These AI art is only trained on human art, so in theory, there's no way for it to create something completely new that's based on human experiences.
Take the example of psychedelic art.
The way that we humans produced it is through taking some LSD or DMT and then drawing whatever shapes and colours we saw whilst we were tripping.
How is an AI ever supposed to do that?
The best you can do is feed an AI reference images of preexisting psychedelic art as training data and let it chop and combine bits and pieces from different works and what you end up with is a Frankenstein mess that the user of the AI then labels as "original art".
It doesn't "understand" the concepts behind different art pieces and where the ideas originated from, it simply looks for patterns in human labeled images and learns to combine them in a semi realistic way.