If you buy that a violation of copyright is stealing, then 1) even cases where a human artist created a fundamentally new work but none the less “copied” the original work, it was considered copyright infringement. “Rogers v. Koons”. 2) The infrastructure of AI requires “copying” the work into their database for learning. This copy is as a fact distributing the work without permission. 3) All techniques of AI are fundamentally “pattern matching” through math where the algorithms and their cost functions attempt to minimize the error of reproducing the pattern. Which is a fancy way of “copying”. 4) As a practical matter if creators are not able to enforce their copyrights by excluding others from copying and distributing their works without their permission they will not have an incentive to produce those works, if everyone goes to ChatGPT for a cake recipe, many of the recipe websites they scrapped from will not get their ad impressions and won’t be able to pay their server architecture, eventually they will close shop. Which will benefit the first movers OpenAI, Google, Microsoft, etc who have already copied the corpus of the web used in these AI implementations. I think creators, authors, and artists have a legitimate complaint. They should be able to extract payment from a license to use their works. Like they would if anyone else was going to create a derivative work from their efforts. If Sampling a baseline for use in your music even if you modify it is a violation of copyright, then it should be atleast as restricted as generating a model artifact that can reproduce your bass line with any transformations requested.