Interesting topic. There was a thread on AG Hotspot on this very same topic. Basic gist was the lack of credit and/or compensation for the use of copyrighted material used for training by AI should make the use of that training either unusable or illegal. Same holds true for image/likeness/voice stipulated in the collective bargaining agreements negotiated by the Screen Actors Guild. And to a different set of agreements negotiated by the Screen Writers Guild as part of their agreement to end the strike. This will all end up in court someday. (Some of it already has.) And it will take years, if not decades of appeals and counter suits before it is resolved.
I have no trouble with a company training AI on its own data in order to better predict the outcome of corporate decisions. I do have a problem though, in AI being trained on data that is not in the public domain in order to replicate any artistic endeavor. Of course that is my opinion.