Artificial Intelligence can be used to autogenerate music, either by generating new notes or samples using generative neural networks. One way is to use a generative adversarial network and another way is to use a variational autoencoder, maybe in combination with a recurrent neural network.
The first AUv3 plugin that I'm working is a wavetable synthesizer where you will be able to generate waves from a neural network variational autoencoder. This autoencoder has been trained on over 4000 waves and has learned an underlying latent space to reconstruct many of these waves using only two parameters.
This graph demonstrates a number of different waves generated by the variational autoencoder. Although it can generate quite complex waves such as the waves above, some parameter settings will also generate simpler waves such as square waves.