Machine-generated text, artwork or music – by definition – cannot be copyrighted. In some cases it looks and feels derivative of other works. That may depend on the range and style of the training data it is using. Obviously, if all the training data is in the public domain, then any argument about its provenance is moot.
I’m currently treating everything as if it was assigned a Creative Commons license. So, when I remember, I attribute its source and label it as machine generated.
Of course, in practical terms, if no copyright can be assigned, it’s a free-for-all for those of you who like this stuff. You can re-mix, edit and republish however you want. Incorporate it into a worship service, concert, movie or computer program.
It would be fun to know where it is used. As the writer of Ecclesiastes said: “Cast your bread on the waters; for you shall find it after many days.”