> Imagine an AI that generates songs. Imagine it generates millions of songs. What are the odds some of those songs will accidentally and independently "copy" features deemed distinctive and commercially valuable in existing songs under copyright?

Another question is: what are the odds that a future song will inadvertantly copy one of the computer-generated songs?

Given that there are only a finite number of notes and chords, that some note-sequences are more "musical" than others, and the computer can easily generate billions of songs, one can imagine a musical copyright troll (like a patent troll, but with copyrights):

1. creating a large number of songs on a computer

2. writing software to detect similarities between new songs and the computer song collection

3. shaking down songwriters/artists for profit

> Copyright law is going to have to change.

People have been saying that for some time. My own take is that it will change when it no longer suits the rich and powerful.