You should be able to judge whether something is a copyright violation based on the resulting work. If a work was produced with or without computer assistance, why would that change whether it infringes?
That's why clean room reverse engineering and all of that even exists.
As a descriptive claim, it isn't correct. Several lawsuits relating to sampling in hip-hop have hinged on whether the sounds in the recording were, in fact, sampled, or instead, recreated independently.
This is interesting from the legal point of view, because AI service providers like OpenAI give you "rights" to the output produced by their systems. E.g. see the "Content" section of https://openai.com/policies/eu-terms-of-use/
Given that output cannot be produced without input, and models have to be trained on something, one could claim the original IP owners could have a reasonable claim against people and entities who use their content without permission.