Is this really that fundamentally different from a human artist? Indeed, the person generating the art has what they want in their head, just as any other artist does, they're building prompts seeking to realise that vision on the screen (with tooling, just like a traditional artist would). Yes this is a new tool that requires far less technical proficiency with traditional techniques - people have constantly criticised new tooling (e.g. the criticisms sometimes still levelled at digital artists)

Further, even if you totally discount the human agency in this process & view it as a machine doing everything: we don't expect human artists to add a credit on each of their works to all the artworks they have viewed in their life and the experiences that came together to form an artwork or even their particular style.

I'd argue the networks are clearly transformative (since the network weights aren't large enough to contain the actual input artworks). It may not be as transformative as a human learning process (although then again, what is the learning 'storage' capacity of a human compared to a 4GB weights file?)

They never credit the developers that wrote the libraries either, nor the researchers that came up with the algorithm, or von Neumann...

Because they don’t know, and they have incentives to not know.

Because that would be a series of tortuously long encyclopedias that nobody would read.