Introduction
Music is data. AI researchers are beginning to use this data to create machines that make music. This project takes jazz improvisation as a (thorny) example of a problem that haunts AI-generated art: what happens to the human? The answers to this question will be in the form of scholarly and creative reflection.
Explaining the science
The project takes the idea of the "social machine" as its point of departure. Jazz is made on the basis of "operations" (interactive improvisation) performed on "data" (song structures, rhythmic conventions etc). The project argues that jazz as a global practice is a "social machine" (e.g. Wikipedia). It aggregates the "energy" of networked participants - their data - and converts this via interaction into meaning.
Good jazz improvisation is the result of a complex social interaction. The improviser’s approach to the “changes” or horizontal chord structures of song tunes or “standards” - classic jazz’s underlying data set - depends on individually acquired historical awareness. A great jazz solo ingeniously “quotes” previous jazz solos and may even shift stylistic register or improvisational language midstream for expressive purposes. Such advanced improvisational constructions are in turn preserved (often by mechanical media) and become part of a given jazz community’s “memory” of a tune. Thus an excellent improvisation “annotates” the original tune and is situated within that tune’s history.
One might claim that this intertextual quality of a great jazz improvisation parallels the validation of financial value by blockchains. The blockchain - numerical data that aggregates over time - requires ever more computer power to authenticate financial transactions. Likewise, to “know” the improvisational history of a tune, an AI would require significant amounts of training data. The result would entail statistically governed appearances of “historical” material in idiomatically and communicatively effective locations or contexts. The problem seems complex: even if it had access to the entire history of jazz, how could an AI, on the basis of statistics alone, know what to do and when?
Project aims
The centerpiece of the project is a public-facing creative-artistic response to the problem of AI jazz. This will be flanked by more traditional academic outputs and networking events involving academic partners from arts and computer science institutions in the UK, US and Taiwan. Both forms of intervention aim to reach a wide and diverse audience, and lay the groundwork for larger research projects.
Applications
Understanding the problem of AI-generated musical composition is critical to the fortunes of the burgeoning global AI music industry, which encompasses streaming services such as Apple Music and Spotify, as well as countless start-ups, many here in the UK. In addition, music practitioners need to learn more about AI-generated music. In the not too distant future “real” musicians may find themselves making music with AI "colleagues.” This is a problem that needs to be understood.