Towards Intelligent Human-Machine Interaction: Learning to Create in a Common Effort
Peter Beyls / Belgium
Interactive composing implies dynamic on-the-fly musical negotiation between a live performer and some musical aptitude captured in a computer program. Much software interfacing human and artificial players involves mapping features in human input to parameters affecting output entailing responsive behavior.
Our work suggests viewing human and machine as convincing creative entities sharing a common biotope with equal authority. No one is in control; man and machine express mutual influence and complex behavior emerges from the interaction of many simple cognitive building blocks. Such systems exhibit unpredictable though coherent, life-like behavior.
Our method takes inspiration from human psychology: we implement artificial relationships, explore qualitative features in perception and competing machine motivations. In addition, a reinforcement-learning algorithm aims (1) to maximize sensitivity and diversity in system behavior and (2) to deeply link human and machine initiative as to sustain a compelling and rewarding interaction format.
Over the years, various systems were implemented approaching intelligent interaction from the above perspectives. Most programs suggest an interaction format of one human soloist interacting with a small virtual chamber orchestra, itself supporting the expression of social affinities between individual players. Our presentation sketches their specific conceptual orientation and computational architecture. We conclude with a live demo.