This guest contribution was written by Dr. Stephanie Santschi, who is a Postdoctoral Fellow at the University of Zurich’s Department of Art History, where Santschi teaches and researches Japanese visual culture, especially ukiyo-e prints, using digital approaches.
Since the publication of this article AI_D has been launched. It can be accessed here.
When our team at the Chair of East Asian Art History confronted an MA student’s shift in writing style, they admitted they used “computational language support” because they feared their German writing would not translate into the course’s English requirements. We faced a new pedagogical challenge: the student honestly sought solutions to navigate linguistic complexity in their multilingual academic environment, between the Swiss, East Asian and English languages at our Chair.
This essay argues that transparent guidelines can position AI as a learning tool rather than an academic crutch. Use permissions depend less on technology than on transparency about its impact on students’ learning trajectories, and should be defined as part of constructively aligned teaching: “teachers need to be clear about what they want their students to learn,” and “students need to be placed in situations that are judged likely to elicit the required learnings” (Biggs 1996, 360). I propose “AI_d,” named for its integration of artificial intelligence with didactic considerations, as a framework that helps lecturers define AI usage parameters for constructively aligned teaching. Currently developing at the University of Zurich, it has evolved from addressing individual student uncertainty into supporting lecturers in defining and communicating pedagogically sustainable AI policies.
The Problem: Language, Scripts, and Student Uncertainty
In the humanities, truth is not a binary. Even if translated metadata were available, students must navigate multiple scripts, historical contexts, and multiple romanization systems (Bellemont 2022). For translation tasks, students must verify AI character-recognition, transcription, and subsequent translation. Consider the complexity students face when analysing artworks like Josetsu 如拙 (active 1405–1496)’s Muromachi-period Hyō-nen-zu 瓢鮎図 (Catching a Catfish with a Gourd). To fully understand the significance of this monochrome ink painting, they must decipher and translate its Japanese poems, and embed findings in Zen iconography (Yamamoto 1997). Equally, students may encounter linguistic particularities such as kanbun kundoku (Japanese reading order of classical Chinese, see Crawcour 1965) in Katsushika Hokusai 葛飾 北斎 (1760–1849)’s illustrated book used as the cover image to this article Tōshisen ehon gogon ritsu 唐詩選画本五言律 (Illustrated “Anthology of Tang-Dynasty Poems” in Five-Syllable Lines), 1836, whose translation should point to both the original verse and interpreted meaning (Yonemura 2014).
With students led to believe they can delegate some of these conceptual tasks to AI, lecturers need to teach them to navigate these processes appropriately. Retaining the original’s ambiguity is not a binary process suited for delegation to an AI. The processes themselves might be underdeveloped, as even LLMs struggle with transforming kanbun kundoku input to comprehensible transcriptions of Chinese poetry with Japanese syntax (Wang et al. 2024). And these questions of understanding stand even before documenting their insights in academic English or German writing.
Developing a Solution: AI_d as an Adaptive Didactic Framework
My spring 2025 BA seminar included different learning scenarios aimed at developing students’ critical skills in evidence-based argumentation, the core learning goal of our discipline. Assessments included image description, object analysis and research presentation. In these works, we noticed that instead of developing independent analytical arguments, students remained stuck in description, struggling to move from what they see to how visual evidence supports art-historical claims. As many gravitated towards generative AI to write analyses, which did not resolve their shortcoming at communicating insights effectively, I developed a guide on reflective AI use. But upon sharing it with fellow lecturers I realized that providing equally apprehensive faculty with tools that support defining AI usage permissions might even be more important. This led to the first draft of AI_d.
AI_d’s core principle is that usage policies should react to tasks, not tools. Assignments are constructive if they can be reached through a series of steps which exercise students’ various research skills. To explain the tasks and associated risks transparently, AI_d provides lecturers with a mentored workflow which prompts them to define the tasks in their course, the intended learning outcome of these tasks, as well as their recommendation on AI use for completing them. When developing AI_d, I started by listing research tasks in East Asian art history. I then added a reflection on what appropriate AI tool use could look like, including the risks of AI use, and workaround possibilities. A permission spectrum that reflects research complexities ranges from “prohibited” for tasks requiring original thinking, to “encouraged” for computational tasks requiring human interpretation. For instance, when students analyse visual motifs across hundreds of ukiyo-e prints, AI pattern detection is encouraged, but students must contextualise their findings within Edo period cultural practices, which is a human interpretive task.
Lecturers stand at the beginning of constructively aligned teaching. Therefore, integrating AI_d with existing teaching infrastructure at the University of Zurich was important to me from the start, regardless of which form it would ultimately develop. I am grateful for the input from the UZH Teaching Fund (ULF) and technical partners. Our conversations led to developing and testing two complementary approaches: an integration into the university’s online learning and teaching platform, where lecturers can define course-specific AI parameters that students access during assignment submission, and as AI chatbot, which offers personalized consultation for complex teaching scenarios.
Lessons Learned: Procedural Thinking and Learning Alignment
Developing AI_d in conversation with students and lecturers, I argue that AI tools, when properly integrated, do not fundamentally change humanities research processes. The essential scholarly skills of critical analysis, source evaluation, argumentation, and cultural interpretation, remain the guiding goals of our teaching processes. Revealing the links in this chain of “procedural thinking” breaks it down into achievable steps. As a result students see AI as part of a larger methodological toolkit. Such transparency, which AI_d fosters, is most helpful: students receive clear guidance on why only certain uses support learning goals. In our course, this increased their proactive consultation about appropriate AI application and led to greater accuracy in declaring and reflecting on their tool use.
Looking Forward: Implementation and Expansion
The AI_d framework, supported by a micro-innovation grant from ULF, is currently under revision and pending refinement. Initial feedback from colleagues emphasized its practical value in connecting AI guidance to specific learning goals, which is what educational theorists call “constructive alignment” (Biggs 1996). Testing will begin in fall 2025, focusing on systematic feedback collection from both faculty and students. The goal is developing scalable approaches that other faculties and institutions facing similar challenges can adapt.
Conclusion: Critical Tool Use in Digital Humanities
AI_d demonstrates that the future of humanities education lies in teaching critical tool use through constructive alignment, focusing on tasks and learning objectives rather than (AI) tools. This empowers lecturers who may not be technology experts but understand their discipline’s learning goals: AI literacy requires the same critical thinking skills we have always taught in Humanities disciplines, namely, source evaluation, bias recognition, methodological transparency, and intellectual honesty.
Student M.H.’s response to AI_d, via email to the author “[f]rom my perspective, the biggest worry is often to accidentally use a tool, not knowing it was not allowed, so a clear overview of what is prohibited would be very appreciated” (March 14, 2025) has been encouraging and validates our approach. When we focus on tasks rather than tools, we enable truly constructive alignment and ensure every assignment supports learning objectives regardless of the technology involved.
Artworks cited
Katsushika Hokusai 葛飾 北斎 (1760–1849). Tōshisen ehon gogon ritsu 唐詩選画本五言律 (Illustrated “Anthology of Tang-Dynasty Poems” in Five-Syllable Lines). 1836. Illustrated book. FSC-GR-780.218.1–5, Freer and Sackler Galleries, Smithsonian Institution, Washington, D.C.
Josetsu 如拙 (active 1405–1496). Hyō-nen-zu 瓢鮎図 (Catching a Catfish with a Gourd). c. 1415. Hanging scroll, ink and light colors on paper, 111.5 x 75.8 cm. Taizō-in Temple, Kyoto. Currently on deposit at Kyoto National Museum.
References
An, Y., J.H. Yu, and S. James. 2025. “Investigating the higher education institutions’ guidelines and policies regarding the use of generative AI in teaching, learning, research, and administration.” International Journal of Educational Technology in Higher Education 22, no. 10. https://doi.org/10.1186/s41239-025-00507-3.
Bellemont, F. 2022. “Resources for the Transliteration of Standard Chinese, Korean and Japanese.” Digital Media in Asia, December 23, 2022. https://digmedia.lucdh.nl/2022/12/23/resources-for-the-transliteration-of-standard-chinese-korean-and-japanese/.
Biggs, John. 1996. “Enhancing teaching through constructive alignment.” Higher Education 32, no. 3: 347–364. https://doi.org/10.1007/BF00138871.
Crawcour, Sydney. 1965. An Introduction to Kambun. University of Michigan.
Foka, Anna, and Gabriele Griffin. 2024. “AI, Cultural Heritage, and Bias: Some Key Queries That Arise from the Use of GenAI.” Heritage 7, no. 11: 6125–6136.
Wang, Hao, Hirofumi Shimizu, and Daisuke Kawahara. 2024. “Kanbun-LM: Reading and Translating Classical Chinese in Japanese Methods by Language Models.” arXiv preprint arXiv:2305.12759. https://arxiv.org/abs/2305.12759.
Yamamoto, Hideo. 1997. “A Mysterious Painting, Josetsu’s Catching a Catfish with a Gourd.” Translated by Melissa M. Rinne. Painting Stories, Kyoto National Museum, April 12, 1997. https://www.kyohaku.go.jp/eng/learn/home/dictio/kaiga/fushigi/.
Yonemura, Ann. 2014. “Commentary on Katsushika Hokusai, Tōshisen ehon gogon ritsu 唐詩選画本五言律 [Illustrated ‘Anthology of Tang-Dynasty Poems’ in Five-Syllable Lines].” Freer and Sackler Galleries, Smithsonian Institution, November 2014. https://pulverer.si.edu/node/284/title.
Cover Image: Illustrated “Anthology of Tang-Dynasty Poems” in Five-Syllable Lines (Tōshisen ehon gogon ritsu hairitsu 画本唐詩選五言律排律), 2013_735_a_e_d_007, from Metropolitan Museum of Art Libraries: Digital Collections, https://libmma.contentdm.oclc.org/digital/collection/p16028coll7/id/12794.
One thought on “Think Tasks Not Tools: Teaching AI Literacy in East Asian Art History”