TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—21 · POSITIONAL · HELD
Seduction is not quality
The test of an image is what decision it lets you make.
The dilemma
The most cinematic image on your wall draws every eye, including yours. Someone asks what it is for. You realise you do not know. It is there because it is beautiful. Is that enough?
The choices
Keep it, because it makes the wall look better. Or put it to one question: what design decision does this image let me make? The court depth, the roof material, the way the street meets the door, anything at all. If the honest answer is "nothing," the image is decoration, however good it looks.
The consequence
Keep images that answer no question and the wall slowly becomes a performance of prompts, output and fluency with the tool, and the performance starts to feel like designing. Apply the question and the wall thins out. What remains is working material: fewer images, each of which moved something. That is a wall a juror can read.
The case
A student's review board for a weavers' cooperative in Ilkal: twenty-two images, three of them stunning. The tutor asks of each stunning one, "what did this let you decide?" The first: the decision to use a sawtooth roof. A real answer. The second: nothing. It was a "vibe." The third: nothing. It was the one the student's friends liked. Twenty-two images became nine, and the nine made the scheme readable for the first time.
Our position
Many image models are tuned on human preference, which means they are built, in part, to flatter the eye. So is the performance of using them in public. Neither is designing. The test of an image is what decision it lets you make. An image that answers no design question is noise, however cinematic.
Why we hold it
The Lab has watched practitioners and students drift from design into display: feeds of outputs, prompt reveals, speed treated as a virtue. It is seductive because it looks like mastery. The decision test cuts through it with one question, and it is the question a client or a juror will ask eventually anyway.
The strongest objection
Beauty is information. An image that makes the room go quiet has told you something about what people respond to, and a designer who files it as "noise" because it does not yet attach to a decision is throwing away evidence about what people want. Sometimes the decision an image enables is only discovered later. Applied rigidly, the test punishes the image that arrived early.
What would make us revise it
The test allows "not yet" as an answer, if the student can say what question they are holding the image against. We would rewrite the card if we found that decision-tested walls produced duller schemes than free walls across a cohort, which would mean the discipline was costing the work its ambition rather than its noise.
Try it
Reject a beautiful output because it answers no design question, and name the question it fails to answer. Then find two images that look equally good, one useful and one empty, and say what separates them. Being able to say that is the skill.
Take it to crit
Ask the student what the prettiest image on their wall lets them decide. If the answer is "nothing," the wall is decoration. Ask them to take it down and see whether the scheme is weaker. Usually it is clearer.
How it works
Several image systems are fine-tuned using human ratings of which output looks better. Over millions of ratings, that training pushes the model towards whatever people click on: saturated light, dramatic angles, polished surfaces. The output is optimised to please, which is not the same as optimised to be useful. The same pull acts on the person using the tool. The reward for posting a striking output arrives faster than the reward for a resolved plan. The decision test is a deliberate counterweight to both.
What this idea builds on
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.