DeepMusic played songs written by artificial intelligence. The way it worked was I first collected a bunch of music files in the midi filetype. Then I used a type of AI called a recurrent neural network (RNN) to learn how to predict notes. I used an RNN because they have a sort of memory and are well-suited to tasks that need to remember things. In this case, it needed to remember the order of music notes. I then trained the AI on the music dataset, and it would generate original songs.
I put the original songs in the DeepMusic Alexa skill. I was surprised by how good the songs sounded. You can hear a clip for yourself here.