II. RE: SHADOWPLAYS: I've a lot of thoughts but not the right words. I question the merit of whether AI (specifically neural-network image generators) generating vague architectural images is that much similar to skiagraphic drawings ... they both have a certain vagueness. The vagueness of skiagraphic drawings and models was chosen by a human, represented by a human. In the case of AI the images are picked by AI in a black box, using the database of images fed to it by the human/s behind its training.
The text is certainly mostly optimistic (at least until the end where some of the wording is more neutral — or even when Witt says something using the verb "erodes". That seemed worrying or generally aware of some sort of negative consequence, but I could be misreading his tone here). The text makes no mention of the energy/sustainability consequences of AI usage. It chooses not to engage in that. It is not the focus of the text. Sure. I figure maybe one or more of the other says in the collection this seems to come from might have done so. Quoting Gerber: "The invisibility of technology often makes its relation to the 'ground' equally invisible."
AI bridges everything through data. It regresses to the mean, often; LLM writing quirks are an example of such. AI architectural images — at least meaning those purely of content by AI with not enough human interventions of clarity — are not "architecture". Indeed, I believe Witt would agree with that. But it is likely he'd also say "but neither are skiagraphic drawings / models". Erosion, sideloading, offloading, lack of spatial thinking in AI architectural images.