Studying AI Itself
Many courses study AI directly or examine its relationship to their field. Literature students might analyze AI-generated text alongside other texts, asking how its sentences work and what they suggest about how AI “thinks.” Courses on politics, social movements, media, culture, gender, and identity examine domains of human life that generative AI is changing. AI’s own history, the figures involved in its development, and the ideologies behind its construction and promotion also invite study across disciplines. In these contexts, we work with faculty to design experiences that help students think critically about AI and look beneath the hood of familiar tools: how they are constructed, what they reveal or obscure, and how they represent the world in language and images. Students can investigate both how models reproduce biases and how companies try to reduce those biases or introduce guardrails, sometimes in counterproductive ways. These activities give students ways to investigate AI using the questions and methods of their fields.
Revealing hidden prompts in image generators (EMR 162)
In EMR 162: Interdisciplinary Perspectives on Race and AI, taught by Leslie J. Fernandez, students examined how AI image generators represent race, gender, ethnicity, and class. Many AI companies rewrite a user's prompt before their image model sees it. The Learning Lab wrote code that showed students these rewritten prompts, which revealed the choices the companies had made about diversity and representation. Students brought their own prompts to the workshop, compared the images they hoped for with what the tools returned, and ran a loop in which a vision model described an image the system had just generated, so they could see what the model treated as the image's salient features. Students who focused on text generation ran into chatbots that refused to respond because of built-in restrictions. Part of the assignment was to test and iterate on prompts to see how one might get around guardrails or system prompts that are meant to mitigate the generation of bias or false information, and to draw students' attention to the potential limitations of those safeguards. The workshop ended in a seminar-style discussion of what these modes of representation mean socially and politically. The underlying move is to treat the AI tool as a primary source, open to the same analysis a course already applies to its texts and images.
To design an exercise that puts an AI tool itself under analysis, contact the Bok Center.
Surfacing bias before a final project (PSY 1409)
Students in Haley Keglovits's PSY 1409: The Psychology of Large Language Models came to the Learning Lab for a workshop on bias in AI systems before starting their final projects. They completed short hands-on exercises that surfaced linguistic, cultural, and technical biases in generative AI tools, then toured the studio tools and workflows available for media-rich projects designed to teach others what they had learned in the course. Students went on to produce web tools, short videos, podcasts, a song cover, a short story, and a small interactive app; none of them had a coding background.
To prepare students with no coding background to build with AI, contact the Bok Center.
Love songs in a poet's voice (COMPLIT 126X)
In Professor Moira Weigel's COMPLIT 126X / HUM 5: Literature and/as Artificial Intelligence, students generate new love songs in the voice of a poet they have studied closely, then defend the results in a short artist's statement. A single request to a chatbot produces a weak imitation, so students work in steps: they gather more poems by their poet, name the recurring images and habits of syntax that make the voice recognizable, and only then instruct the tool, judging each draft against the features they named. The Bok Center ran a workshop with the class on breaking the task into steps like these, and built starter materials so students with no technical background begin from a working setup. Writing good instructions requires close reading first, and the finished song, with its artist's statement, shows how well the student knows the poet. The sequence asks for close knowledge of a voice or style; which voice is up to the course.
To design an assignment in this shape for your course, contact the Bok Center.
Studying AI's history by building with it (ENGLISH 298AI)
In Professor Martin Puchner's ENGLISH 298AI: AI Humanities Lab, students pair a cultural history of AI and of dialogue with hands-on building, from chatbot characters to working websites and tools, with no coding background assumed. The Learning Lab runs two workshops for the course, one on chatbot characters and one on building toward the final project. Students also interact with systems from different stages of chatbot development: ELIZA, built decades before large language models, then GPT-2, early ChatGPT models, and contemporary systems. Each generation answers differently and fails differently, so students can examine the technology's development directly.
To design a course that studies AI while building with it, contact the Bok Center.