The original note, as written at the time
What if we were to start by teaching a machine to judge designers … to judge what good or fine design is … we could start with Architecture students etc … the machine can accumulate all the data it needs to finally start designing on its own. Things like structure and environmental consciousness should be easier to factor in before we get to the sentimental-sociological-aesthetic stuff.
Ongoing development
What if we were to start by teaching a machine to judge designers, rather than to be one.
It sounds backwards and it is the only route that could work. We have spent years trying to generate design and the output is impressive and unusable, because nobody has taught the system what good looks like in a way that survives contact with a client. Judgement is the missing faculty, and judgement is trainable in a way that taste is not.
Start with students. Their work is graded, argued over, torn apart in crits, and the reasoning is recorded in the marks and the comments. That is a corpus of comparative judgement about buildings — not preference, but justification — and it is the closest thing our discipline has to labelled data.
And the order of learning matters. Structure and environmental consciousness are easier to factor in than the sentimental-sociological-aesthetic business, and they should be learned first, not because they matter more but because they are checkable. A machine that can say this span will not work, this orientation will cook the west rooms, this stair does not discharge, has earned the right to be listened to about proportion.
Get that right and the aesthetic layer is not far, because the aesthetic layer is mostly a conversation about consequences. A building that is badly structured cannot be beautiful in any sense that survives. The sentimental stuff is the last ten percent, and the last ten percent is where all the argument is, and the argument gets much easier when the first ninety is settled.
I am not proposing a design machine. I am proposing a critic — one that has read every crit, remembers why the panel said no, and can tell a student on a Tuesday evening that the scheme was rejected for a reason that has a name.