Using AI Isn’t AI Literacy: The Mistake Fashion Schools Keep Making
Fashion programs are rushing to add AI tools to the studio. The harder work, and the more important one, is teaching students to understand them.
Imagine a second-year fashion design student with a Friday deadline. By lunch she has forty mood boards, a dozen silhouette studies, and a color story she likes, all generated in less time than it once took to pull tear sheets from magazines. Now ask her three questions. What was the image generator trained on? Why do nearly all the models on her boards share the same body? Who owns what she has just made?
If she cannot answer, she has used AI. She is not yet AI literate.
That distinction, between operating a tool and understanding a system, sits at the center of a paper I published last month in Discover Artificial Intelligence. My argument is that fashion higher education faces a literacy gap rather than a tooling gap, and that most current responses treat the second as if it were the first.
An industry moving faster than its classrooms
The fashion industry is not waiting. In October 2023, LVMH joined the corporate affiliate program of Stanford’s Institute for Human-Centered Artificial Intelligence, in its consumer goods, retail, and AI focus area. Generative tools now sit at the front of the design process, producing references, silhouettes, and colorways at speeds that were impossible a few years ago. Machine learning systems trained on social media, sales, and search data can flag micro-trends before human forecasters notice them. Design software such as CLO3D and Browzwear lets teams sample in 3D, cutting the number of physical samples a collection needs, and recommendation engines and virtual try-on are changing how people shop.
The executives running these companies know where this is heading. In the most recent State of Fashion report from The Business of Fashion and McKinsey, they ranked scaling AI as the industry’s biggest opportunity for 2026, ahead of product differentiation and sustainability. The same survey, two years earlier, captured the problem underneath that ambition: 73 percent of fashion executives called generative AI a priority, yet only 5 percent said they were ready to make the best use of it.
Education has not closed that gap. A 2024 study of fashion education by Kohli concluded that the skills current fashion design curricula impart are insufficient for work in a digital world, and called for bachelor’s programs to be revised. A qualitative study of 21 participants across five countries by Krasley, Cobb, and Gangoda (2023) found that automation is driving sharp demand for intellectual, innovative, and creative skills in the apparel workforce. The trade publication The Interline has described fashion’s pool of digital talent as less a gap than a “talent puddle.”
The default response: add the tool
When AI does arrive in fashion programs, it usually arrives as a tool. A Midjourney workshop appears in the design studio, a ChatGPT policy goes into the syllabus, a CLO3D elective opens in merchandising. Each of these is useful. Together they teach students which buttons to press rather than what the buttons do, treating AI as something to be wielded rather than something to be questioned.
The mistake is easy to make because using AI looks like competence. The student with forty mood boards appears productive, and in a narrow sense she is. Reading offers a useful parallel. A child who can sound out every word in a newspaper has mastered decoding, but that is not the same as being able to tell when an article is misleading, whose voice is missing, or what it wants its reader to believe. Nobody would call decoding alone literacy. Tool fluency is the decoding stage of AI literacy, and fashion education risks stopping there.
What literacy actually asks for
Researchers do not agree on a single definition of AI literacy. An integrative review of 124 studies by Gu and Ericson (2025) found the term still vague, stretched to cover everything from kindergartners interacting with social robots to graduate students engineering prompts for large language models. The most widely cited definition, from Long and Magerko (2020), describes a set of competencies that let people critically evaluate AI technologies, communicate and collaborate effectively with AI, and use AI as a tool at home, online, and at work. Using AI is one of the three parts, and it comes last.
The older literacy tradition explains why the other two parts matter. The psychologist Sylvia Scribner described literacy through three metaphors. As adaptation, literacy is the functional ability to do what a job or a society requires. As power, it is the ability to question, to critique, and to advocate. As a state of grace, it is a form of self-cultivation, the mark of a person who engages with the world on deeper terms. A tool workshop delivers the first. It rarely touches the second or the third.
The same narrowing happens with how AI itself is understood. Gu and Ericson, drawing on a model from media education known as the Dagstuhl Triangle, describe three ways of looking at AI: as a technical system that learns from data, as a tool that people use, and as a sociocultural artifact that carries the values, gaps, and politics of the people and data that shaped it. Teaching prompts sees only the middle view. A literate student holds all three at once, which is what lets her ask why her image generator keeps producing the same body, and what that default says about whose bodies were in the training data.
In the paper, I define AI literacy in broader terms: a set of competencies, critical dispositions, and ethical orientations that must be cultivated deliberately. Using AI is a functional act. Being AI literate is a habit of mind.
Why the difference matters in fashion
None of this argues against the tools. It is an argument about what happens when graduates hold the tools without the understanding, and fashion supplies no shortage of examples.
Get Sri Manchiraju, Ph.D.’s stories in your inbox
Join Medium for free to get updates from this writer.
Consider a merchandiser handed an AI trend forecast. It arrives polished and precise-looking. If she does not know what data the system was trained on or what it cannot see, she will treat its output as fact, and in an industry whose chronic problem is overproduction, an uncritically trusted forecast is an expensive mistake.
Consider a marketing graduate deploying a personalization engine. Recommendation systems and dynamic pricing raise real questions about privacy, consumer autonomy, and manipulation. A graduate who cannot read how these systems work, what data they require, and what harm they might cause will be neither an effective practitioner nor a responsible one.
Consider a sustainability team adopting AI to cut waste. Demand forecasting and 3D sampling can reduce overproduction and material use, and that promise is real. Every such tool, however, optimizes for particular metrics and quietly encodes trade-offs. Someone on the team has to know which ones.
And consider the designer. Generative systems are shifting her role from sole creator toward what Spaseska (2024) calls a co-creative sense-maker, the person who decides which problems are worth solving and which AI outputs are worth developing. That is a judgment skill rather than a button skill, and it is precisely the part a tool workshop leaves out.
Five questions that separate a user from a literate user
A practical test helps. A student who is literate, not merely fluent, should be able to answer five questions about any AI system she works with.
What was this system trained on, and what is missing from that data? Every output inherits the limits of its inputs.
Why did it give me this output rather than another? Generative systems produce plausible defaults, and defaults are choices someone else made.
Whose bodies, skin tones, cultures, and genders does it show by default, and whose does it leave out? In fashion, representation is never a side issue.
Who owns what I made with it? Questions of authorship and intellectual property remain unsettled, and a professional should know where the open questions lie.
What is it optimizing for, and what is it ignoring? This applies as much to a trend forecast or a sustainability dashboard as to an image generator.
A student who can answer these has moved from adaptation toward power, in Scribner’s terms. She can still produce forty mood boards before lunch. She can also tell you which three are worth keeping, and why.
What changes when literacy comes first
Moving from tool training to literacy does not mean removing the workshops. It means placing them inside a curriculum that asks students to explain, critique, and decide. Reviewing 29 studies, Ahn, Lee, and Chee (2024) found that at the university level, AI literacy should move past basic knowledge and device use toward understanding data and algorithms, solving problems, and building career-relevant competencies, with the emphasis shifting from how to use AI to how to integrate it critically, strategically, and responsibly. In a design studio, that might mean grading a student’s rationale for which AI outputs she kept and which she discarded, rather than only the finished board. In a merchandising course, it might mean asking students to audit a forecast before acting on it.
In the paper, I propose a framework for doing this, the Fashion AI Literacy Framework (FALF), which I will walk through in a later piece. Its status is worth stating plainly. The gap it responds to is well documented in the research I reviewed; the framework itself is a conceptual proposal that still needs testing through curriculum pilots and studies with students and faculty.
The stakes
Fashion is one of the world’s most powerful cultural industries, shaping how billions of people understand beauty, identity, and belonging. If AI is going to reshape fashion, and it already is, the people guiding that reshaping need to do it with intelligence, creativity, and conscience. Teaching students to use AI is the easy part. Teaching them to understand it is the work.
The full paper, A fashion specific framework for artificial intelligence literacy in higher education, is open access in Discover Artificial Intelligence.
