The non-linear medium of computer gaming can lead a player down an enormous number of pathways to an enormous number of resolutions. From the standpoint of music composition, this means that a single piece may resolve in one of an enormous number of ways. Event-driven music engines (or adaptive audio engines) allow music to change along with game state changes. Event-driven music isn't composed for linear playback; instead, it's written in such a way as to allow a certain music sequence (ranging in size from one note to several minutes of music) to transition into one or more other music sequences at any point in time.
A separate definition as coined by Christopher Yavelow concerns the reliance on a continual feedback loop consisting of a user or users evaluating music generated by seed material provided by themselves (or itself generated). The software incorporates their preferences (opinions, tastes) into subsequent compositions by adapting the models it uses to compose in the future based upon this user feedback. Adaptive Music models are collections of compositional methods, rules, and constraints. The entire learning process allows the software to continually move closer to the ideal composition for an individual or collective group of users.