Improvisation without representation: artificial intelligence and music

摘要

Abstract—The title is meant in two senses: both as a reference to the fact that contemporary freely improvised music is underrepresented in the study of AI and music; and, as a reference to the behavioral robotics research of Rodney Brooks, who famously stated that in place of computational representation, “the world is its own best model”. Brooks cites the theoretical work of Marvin Minsky as a source for his approach to designing intelligent robots. Both Minsky and Brooks describe decentralized, agent-based systems that could arguably be regarded as improvisational, in that they are designed to cope with a dynamic external world. Though neither author explicitly identifies improvisation as a key aim of their approach, some AI systems are, in fact, explicitly designed to engage in improvisation. One example is George Lewis ’ Voyager, a humancomputer interactive musical improvisation system that negotiates and contributes to complex sonic environments in real time. This paper describes a new system, Adam Linson’s Odessa, that interacts with human musicians to perform freely improvised music, following Lewis ’ interaction model. Odessa is designed using Brooks ’ subsumption architecture, thus bypassing central representation and control in favor of a decentralized and environmentally situated approach. Artificial intelligence (AI); music; improvisation; humancomputer interaction (HCI); interactive systems; robotics; subsumption architecture I.