Designing Worlds from Words: Casandra Silva Sibilin on AI, Image Generation, and Deeper Reading
How a philosophy lecturer uses AI image prompts to unlock student engagement and critical thinking in Western Civilization
Casandra Silva Sibilin is a Lecturer in Philosophy at York College, City University of New York (CUNY). In this interview, Sibilin discusses how AI is being used in Western Civilization (CLDV 210).
Source note: This is an edited interview adapted from a narrated video submitted to OpenAI. This was shared last fall and is even more relevant now since ChatGPT Images 2.0 launched.
Intro
What if students could step into the shoes of virtual reality designers, using AI to visualize the worlds described in classic texts? Casandra Silva Sibilin, Lecturer in Philosophy at York College, City University of New York, has been experimenting with just that. In her Western Civilization course, Silva Sibilin invites students to generate images from primary source passages using AI, transforming dense readings into creative, collaborative, and critical exercises. In this conversation, she shares the theory, design, and surprising outcomes of this assignment—and why she believes it’s a model for deeper engagement and AI literacy across disciplines.
The Interview
Q: Casandra, you’ve described a writing assignment where students use AI to generate images from primary texts. What inspired you to try this approach?
Sibilin: The idea came from the buzz around the release of Image 4.0*. I saw a lot of educators using AI to illustrate their own lessons or course modules, but I wanted to do something different—put students in the driver’s seat. Instead of just showing them images, I wanted them to become the builders, the interpreters, the ones responsible for shaping what the AI produced. That way, they’d have more agency, and it would push them to engage more deeply with the readings.
Q: What challenges in your Western Civilization course were you hoping this would address?
Sibilin: Courses like mine—surveying philosophy, literature, science, and religion from ancient to modern times—rely heavily on primary sources. These texts can feel long, dense, and remote to students. The big challenge is getting them motivated to read closely and persist, even when the material feels distant from their own experiences. I also wanted an assignment that would help them develop their own voice in writing, while encouraging critical and responsible AI use.
Q: How does the assignment actually work, step by step?
Sibilin: The core activity is simple but powerful: each student or team picks a passage from a primary text, then develops a prompt to generate an image with AI based on that passage. For instance, if they’re working with Dante’s Inferno, they might start with a straightforward prompt, but quickly realize the image doesn’t quite capture what they imagined. That leads them back to the text—to rethink, revise, and iterate on the prompt. It’s a cycle: read, interpret, prompt, reflect, and repeat. This process really draws them into the material.
Q: Why did you decide to have students work in teams?
Sibilin: While the assignment could be done individually, I found that teams made it more fun and intellectually stimulating. In my online synchronous class, students worked in breakout rooms, discussing which passages to choose and how to craft their prompts. I noticed some groups started looking for common themes across different texts, which was a delightful surprise—and exactly the kind of cross-textual thinking I want to encourage.
Q: What did students discover as they worked with the AI image tools?
Sibilin: Many realized how much the prompt matters. They saw that small changes in wording could dramatically alter the image. This forced them to think about what details, moods, or symbols were truly central to the passage. Students reflected on elements like personality, mood, and symbolism—things they might not have explored if they were just writing about the text in a traditional way. It also sparked conversations about how AI-generated images can differ from human interpretations, and why that matters.
Q: How did you structure the reflection and assessment aspect?
Sibilin: After generating the images, students shared them on a class Padlet. The writing assignment asked them to reflect on the whole process: why they chose their passage, how they collaborated on prompts, and what they learned from the iterations. A key part was considering counter-arguments—how might the AI’s illustration differ from a human’s experience of the text? This kept the focus on the human element and on the value of discussion and teamwork.
Q: Were there any unexpected moments or outcomes?
Sibilin: Absolutely. One unplanned activity emerged when students posted images without the passages, and we turned it into a guessing game—trying to match images to texts. It became a fun way to review material. More broadly, I saw students engaging with readings and each other in ways that were deeper and more creative than in previous semesters. Their reflections showed real growth in both textual analysis and AI literacy.
Students collaborate on AI image prompts, iterating and discussing interpretations in breakout rooms.
What Stands Out
Core idea: Students gain agency and deepen their engagement by becoming the creators of AI-generated images, using their interpretations of primary texts to drive the process.
Classroom design: Collaborative teamwork and iterative prompting encourage critical thinking, close reading, and lively discussion, making dense material more accessible and relevant.
Student impact: Learners report new insights into texts and AI tools, discovering how prompts shape meaning and how technology can both illuminate and distort human understanding.
Transferable lesson: This approach can be adapted to any discipline with reading material, offering a model for integrating AI in ways that foster creativity, collaboration, and critical reflection.
Bio
Casandra Silva Sibilin is a Lecturer in Philosophy at York College, City University of New York (CUNY). She co-founded “Don’t AI Alone,” a grassroots group bringing together faculty, staff, and students into inclusive, interdisciplinary dialogue across 26 CUNY campuses. She has developed open educational resources to empower educators and students to build their own AI tools, including “Student-Made Bots: An Educator’s Toolkit.” Beyond CUNY, she leads the Custom Bots/GPTs subgroup within the global “AI in Education” community. She is currently writing a book on AI and philosophy under contract with MIT Press.
What strikes me most about Sibilin's design is the inversion of agency… students aren't consumers of AI illustrations but producers wrestling with their own interpretations through the medium of the prompt. The "read, interpret, prompt, reflect, repeat" loop she describes is essentially a design process applied to hermeneutics. The prompt becomes a specification document; the AI's output, a prototype the student must evaluate against their reading.
I've been working with a similar premise in my courses, asking students to examine rather than accept the claims of complex texts. Sibilin's approach gives me a tool I hadn't considered: the gap between what a student writes in a prompt and what the AI returns is itself a diagnostic. If the image misses, where exactly did the language fail to carry the passage? That gap is where close reading actually lives.
The piece I'd want to push on is the question of what AI illustration leaves out. There's a contemplative tradition of Ibn ʿArabī's khayāl (the imaginal faculty) that treats imagination as a way of knowing rather than merely picturing. When the AI does the rendering, do we surface that faculty or substitute for it? My read of her account is that the prompt-iteration cycle actually exercises it more than passive reading would, because the student has to commit to a specific imagining and watch it succeed or fail. But it's worth holding the question open as the practice spreads, since imagination is a conscious quality that does not translate cleanly into code and cannot be fully delegated to the machine doing the rendering on our behalf.
The Padlet-as-guessing-game emergence is the kind of thing you can't plan but you can design conditions for. That, to me, is the deeper transferable lesson… Sibilin built an assignment with enough slack that the class could surprise itself.
Casandra, this is an excellent operational blueprint for transforming a static text into a dynamic feedback loop. As someone with a background in physics and engineering, your classroom design maps perfectly to a high-fidelity closed-loop controller.
The iterative cycle you described—read, interpret, prompt, reflect, repeat—is essentially a debugging process for comprehension. When the initial AI output fails to match the student's internal conceptual layout, it forces a system error. To fix it, they can't just guess; they have to pull higher-resolution data directly from the primary source text.
Your unplanned "guessing game" is a brilliant example of reverse engineering. By stripping away the text and forcing the class to decode the visual output back into its structural source material, you are training deep pattern recognition.
It moves students from passive data ingestion into active, structural systems analysis. Thank you for a grounded reminder that AI shouldn't be a shortcut to bypass thinking, but a mirror that forces us to look closer at the original code of human thought.