Three Sirens sitting on their island's shore surrounded by the modering bones of those not able to withstand seduction.
The Three Sirenes are partly robots, partly musical instruments. They teach themselves how to play. The implemented computer programs contain no information on notation, harmonies or musical systems. So called 'neural networks' (nerve cell stimulations) use self-organized and unsupervised learning to find out about improvisation and instrumental virtuosity.
The neuronets learn by allocating related stimuli to neighboring groups of neurons. Using this mechanism it is possible to reinforce similarities and to find hidden information within the presented sensor patterns (audio input). They learn to control their motors and mechanic characteristics (position sensor input). To give the instruments a chance to interact, their audio inputs are connected not only to their own amplifiers or microphones but also to each other.
They first play at random, listen to their own sounds and to the other sirenes. In the course of the performance, they develop rhythms, invent melodies and improvise. Through the human's user interface (electric guitar, microphone etc.) session-like experiences are made available to everyone.
Saying "this machinery is creative as musicians are" feels like touching one of the very human abilities since music appears to be extremely related to emotional response.