I talked about this in a different post so just gonna copy/paste that answer here
So here's the thing. We're all humans and without even thinking about it we anthropomorphize things, especially AI partially because of the marketing and the name, but contrary to the name, it's not an intelligence. It's not referencing and it's not being inspired. These are all human things that humans do. It's just a piece of software. It is closer to an insanely sophisticated Photoshop filter than it is to any actual intelligence that exists.
Machine learning might be better called machine processing, or machine neural analysis or something like that because it's nothing like what a human is doing when they learn something. Even when a human can't draw well and has no training, they can still draw and communicate ideas with symbols. This software literally can't do anything without first processing artists artwork or photos or whatever. It has exactly zero value without the artists work adding all the value. The data is everything.
The worst skilled artist, while their work might not be aesthetically pleasing, it's still an expression of that person. These diffusion models are just a laundering system for IP infringement.
I don't think that is an entirely accurate description of what AI is though. The whole point of AI is to model human thinking with some techniques very literally doing so (neural nets to mimic neuron activity for example).
On the other side of it, what distinguishes a software "brain" from a biological one, conceptually speaking? Couldn't you argue that the brain is simply biological hardware and software executing its own biological algorithms? Just because we don't fully understand the processing of the brain doesn't mean it isn't happening. While you can come up with all kinds of technical differences as far as what is happening in the brain vs. in software/hardware, conceptually I don't see that the difference is significant. Both take inputs, learn from them in some way, and produce outputs based on that learning. It seems like it would be hard to come up with a meaningful legal distinction. Copyright protects a work of art. It does not protect someone (or something) from deriving information (style, etc.) from that art. At least I don't see how it would.
A baby doesn't come out of the womb knowing how to draw a circle. That's learned through observation and experience. AI does the same thing even if the methodology isn't exactly the same. If I like a particular artist and decide I want to draw something in the same style (or combine styles from multiple artists based on paintings I've seen), is that IP infringement? If not, then why should it be for AI? We aren't (typically) talking about literally copying a painting or photo (though I'm sure an AI could be made to do that but why bother? If you want a copy you would just make a copy). I just don't see how there is some sort of blanket IP infringement here.
Quite frankly, I think it's all rather moot. Whether or not an AI data set has copyrighted material or not isn't going to have much of an impact on what it is capable of. There's virtually unlimited non-copyrighted data to chose from. It just seems analogous to banning human artists from looking at copyrighted material because it will influence their output. The data sets used probably have a lot more to do with convenience than with some strong desire to use copyrighted material. For opponents of AI art, forcing out copyrighted IP won't really change anything. It isn't going anywhere either way.