TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 19—22 · POSITIONAL · HELD
Defending AI work in jury
You defend the chain of judgment, not the image.
The dilemma
A juror points at your best image and asks: "Is this real?" You have three possible answers. Defend the image: it is close, it is what I intended. Disown it: it is just AI, the real work is in the drawings. Or answer the question that was actually asked: say what this image is, what in it is settled and what is still a guess, what was generated and what was drawn, and what decision it carried. The first two are easier. Only the third survives the follow-up question.
The choices
Defend the render. Disown the render. Or defend the chain of judgment that made it, and be ready for the juror to find the one place in the chain where you cannot account for a decision, because that is what a good jury is for.
The consequence
Defend the image and the jury will find the load path that is not there, and the scheme will fall with it. Disown it and you have told the room that the most persuasive thing on your wall is not yours, and the room will believe you. Defend the chain and the image becomes what it always was, evidence of a decision, and the jury moves on to arguing about the decision, which is the argument you wanted.
The case
A student presents a housing scheme with six generated views. "Is this real?" The answer, practised: "This view is a mood study from cycle three. The court proportion in it is settled and matches the section. The balcony depth is not settled. I have since changed it, and the drawing on the left shows the current depth. The material in the image is wrong. The model gave me exposed concrete and the scheme is brick. I kept it because the light was what I wanted to test." The juror's next question is about the balcony depth. The jury has begun.
Our position
Presenting AI-assisted work to a jury means saying what is still a guess and what is settled, what was generated and what was drawn. You defend the chain of judgment, not the image. The pin-up is not the proof. The decisions behind it are. The student who can give that account has nothing to hide and nothing to apologise for.
Why we hold it
A jury exists to test decisions. When the image is defended, the test is wasted on a likeness. When the image is disowned, the student has given up authorship of their most persuasive material. The account (settled, guessed, generated, drawn) is the only answer that lets the jury do its job, and it is the same account a client and a building authority will eventually require.
The strongest objection
A narrated chain of judgment is evidence about a process. It is not the process. A fluent student can build an impeccable account after the fact — settled, guess, generated, drawn — around decisions that were never made in that order, or never made at all; a weaker speaker can have judged well and explain it badly. A jury that rewards the account rewards the narration, and the student who learns this card best may be the one who learns to rationalise best. Prejudice against generated images is the smaller problem, though it is real and it costs marks in rooms where it should not. The larger one is that the defence this card teaches can be manufactured.
What would make us revise it
The objection is why the account must rest on artefacts made at the time — the iteration log, the versioned briefs, the trace — and not on memory. A narrated chain with no contemporaneous record is a story, and a jury should treat it as one. We would revise the card if logged and unlogged accounts scored alike at juries over several cohorts, which would mean the record was not doing the work we claim. And we would revise the advice, not the position, if full accounts kept scoring below concealment: then the Lab would have to say so, and to work on the juries rather than on the students. It is reviewed every edition against what juries actually do.
Try it
Present an AI-assisted scheme to a classmate acting as juror. Under questioning, say what each image is (settled, guess, generated, drawn) and get it right. Then answer "is this real?" with the decision chain, not with a defence of the render.
Take it to crit
Under jury pressure, does the student defend the image, or the judgment that made it? Ask "is this real?" and listen for whether the answer is a defence, a disowning, or an account.
How it works
The account has four parts, and they can be rehearsed. Settled: what in this image matches a drawing or a decision you have made. Guess: what in it you have not yet resolved, and what you are doing about it. Generated: which elements the model supplied and which you directed. Drawn: what exists in your own hand, and where. A student who can give those four for every image on the wall has done the work this strand describes (forensics, the six layers, the log), and the jury becomes a conversation about the scheme. The wider argument, that the studio must now evaluate reasoning rather than output, is made in the Lab's essay on architectural thinking after AI.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
- judgment_register
- iiid-saurashtra_ai-fundamentals
- ETH.A
- What Counts as Architectural Thinking After AI
- The AI Design Director Protocol
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.