I think is helpful — they can see other minds at work, and it helps normalize difficulty and helps them realize they're learning in community.

Derek Bruff: I want to highlight a couple of, not quite recipes, but templates we talk about in the book. The analog sandwich is a version of what I'd call a red-light-then-green-light assignment, where you tell students not to use AI for the first portions — the brainstorming, the ideation, the drafting — but they can use it later in strategic, targeted ways. PAIR is a nice example of that. One advantage is that you're bringing AI use out into the light — you're having a conversation about it with your students, which I think is an important part of AI-aware teaching. And some students are less likely to shortcut with AI early on if they know they'll get a chance to use it later in more thoughtful ways — it helps them resist the temptation early on.

I'd also say that metacognitive reflection is an important part of the processes we develop for students. I have a colleague at UVA, Kira Allison, who teaches a communications course, and she has students do something she calls a "Mission Impossible" persuasion task — do some kind of persuasive task you never thought you could do before, but see if AI can help you do it. What I like about this is that students plan out beforehand how they think they might use AI, then document how they actually use it as they do the work. Even when they present to the class — it's a public-speaking class — part of their presentation is how they used AI, where it helped, and where it didn't. That's really important for students; it brings the AI use out into the light and helps them navigate the struggle between getting help from AI and getting too much help from it, and gives us a chance to speak into that.

Eric Mazur: I see a question here that's been upvoted 14 times, so it's time to address it. Brian Dunst asks: "I work at a community college, and students have begun to overwhelmingly prefer online asynchronous courses. Colleges are also facing the so-called enrollment cliff, which makes them reticent to push back against consumer preferences. That means the pedagogical environment we're facing in this online space is a choice between policing, surveillance, an oppositional relationship with students — versus struggling with generative AI being used contrary to the spirit of learning. Could the panelists speak to this?"

Marc Watkins: Yeah, I can take that. It's deeply unfortunate to see how exposed asynchronous online learning has been to AI. I teach asynchronously online too — we're a residential college at the University of Mississippi, but we're seeing students prefer asynchronous classes even when they're physically here in the dorms, which is wild to think about. I'm in my office right now, 150 yards from the nearest dorm, and I could, in theory, be running a class where I never see my students' faces live. So I think this is something we want to reconsider.

I'm in no way saying we need to do away with online learning — it's an accessibility issue, and also an issue about getting students engaged. I think we can change how we handle these materials. My online discussion board is no longer text-based; it's been video-based for a while now, and it's not just talking-head videos like you see us doing right now. I have students share their screen — when they're working on project two, for example, that project is shared with them, and they go through it just like a social annotation exercise, but asynchronously: they record themselves working through it live, and then interact with each other around it. For readings, Perusall is a wonderful tool we've used for that too. The whole point is: we need to be aware that if we don't have a good relationship with students, if we don't have some consistent face-time in some form, we're really inviting them to automate that process — especially with the new AI agents out there.

One bit of good news — Anna Mills was actually talking about this on LinkedIn — is that Amazon has successfully sued Perplexity AI over its Comet browser, to block it from operating within Amazon's system. It's not in your LMS yet, but it's based on a user's login. That gives me some hope that we might see legal action that could block learning-management systems from being invaded by these agentic browsers, which I find very difficult to see as pedagogically sound.

Eric Mazur: I'd add to that — I looked into this when we pivoted online in 2020 — that there are examples of extremely successful online educational programs that have dealt with these problems of assessment, accountability, and keeping students engaged. I'll mention two examples from almost opposite ends of the spectrum. One is Minerva University, which has a residential program — students live near each other — but the education itself is completely online, because they move from city to city around the globe and instructors don't travel with them. So their social program is live and in-person, but their learning program runs on a platform called Forum, where you can't even turn off your video, and which has a complete assessment tool built into the video platform. It's very impressive. On the other end of the scale is Western Governors University, which has taught online for a very long time and dealt with many of the issues that AI is now suddenly bringing to the forefront.

Annette, in writing, the struggle is often where the thinking happens — you've already alluded to that. I don't know if you heard my podcast with Flower Darby, where she talked about "keeping the fizz" in education. So how do you keep the fizz in writing when a bot can do the heavy lifting?

Annette Vee: This is the question every writing class has to answer right now: why write when AI can do it? Teachers need to engage with that question, but they also need to help students answer it. Ultimately, we're not going to get through this by policing and wagging our fingers at students, telling them what they should be doing. They have to be involved in it somehow, and they have to feel like their education matters.

One of the things we talk about in the book: some of the generic advice about AI is that it's great for brainstorming, that you can get ideas from it. I think that's absolutely the wrong way to go. It's better to use AI on the back end of writing rather than the front end. We have lots of examples of writing teachers doing that — including, if you're resisting AI and can't fit an entire paper into class time, starting students out brainstorming in class, thinking on paper or in focus mode on their laptops. Because when students use AI's ideas, they don't feel the same attachment to them — they see those ideas, think they're potentially better than their own, and start to second-guess their own creativity.

Ethan Mollick calls this anchoring: if you start with an idea from AI, it's very difficult for students to deviate from it and come up with something new. Whereas if they start with their own ideas, you get much more diversity. There's a great article by Piers Gelly in LitHub from a couple of years ago about using ChatGPT to replace himself in grading papers — one of the things he does that I think is such a great teaching move: he'd received a bunch of papers where students had written pretty generic things, and what he realized, going back to that idea of social conversation, is that the students thought their ideas were interesting because they never saw each other's papers as a whole. So he had students read the titles of their papers aloud — and nearly all of them were some version of "Navigating the Digital Age: AI and..." whatever. As they went through all of these titles, there was a kind of collective realization among the students that their ideas were super generic. So I think convincing students that their ideas are interesting — that we're interested in their ideas, that we want to hear from them — really matters.

The other thing I'll add is that I've done a lot of empirical research on why students use AI and why they don't. The top reason they turn away from it is that they value their own originality. I think that's really crucial for us to lean into — you can have original ideas while also working in conjunction with AI. But that's where the fizz is for students, and for writing, that matters.

Eric Mazur: I've recently moved toward ungrading this semester for the first time. I went into it kind of in a panic — first shifted to specifications grading, then, after having several guests on my podcast talk about ungrading, I thought, "Can I do this?" Ideally I'd like to move to a fully narrative evaluation, but for now I've moved to ungrading this semester, and I'm quite happy with the results, because I believe motivation is the key to stopping students from outsourcing their work to AI. If you focus on learning — and all human beings are wired to learn, we're all born constantly asking "why," and in a sense that gets beaten out of us — if you own the learning, you won't want to skip it.

So does your book suggest that the AI-aware teacher also needs to be an assessment-aware teacher? In other words, can we have AI awareness within a high-stakes, hundred-point grading system?

Marc Watkins: That's a wonderful question, and I have some wonderful colleagues, Emily Donahoe and Josh Eyler, who are real experts in alternative grading strategies. I've been using a version of labor-based grading since around 2019 in my classes. It's high-contact, which is how we teach our courses, since they're low-enrollment. For a high-enrollment class, you'd have to think about that differently — and you also have to think about students' perceptions of grades. One interesting thing I've seen is that some students have panic attacks when you tell them you're not using grades, just feedback. I've had to sit down with students in my office and say, "Look, you need to tell me what my grade is, I have to know" — even though they have literally glowing feedback from me all semester. It's like, "You're not an A or a B, you're a person," and the student just says, "Just tell me if I'm an A or a B."

We have to navigate the fact that this has been instilled in students, and give them a process for working through it. But I think the umbrella this all falls under is that there are so many different ways we're going to have to rethink and reevaluate how we teach and how we assess learning — and we have options, which is a wonderful thing to have.

The challenge is that no one's going to tell you that you have to do this — you'll have to come to it on your own, and think about your own capacity, your resources, and the support you have from your department and school. You can tell your provost — I'll probably get some wonderful letters about this — that I very much believe non-tenure-track faculty should also have course releases and things like sabbaticals to help them teach, not just about AI but in general. This is a wonderful opportunity to make that labor visible to upper administration, to get some of those resources.

Eric Mazur: Time for another audience question. Eric Grunwald asks: one impact of AI that seems to be emerging is that for novices, AI often leaves users with an illusion of mastery, whereas for experts, it serves as a real accelerator — hinting at a new kind of divide. Any ideas on how we can show students that danger as a motivator, and help them move into the latter category?

Derek Bruff: I'll jump in, because I was thinking about a question you raised earlier about large classes, and where AI shows up there. A lot of my colleagues in the sciences and mathematics are thinking about how to create a custom AI chatbot that would be a better tutor for students. Most AI chatbots — like when you log onto ChatGPT — aren't designed to help you learn; they're designed to answer your questions and give you a feeling of fluency and ease. That's not actually how learning works. So my answer is: how can we help students see the flaws in that easy, fluent feeling, and shift them toward interacting with AI in ways informed by what we know about how learning actually works — leaning into the desirable difficulties of learning?

Some faculty are sharing prompts with students — here's a prompt that gets your chatbot to act like a student, and your job is to teach it what you know, and it gives you feedback on that. That's a different way of using AI. Other prompts have the AI generate practice questions so you can do retrieval practice — a more learning-science-informed way of interacting with it. As I said, some faculty are trying to build that into tutor bots they're designing, so the bots behave more like tutors and less like the answer machine ChatGPT presents itself as.

I was talking to educational researcher Leon Furze recently, and he said if there's one thing we can have students unlearn, it's that AI isn't Google — even though it looks like Google, a little box that answers all your questions. That's not a helpful way to use AI in most cases. We need other mental models for what these tools can do that actually contribute to learning, rather than giving this fake feeling of mastery. It's like what you talked about years ago, Eric — I show up to your class, you lecture for fifty minutes, it all makes sense while you're saying it, but that doesn't mean it's in my head. AI does that same kind of thing, and we have to find structures to shift students away from it.

Eric Mazur: I'm also thinking more and more that AI devalues artifacts. If you think about assessments, as I heard one of my podcast guests, Janick Guerio, put it, they fall into two buckets: the performance bucket and the artifact bucket. An essay, a homework set, a problem set, even a project — those are all artifacts. A performance would be an oral exam, a musical performance, a theater play, or an oral justification of an artifact. I think AI is pushing us more and more toward that performance bucket, to make things more insightful into what's going on in the student's head. The big question is how we scale that up, since most problems seem to occur in larger classes.

Derek Bruff: I want to jump in, because one of the fun parts of writing this book was that we have a chapter on authenticating student learning and academic integrity issues, and I —

Eric Mazur: That falls into policing, right?

Derek Bruff: Right, right, right.

Eric Mazur: It harms the trust in the relationship between student and teacher — I think I alluded to that earlier.

Derek Bruff: Well, I half-volunteered for the section on oral exams, because I was a little skeptical of them — I'd heard from faculty saying, "We can't trust the artifacts anymore, so now we're going to have to go to oral exams," and I had in my head my own PhD dissertation defense, which was a very high-stress situation. I don't want to inflict that on my students — my advisor was fantastic, but I brought a lot of stress into it myself. But as I researched that section of the book, I found a lot of literature on using oral exams in ways that are very humane, supportive, and aligned with the rest of a course and its learning objectives — actually there to help students learn.

There's research showing that when students prepare for an oral exam, they study differently, in ways that can be very helpful. Often a five- or ten-minute oral interaction — like Marc was saying, even over Zoom — a face-to-face interaction with a student, not letting the entire assessment ride on it, but building it into the whole process, works well. I have a colleague at UVA who started using oral exams as a kind of AI check, teaching an education-policy course where communication about policy is itself a learning objective. So he built oral communication practice throughout the whole semester, rather than tacking it on at the end just to authenticate or surveil students. I think that's a smart approach.

Marc Watkins: Yeah, I highly agree. One thing we've seen pretty consistently since the pandemic is that interpersonal, professional communication has fallen off, and any time we can encourage students to practice it is a skill that will follow them into higher education and their careers — that's exactly what we're hearing from industry, that they want that level of communication from future graduates.

There are AI tools that can help with that too — a lot of our speech faculty are experimenting with a Microsoft tool called Speaking Coach, so you can be thoughtful about integrating that. It's not going to be a cure-all for making your classes "AI-proof" — if you're using it purely to secure assessments, it'll be very labor-intensive. In my own discipline of writing, there isn't much background yet on how to use these tools pedagogically for oral assessment, so you have to look at how it's done in other fields, like Derek has. That's what we try to do in the book — give people options to consider.

Eric Mazur: Derek, I was going to react to what you said — I'm really eager to read that chapter on oral exams. I guess I'll have to wait until July 6th for the Engage event. I'll take one more question from the Q&A box. This one's from Jennifer Carter, who asks: "A few of my friends outside academia have shared their distaste at being asked to use AI, like Claude, for coding and drafting web layouts in the workplace. How can we, as mentors, help prepare our students to use AI reflectively and effectively in the workplace and in their everyday lives?" Who wants to take that one?

Marc Watkins: I think part of the process here is that we're all exposed to this together — Derek, Annette, and myself. Annette has a computational background, she's worked in this space, but Derek and I didn't grow up with generative AI as part of our lives. Same with a lot of our faculty, and probably a lot of people in this webinar — you're not trained in this. So how do we respond to it? Do we even engage with it, and start preparing students for it? Those are questions we all have to answer for ourselves.

For me, it comes back to the question of value — personal value for students, being honest with themselves about their motivations for using these tools or resisting them — and having a conversation about that, one that crosses disciplines. It's not just about writing; these tools are completely multimodal, and have been for three or four years now. You can put the app on your phone, talk to it, activate the camera to look at your surroundings, and have it assess or give you live feedback. Part of being aware means understanding how embedded these tools already are, and understanding the use cases.

And really, it just means having open communication and dialogue with your students about that. That's the big thing we all want: for students to talk openly with us about whether they're using these tools. I need to know if my students are using AI to read or study for homework, because I need to know whether it's an effective use case — and I won't know that unless they're open with me. There's no AI detector that can tell you whether a student is reading or not. That's the big thing.

Annette Vee: I just want to underscore what Marc's saying about talking to students — this is something we emphasize throughout the book. We see it in surveys and in other research we've done: students desperately want to talk to us about AI. They're often navigating it alone, getting constantly bombarded with ads for Gemini from TikTokers and influencers — Marc actually named this dynamic a couple of years ago, really influentially. It's important for us to understand that context. I think one of the reasons we might want to engage with AI in the classroom is that it gives students a space to practice ethical use of it, since almost all of them are already using it. So introducing it in class isn't introducing something new — it's giving them a space to engage with it with some scaffolding, which can be really valuable.

There are a lot of ways to run those conversations. If you want them anonymous, you can use a Google Doc, or Mentimeter, or QR codes students can answer through anonymously. You can pass around paper in class, or discuss it openly. Carly Schnitzler at Johns Hopkins, and colleagues at Boston College, have examples using specific case scenarios — "Is it okay to use AI to do this thing?" — and have students work through a list of them together. It turns out students have really varied ideas about what's okay and what isn't.

We all know the examples that get shared publicly — the "ChatGPT lawyer," or using AI for a condolence message — clearly bad uses of AI. But the reason those keep getting publicly shamed is that we're all still working this out together; we don't have a widely shared ethical framework for it yet. As we prepare students to move out into the world, we need them to develop that kind of internal ethical orientation — one that's well thought out, not just a default. Giving them space to work that out in a classroom, in community, is really crucial.

Eric Mazur: We're nearing the end of our hour together, and I know people appreciate practical ideas. So for the educators listening at home who are feeling overwhelmed — and I know a lot of people are expressing anxiety here — what is one small move they can make in their next class or course to be more AI-aware?

Marc Watkins: I actually designed a chapter around in-class activities built around time, more than anything else, because one piece of feedback we've heard from faculty all around the country is that this is so time-consuming. A lot of the activities in that chapter are built around a ten-to-fifteen-minute window if you're in person — and they can also be translated into online activities; Derek did a wonderful job on our online-activities chapter as well.

You can take an activity like John Epilato's "What Uses More," which looks at AI's environmental impact — bring it into the classroom, pair it with a critical reading on energy usage in AI, and get students thinking about their own use of generative AI. Maybe pair it with something like how many hours of Netflix they watch, to compare "what uses more" and start these critical conversations. That's the secret sauce: something active, involving a bit of critical reading or discussion and metacognitive reflection — written or discussed in class. Taking a page from Jim Lang, those are the kinds of small, teachable moments you can start incorporating into your teaching.

Derek Bruff: This operates on a couple of levels for me. In our AI-aware framework, there's: what are your learning goals, how are your students using AI, and how do they feel about it — and there can be small ways to open up that conversation. I'm a big fan of some form of transparency statement, where you ask students to disclose and reflect on their AI use on just one assignment — open that box just a little, that way.

I also think part of it is learning about what AI actually does and how it works. I'm a big fan of taking fifteen minutes to interact thoughtfully with a chatbot. And I have to share one concept quickly, because it gets at this mental model of what AI is — the rubber duck effect. It's a notion from computer programming: if you're working through a program and can't get it to work, you pick up an inanimate object on your desk, like a rubber duck, and talk your problem out loud to it. It doesn't respond — it just quacks — but the act of articulating your problem helps you see it differently and maybe find a solution. If we see AI as an answer machine, that's not productive. But if we see it as a rubber duck we can interact with to prompt our own thinking, that's a much more helpful approach. One way to get started is to try that in your own work — I've got a lesson plan coming up next week, so let me talk to a chatbot about some ideas to make it more active, and see how that conversation goes. That helps you become more familiar with how these systems work in your own practice, which prepares you to work with your students too.

Eric Mazur: That's a great use of AI — and it's exactly what students in my course are doing; I've been encouraging them to articulate their thoughts that way too. Unlike the rubber duck, which only quacks back.

There's something really interesting that came up in the chat from June Griffin — not in the Q&A. It actually came up when I brought it up during the student panel yesterday too. I'm hoping you might comment on faculty using AI to assess and/or give feedback. One concern is that this looks very different across disciplines — humanities faculty will likely continue to make student-focused assessment decisions that take more time and labor than science and social-science faculty.

Annette Vee: One thing we know from students is that they don't want us to use AI for feedback — in terms of judgment, they want to hear from us as faculty. I do think there are ways to coach students into using AI feedback for themselves, so that it's done well and transparently.

One note, since we talked about blue books and students' handwriting: one way you can use AI to save labor in humanities classes — for people who say, "I can't do that because I can't read my students' handwriting" — is that if you're honest and transparent with your students, you can take pictures of their writing, or have them take the pictures, and have AI transcribe it. That's one of the simplest ways to integrate AI into assessment in a way that saves labor in writing-based fields.

Eric Mazur: Yeah.

Derek Bruff: I just want to hear Marc respond to this, because I think he thinks a lot about this.

Marc Watkins: Yeah, I'm very opposed to using AI to grade, because that's a critical judgment we're trying to communicate to students that should come from us. I think there's a real risk of devaluing the relationship between professor and student if we let AI do that. I'm perfectly happy with projects like PAIR, which I mentioned, where it's done transparently and above board. But my real fear is that this technology is being sold to faculty in both higher ed and K-12 as a labor-saving tool — and one of the most labor-intensive things we do is actually assess student learning. Having AI do that, not just through text but even oral exams — it can look at you live, the way I'm looking at you on this camera right now, and make judgments using a system that's opaque, a black box you can't ask how it arrived at its decision.

That might change someday, but for right now, it's not appropriate for me. I'm teaching 30 students right now; when I was teaching 125, that raises different questions you'll have to ask yourself. But to me, these are the conversations departments need to be having openly: if you're using AI to grade, how do your students feel about that? How do I feel about it as a colleague, knowing you're doing that when I don't want to do it at all? Those are some of the big conversations I hope this book helps people start having.

Eric Mazur: Anything to add, very briefly, since we're running out of time? All right — let's end with a quick round. I'd like each of the three panelists to complete this sentence: "Teaching in an AI-aware way means..."

Annette Vee: ...listening to your students' engagement with AI and calibrating your pedagogy appropriately.

Marc Watkins: ...acknowledging that AI-aware teaching is not a solution for AI to solve.

Derek Bruff: ...helping our students learn how to learn, with or without AI.

Eric Mazur: Annette, Marc, Derek — thank you for this incredible preview of your upcoming book, The Norton Guide to AI-Aware Teaching. I can't wait to see it in print this summer. And a final reminder to our live and virtual audience: don't miss the Perusall Engage event with these three authors — coming up soon, on July 6th. You'll be able to read and discuss the book directly with them on the Perusall platform, and you can register at perusall.com/engage. Thank you to everyone at the Perusall Exchange for joining us for this live recording. On behalf of all our listeners, thank you, Annette, Derek, and Marc.

Annette Vee: Thank you, Eric. Thanks so much.

Marc Watkins: Thank you.

Derek Bruff: Yeah, thanks for some great questions and discussion.

Annette Vee: Yeah, absolutely.

Eric Mazur: To find more episodes of Social Learning Amplified, go to perusall.com/sociallearningamplified, and please subscribe to find out about upcoming episodes. I hope to welcome you back on a future episode.