By Hengchang (Alan) Liu, NYU Cinema Studies MA
Research focus: AI & the humanities; digital/public culture; identity
Website: https://alanliu2.wordpress.com | Email: Hl5843@nyu.edu
Reflect further after exploring Alan’s work: Read this recent article, “The Cinematograph, the ‘Noematograph,’ and the Future of AI Art,” where author Ken Liu explores what Chinese poetry and early cinema can teach us about AI’s potential as an artistic medium.
PROJECTS:
1) Digital Humanities Final — “Translate” Reading Material
A proof-of-concept mechanism that “translates” dense theory into short AI-generated videos under light human supervision. Using ChatGPT and LTX, I rendered Stuart Hall’s “Cultural Identity and Diaspora” into a three-part video series (mechanical reading → script → visuals), testing both accessibility gains and losses of nuance.
2) Alternative Theory — Translating My Own Essay
I reused the same pipeline to visualize the argument of my Film Theory final (a critique of sensory film theory that returns emphasis to “meaning”). The 5-minute video applies the same workflow to my text, allowing me to observe how ideas shift when re-mediated as images and narration.
Website: https://alanliu2.wordpress.com/2025/04/30/alternative-theory/
Reflection:
Academic texts in the humanities often lack the visual and interactive appeal that resonates with today’s digital audience. These two projects explore the transformative potential of AI tools to reimagine how dense academic material is consumed. In this process, AI tools (ChatGpt, LTX) for text summarization, image generation, and video creation demonstrate how AI tools can “translate” traditional academic reading into engaging visual narratives with minimal human intervention. To test this mechanism, the first project “translates” Stuart Hall’s article Cultural Identity and Diaspora（1990) into a three-part video series. These videos explore themes of cultural identity, representation, and hybridity, demonstrating how AI-generated narratives can maintain academic depth while enhancing accessibility and engagement. For the second project, I continue the mechanism and “translate” my own film theory into a visualized AI video.
Ultimately, these project serves as a proof of concept, demonstrating AI’s potential to transform dense academic texts into engaging narratives. However, it also reveals limitations, such as the loss of rhetorical nuance and depth.
In my practice, AI functions as a method and experiment rather than a substitute for interpretation: a way to prototype arguments, expose assumptions, and make mediation visible. Both projects use a deliberately constrained pipeline—mechanical summarization → script → image/video assembly—so I can compare how meaning travels when the source is a canonical text (Hall) versus my own writing. The bluntness of “mechanical reading” is the point: by staging a controlled reduction into claim-and-evidence units, then iterating rapidly with text-to-image/video tools, I can see how small prompt and sequencing shifts reweight concepts, tone, and emphasis. I keep a lightweight version log (prompts and output snapshots), mark where nuance drops out or unintended connotations appear, and reinsert key transitions myself; this keeps authorship accountable and prevents AI from displacing analysis. Ethically, I label AI-generated assets, cite sources, avoid likeness synthesis without consent, and maintain a record of biases and failures when outputs deviate.
The two projects also bracket time: one finished in December 2024 and the other in May 2025; now, roughly six months later, I want to probe how the intervening advances in models, tooling, and interfaces reshape the very same mechanism—what improves (coherence, pacing, visual legibility), what still fails (register, ambiguity), and which editorial interventions remain necessary. For AI Pedagogies specifically, I think the mechanism offers a tractable template for classroom use. Still, the Hall case matters: that text is not neutral and carries its own politics and rhetoric. A next step, then, is comparative translation across different genres of academic writing—for example, policy briefs, archival descriptions, methodological essays, ethnographic excerpts, or film-analytic prose—to test whether the losses and gains I observed are properties of the pipeline or of the source’s discourse. In parallel, I aim to make the pipeline more auditable (exportable metadata for prompts, sources, and edits) so students and readers can inspect the transformation chain. The goal is not to “automate” reading, but to utilize AI to identify where meaning is created, altered, and debated.
My interest in AI stems from a genuine curiosity, not fashion or hype. In the early years of my BA, AI tools were weak—we were effectively in a non-AI academic milieu. When systems like ChatGPT drew wide attention in 2022, I experienced—not as gossip but as a concrete shift—their disruptive force. Like the camera’s challenge to oil painting, or mechanical reproduction to handcraft, this felt irreversible. Precisely because it is unstoppable, it warrants study rather than applause. At the end of my undergraduate degree, I wrote a thesis on the ethics of AI-generated text (with a focus on literature). Later, when I encountered digital humanities more systematically, I recognized the continuation of that inevitability. As traditional humanistic infrastructures contract and AI produces new forms of “alienation” in how we read, write, and circulate meaning, I began looking for workable new methods inside the humanities.
Expressive AI 