It seems that Valve’s issue is mainly with AI generated assets from tools whose data set is of questionable origin and even then it seems to be to limit legal liability until there’s more clear precedent.

As for analogies of humans learning based on other people’s work, those might not work depending on how the AI tool in question works.

If it’s a tool that purely uses the data set to understand concepts and generates completely unique products then it might be a good analogy.

If however it’s a tool that essentially produces a collage of parts copy and pasted from different sources and smooths things over bit, like how Stable Diffusion did to the point of copying the Getty images logo, then the analogy might fall flat on its face and it might be a more murkier area from a legal perspective.