Using machine learning to combine (or deepfake) two songs into one?

I have always had enthusiasm for the future of audio processing by neural net machine learning algorithms. I have wondered what results might yield should "AI" be designed to deepfake music? I know inception nets and CNNs accomplish some of this already. But what would it be like to take two or three of your favorite songs and ask a machine to combine them and make 7 different variations (one for each day of the week)?

The copyright nightmares that technology could potentially cause could be big.

I wonder how artists will react to their music being remastered practically on demand?