There are two issues -- (1) feeding copyrighted material in to an AI model, and (2) getting copyrighted material out.
The latter is obviously a violation of copyright, full stop.
The former, to me, is obviously not a violation. If it were, that would massively tilt the playing field in favor of large corporations. It would become very hard to independently train your own models. Philosophically, I go by the principle that if it's (il)legal to do yourself, then it should be (il)legal to do the same thing with an AI's assistance.
The massive complicating factor is that nobody knows how to do (1) without also doing (2) as a side effect, because we don't understand how deep learning works well enough to control it.
The latter is obviously a violation of copyright, full stop.
The former, to me, is obviously not a violation. If it were, that would massively tilt the playing field in favor of large corporations. It would become very hard to independently train your own models. Philosophically, I go by the principle that if it's (il)legal to do yourself, then it should be (il)legal to do the same thing with an AI's assistance.
The massive complicating factor is that nobody knows how to do (1) without also doing (2) as a side effect, because we don't understand how deep learning works well enough to control it.