Four professors share their experiences and strategies for teaching critically using AI tools.

Amirpouyan Shiva: Using Generative AI to Teach Ethical Reasoning in Ethnographic Methods

I designed a group exercise for my undergraduate methods course (linked here) in which students use generative AI (ChatGPT without login, with privacy safeguards) as a fallible co-author to produce a fieldwork ethics case study. Groups generate an initial case, critique it for generic content and unexamined assumptions about fieldwork, refine the prompt and manually edit the result, and then analyze the case against TCPS 2 (Canada’s national research ethics framework) and exchange cases with another group for feedback. They close with an attribution statement documenting their AI use, plus a whole-class discussion of patterns across AI outputs. A no-AI alternative meets the same core learning outcomes for students who opt out. Several factors shaped how I designed this activity.

The first was a recurring problem: Students were using AI to bypass analytical work, such as asking for definitions, outlines, and summaries, rather than thinking through the material themselves. Asking them to create cases with AI rather than analyze preexisting ones made that shortcut unavailable. Case-creation cannot be fully delegated to AI: Judging whether a generated case is worth reasoning with requires the disciplinary knowledge the exercise is meant to activate.

The second factor was the specific character of ethics training. Because ethical reasoning is situational, cases are not convenient illustrations of principles; they are the medium through which principles become meaningful. Designing cases forces students to make explicit what they often leave implicit: what counts as an ethical problem, whose perspectives matter, how a dilemma is structured. For example, consider a case a group might build: An ethnographer studying working conditions of undocumented immigrants in a garment factory takes a job there under a false identity. The workers know about the research; the owner does not. Designing this case means deciding whether the ethical problem is the deception of the owner or the risk transferred to workers if the cover fails. And each choice produces a different case. It also means deciding whether the workers speak with one voice or are split about whether they want the researcher inside. A case handed to students arrives with those decisions already made, and made invisibly: The framing reads as a description of the situation rather than as a construction, so students argue vigorously inside a frame they never had occasion to notice. AI’s tendency to produce generic scenarios also proved useful here: Identifying what frames were at work and what was wrong required students to articulate what a better case would look like.

The third factor was a concern about how AI represents anthropology. The patterns students identified in AI-generated cases (“remote” field sites, always outsider researchers, coherent communities, discrete dilemmas) are not neutral. They reflect how the discipline has historically imagined itself, including its blind spots. Making those patterns visible became an entry point into questions about representation, positionality, and knowledge production already central to the course.

The activity, in this sense, does not treat AI as a tool to be used correctly. It treats AI as an object worth scrutinizing, and students’ own disciplinary formation as the resource that makes scrutiny possible. Student responses to the exercise varied. Several described themselves as “AI vegans” but tried the exercise; a smaller group opted out, mostly for environmental reasons, and completed the no-AI alternative. The alternative mattered not only as an accommodation but as a signal that refusal was a legitimate critical position rather than a failure to participate. Several students said they had read about AI bias but had never seen it operate on material they knew well. Watching a generated case flatten a recognizable fieldwork setting made the problem concrete and helped them articulate what their disciplinary training enabled them to do: resist the first plausible output and notice assumptions that a user without that grounding might reproduce. I cannot say whether this changed their subsequent AI use, but it changed the terms in which several discussed it. Several concluded that critical and historical training in the humanities and social sciences becomes more valuable, not less, as these tools spread.

For instructors considering this activity, my main suggestion is to design the opt-out first, and to make the critique rather than the generation the graded work: In my class, some of the sharpest analyses of its output came from students who were most skeptical of AI, and the exercise should reward that rather than treat enthusiasm as engagement. It also helps to keep expectations low about what the AI will produce. The thin, generic first case is not a failure of prompting but the material the exercise runs on. What struck me most in watching students work through the exercise was not how much they knew about AI, but how quickly they were able to draw on their disciplinary knowledge to question and push back against it.

Sita Venkateswar: Teaching Public Anthropology—Taking Anthropology to the World from Aotearoa/New Zealand

Linked here is a sample of AI guidelines. These are developed by the institution and require completion by the student. The onus is on them to maintain detailed documentation of the prompts, used sequentially, and the version history of the drafts and changes applied, while clearly demarcating the provenance of everything on the page. That is a lot of work for the student to take on, on top of completing the assignment! None of this is perfect or foolproof, but a snapshot of an institutional response in progress that is adapting with the technology as it evolves. The policy is principles-based, not tool-based. Thus, the same expectations still apply, including academic integrity, professional responsibility, and legal compliance.

The focus of the 300-level course Taking Anthropology to the World is for students to learn and practice positioning themselves as public anthropologists, writing to the public about current issues, local and global. Students cultivate a distinct anthropological voice that is recognizably theirs, grounded in a particular place, Aotearoa/New Zealand, to craft an anthropological sensibility responsive to Te Tiriti o Waitangi (the treaty signed between Māori chiefs and the Crown, inviting British settlement and shared futures, without relinquishing sovereignty), to address contemporary topical issues on social media. Long-form on Substack or visual and shorter on Instagram, and earlier on Twitter/X, students develop six posts over the semester, working iteratively with me on a class discussion forum in the Moodle-based online learning environment. It is an intensive process of developing, fine-tuning, and responding to my feedback, writing and rewriting over the semester. But the rewards are visible in the improvements in their narratives partway through the semester, requiring minimal rewriting or further feedback from me.

There are no marks assigned to these posts; the public iterative process enables the emergence of the student’s authentic voice and individual mode of writing, both in the asynchronous Discussion Forum and during the in-class live workshops, to speak about the posts in process—which appear to counter the affordances offered by AI! What is assessed instead are the final reflections that students submit of their learnings, challenges, and perspectives derived from their practice as public anthropologists on the class accounts on social media, often demonstrating profound anthropological reflexivity. On social media, the students remain anonymous and protected as they craft and draft their public voice, their individual perspectives available to the discerning viewer, from the series of posts on those social media platforms.

The students discover their distinct public anthropology voice, accompanied by a quiet pride in their portfolio of posts. The iterative process of working closely and intensively with every student, and the intellectual maturity of the students in this final capstone course in the anthropology major, appear to avert the possibility of AI use for the time being. But it is demanding and only sustainable in a relatively small class of approximately fifteen students, much like creative writing courses.

Saad Quasem: Keywords

The rupture of the planet warrants the study of anthropology and climate change, the import of which lies in challenging categories, hierarchies, and chronologies that have brought us panicking into the present. Generative AI has blurred my relationship to students’ writing. Once I knew them through their essays; now I often mark machines. Dragging everyone back to pen and paper risks alienating students who never lived without digital tools. The paradox is how to keep demanding rigorous, inventive, and creative engagement with climate change while acknowledging the animacy of AI in the social life of the classroom. I turn to Raymond Williams’s classic 1976 text Keywords, which insists that we attend to how loaded terms change meaning over time, carrying deep histories and embedded power. Borrowing this approach, I put together a “keywords” assessment in which students either coin terms or draw them from class material to think anthropologically about climate change. Each entry becomes part of a small “keywords lexicon” that draws directly on lectures, images, models, theories, concepts from class, ethnographic case studies, news fragments, and emergent ideas from our discussions, and must be grounded in set course readings rather than external sources. AI may be used to generate ideas or tidy language, but the core task is a form of anthropological “conceptual stitching” identifying discourse around a keyword, tracing patterns of similarity and difference across specific readings, lectures, and case studies, and tying those patterns back to the events that unfold in the liminal space of the classroom, work that generative models, trained on decontextualized text and absent from the classroom, cannot easily reproduce. I learned this assignment from my mentor, Professor Jim Igoe, who had us build keywords across the term in short, regular submissions; I keep that slow accretion of ideas but turn it into a single end-of-term assignment, using AI only to refine. We learn to think densely but present precisely the socio-technical and political-economic systems that perpetuate climate breakdown and enable contemporary AI, drawing out their connections and making them available for critique.

Tiatemsu Longkumer: A Case for AI Use in the Classroom

My approach is to allow students to use AI from the beginning to the end of an assignment. Rather than treating AI as something to ban, limit, or hide, I design the assignment around full AI use. Many students appreciate having clear permission to use AI. It removes the uncertainty and secrecy that often surrounds AI use in assessment. They know that they are not being punished merely for using the technology; instead, they are being evaluated on how responsibly and intelligently they use it.

Students complete their work through Google AI Studio (since it is free), and the goal is not simply to produce a polished answer, but to show how they engage AI as part of their learning process. My rubric (linked here) focuses less on final output alone and more on how students interpret AI responses, ask strategic questions, sustain the conversation, remain aligned with the assignment objectives, and critically evaluate what the model produces.

In my experience, students often become more thoughtful because they must guide and evaluate AI rather than passively receive information. The main challenge is that this method takes time. Undergraduate students need training, and the learning curve can be long, but I believe they learn better this way than through more traditional approaches. My experience of grading these assignments (see a sample here) has been quite different from grading conventional papers. I am not only evaluating the final written answer; I am examining the intellectual process through which the student arrived at that answer. This includes how the student framed the initial problem, what instructions they gave the AI, how they responded when the AI produced something inaccurate or superficial, whether they asked meaningful follow-up questions, and how they revised or rejected the model’s suggestions.

In a conventional assignment, much of the student’s reasoning remains invisible. I usually see only the final product. In this approach, the interaction history gives me access to parts of the learning process that would otherwise remain hidden. I can see, for example, whether a student recognized that an AI-generated explanation was too general, whether they asked for evidence, whether they introduced concepts from the module, and whether they were able to redirect the conversation toward the actual learning objectives.

The grading process therefore becomes less about asking, “Did the student write every sentence independently?” and more about asking, “Did the student exercise intellectual judgment throughout the process?” Two students may submit similarly polished final answers, but their interaction records can reveal very different levels of understanding. One student may simply accept the first response generated by the model, while another may question its assumptions, compare perspectives, identify missing information, and substantially reshape the final argument. The rubric helps me distinguish between these forms of engagement.

At the same time, this approach makes grading more demanding. AI conversations can be long and non-linear. The teacher must examine not only the final answer but also the quality of the interaction. Students also need clear guidance about what parts of the conversation are relevant and how they should document their decisions. I would argue that this form of assessment addresses many of the central dimensions of academic learning, including critical thinking, metacognition, argumentation, information verification, and reflective judgment. 

Students gradually learn that they must decide whether a response is relevant, whether it reflects the concepts taught in the module, whether it contains unsupported claims, and whether another interpretation is possible. They also learn that a good academic question often develops through several stages. The first question may be vague, but through interaction, the student refines the problem, introduces context, requests clarification, challenges assumptions, and identifies what is still missing.

This iterative process can develop metacognition. Students are not only thinking about the topic; they are thinking about how they are approaching the topic. They have to explain why they accepted one AI suggestion but rejected another, why they changed their question, and how the interaction affected their final position. In this sense, the assignment encourages students to become more aware of their own reasoning.

My broader argument is therefore not that AI itself improves learning. Rather, when assessment is redesigned around interpretation, verification, reflection, and intellectual judgment, AI can become an environment through which these capacities are developed.