ImprovCues: Musical Creativity and Ideation with Machine Learning

ImprovCues is an installation that engages the audience with machine-learning generated musical cues for group improvisation. The audience helps to train the system by suggesting new cues and trading them with the system - for each cue suggested by the audience the system generates a new cue in return. A spontaneous performance is created through the interplay of suggesting, interpreting and attempting to play the prompts. Our aim is that the installation allows the hands-on exploration of machine-learning mediated creativity. In particular, rather than asking if machine learning can be as creative as a human musician, we instead ask if machine learning can be an equal partner in the stimulation of human creativity. ImprovCues explores how the act of training machine learning can in itself lead to a creative feedback loop, where participants ideas and machine generated ideas become inextricably tangled, raising questions about creative ownership and the locus of creativity