Robot Class
What an AI design course in 2021 taught me about language as the designer’s last material
In fall 2021, I wrote a graduate design course at SCAD around artificial intelligence.
The course wasn’t a novelty class, and it was not about chasing tools. The question was more basic and more unsettling: what happens to design education when the tool can generate?
If image, voice, variation, and style can be produced outside the designer’s hand, then the one material that survives is language. Every assignment in the course, I realized later, was the same assignment: could you describe a thing precisely enough that someone (or something) that could not see your intent would build it anyway?
At the time, generative AI did not yet have a mainstream public language. ChatGPT was still a year away. As Chair of Graphic Design, I wanted students to meet the shift before it arrived fully packaged by industry. Formal approval for a new course would have taken longer than the moment allowed, so I took an existing graduate shell — GDVX 784, Visual Design for Interactive Contexts — and rebuilt it from the ground up.
It quickly earned its informal name: Robot Class.
A longer history than it looks
The course was not designed to tell students whether AI was good or bad. It was designed to help them recognize what kind of decision they were making.
The skepticism wasn’t new for me. My 2011 graduate thesis at Yale, Ready Made to Order, had argued that algorithmic personalization would fracture shared reality — and the decade since had not argued back. So the course began where that argument left off: not with the tools, but with the long history of machines that force a renegotiation.
We read and discussed work connecting automation to older questions of labor, authorship, imitation, and power: Frankenstein, the Golem, R.U.R., the Mechanical Turk, ELIZA, the Uncanny Valley, Lo and Behold, Klaus Schwab’s The Fourth Industrial Revolution, and histories of technological disruption, including the cotton gin and Captain Swing. Students brought in a current article each week.
The point was not a fixed canon. The point was that AI did not appear from nowhere. It belongs to a long line of machines that promise efficiency while forcing society to renegotiate work, trust, and identity.
One and Five Logos
The first major exercise, One and Five Logos, gave students a fictional shipping logistics company. The context was not abstract: Savannah has one of the largest ports in the United States, and from our classroom window we watched cargo ships move in and out all day. The assignment asked students to design a logo for the company five ways, each through a different model of authorship.
First, they designed it themselves from the brief I had written.
Second, they hired a designer on Fiverr for less than ten dollars and directed that person using language only — no visual references, no glimpse of their own logo, no feedback. If you cannot show, can you direct?
Third, they wrote a language-only survey asking respondents to choose the best logo for the company, with formal decisions embedded in the questions — things like what encasement should the logo have; circle, square, diamond, or no encasement at all — then designed the logo the audience had effectively requested.
Fourth, they designed against the survey: the version respondents had least chosen.
Fifth, they used a platform that could take language and generate a logo from the description.
One assignment, five models of authorship: maker, director, audience, contrarian, machine.
It was also fun, which mattered. Students came in with fear and skepticism. This project let them laugh, test, compare, and see their own value more clearly. The question was no longer whether a tool or a cheap labor platform could make something. Of course it could. It was whether the designer could frame the problem, direct the process, interpret the result, and defend the decision.
Where the illusion breaks
Another early exercise asked students to find a chatbot and talk with it until they could identify the exact moment it stopped feeling human.
Was it repetition? A failure of memory? A tone that felt almost right but not quite alive?
The students treated it as sport. They worked hard to pull the illusion down, and the results were genuinely funny. But the assignment opened a serious problem for designers: interfaces do not only function. They imply character. They ask for trust. They train people to accept certain forms of presence as normal. The students were not asked to admire the machine. They were asked to corner it and notice the seam.
Designing a synthetic character
The central project asked students to create an AI chatbot that solved a specific human problem.
We looked back to chatbots like ELIZA, an early example that simulated a Rogerian psychotherapist, because it raised a question that still matters: what happens when a system offers emotional structure without human understanding? Students imagined their own chatbot not as a generic assistant but as a designed presence with a purpose, a voice, a form, and a point of view. Some explored decision making based on prior behavior. Others took on practical or social needs. The constraint was constant: it could not be a logo with a Siri-like circle attached. It had to feel authored and alive.
Inside the project sat a smaller exercise, Write / Sit / Make. Students wrote by hand for five minutes without stopping, knowing no one would read it. They sat silently for three minutes. Then I gave them character prompts borrowed from screenwriting: What does this thing sound like? What does it do when you’re not around? Does it dance? Who is its best friend? What is its first memory? A designed intelligence needed more than a skin. It needed a mind, or at least the carefully constructed illusion of one. We were, of course, teaching them to build the very illusion the readings had taught them to distrust. That was deliberate: you cannot judge a seduction you have never constructed.
Students built prototypes, hired voice actors, animated states of thinking and response, and staged live simulations: they sat at the front of the room asking questions they had written while the character answered in voice and the visual system shifted between thinking, answering, emotion, and logic. Then we changed the scale.
The final was presented twice: first in class, at the scale of a device in the palm of the hand; then on a football field, where the same character became enormous and the voice became environmental. A friendly assistant in your hand becomes the voice of God when projected across a stadium. Scale changes meaning, and the same design choices carry different ethical weight when they become public and architectural.
Learning in public
The most surprising part was that I was learning almost as quickly as the students. I woke at four or five in the morning on class days and built the lecture from new material. That instability became part of the pedagogy: students could see the subject was not settled, and that was the point.
The course also made a broader curriculum philosophy immediate. I had built the department’s curriculum around a trajectory — the designer as producer, then director, then author — not as separate career tracks but as overlapping modes of practice. AI made all three visible at once. The designer still had to produce: forms made, prototypes shown. The designer had to direct: briefs, constraints, survey questions, voice. And the designer had to author — not in the romantic sense of total control, but in the responsible sense of deciding what a system means, how it should behave, and what consequences it might create.
What I would keep
If I taught the course now, I would update the readings and tools. But I would keep the structure almost intact. I would still begin with authorship rather than novelty. I would still make students compare their own work with outsourced, crowd-influenced, and machine-generated alternatives, test where the illusion breaks, build characters rather than interfaces, and present at more than one scale.
What has changed since 2021 is not the need for design judgment. It is the urgency of it. AI can make bad decisions look polished, shallow systems feel sophisticated, borrowed style appear intentional. The danger is not that AI will make design irrelevant. The danger is that it will make surface fluency easier to mistake for thought.
In the age of generative tools, the designer’s language may matter more than the designer’s hand. But the more important question may be what the designer is willing to be responsible for.
Michael Mikulec is a brand strategist, creative director, and writer whose work focuses on identity systems for emerging and complex categories. He served as Chair of Graphic Design at SCAD, where he rebuilt curriculum around the changing role of the designer and authored the school’s first course on AI and design in 2021. His work has included brand and design systems for Curaleaf, NBC, ESPN, Fenway Park’s centennial, and other sports, media, and cultural organizations. He writes American Alchemy, an essay series on memory, design, and American identity. More at michaelmikulec.com.