Doug Eck Advances Neural Network Creative Expression

These early creative expressions with neural networks continued decades of previous research covered in the Timeline by Peter M. Todd, J. P. Lewis, Jürgen Schmidhuber and others. A key figure in the foundation of modern AI art, Doug Eck created four “Blues Improvisations” in the essay “Composing Music with LSTM Recurrent Networks - Blues Improvisation.” In this unassuming but fascinating paper, Eck, a former postdoc of Schmidhuber, explored using “Long Short-Term Memory” [LSTM] recurrent neural networks for music composition. The project focused on creating blues improvisations, training the network with a 12-bar jazz bebop blues structure. The LSTM learned to generate coherent melodies and harmonize with fixed chords, showcasing its ability to capture long-term musical structure. Often the first experiments with new technologies and system appear in music or text with their pre-set rules. Christopher Strachey and Max Mathew working with the earliest computers in text (1952) and sound (1957) respectively both pre-date Michael Noll’s first visual experimentation. As it is with neural network art, where artistic experimentation with music preceded the visual.