In a typical neural network application, layers of simulated neurons are used to compute an output based on some input. The Ecstacy of Communication explodes the network to a human scale, both in terms of space and in terms of time. The result is a spatially-situated conflation of neural network and swarm intelligence, a population of sound-making devices interacting with each other and the sounds in their environment. Our focus shifts to the very activity of the network; the generative process itself becomes the art.

See movie: http://machinatus.net/tec/tec_240.mov

In the adjoining text, I am tracing the development of aural art in the last century, focusing on the role of the artist working with increasingly mechanized composition methods. From serialism, through chance operations, to recursive grammars and neural networks, the composing subject is progressively removed from local decisions and begins designing processes and systems, developing form in the abstract, playing with intensities, with intuition. Combining ideas from Attali, DeLanda, and Deleuze, I am suggesting that recent non-symbolic approaches to AI are more appropriate in the making of art and perhaps also in the modeling of intelligence. I, furthermore, consider the spatiotemporal specificity of sound, examining the behavior of sound empirically, rather than in abstract theoretical terms, exploring some of the codetermining interplay between sound and architecture.

Software simulation: http://machinatus.net/tec/software_a.html
This simulation was originally developed in order to model the expected behavior of the Ecstasy of Communication physical installation. Although the perfect world of software lost the inherent variability of physical systems, the tool proved quite useful in designing the hardware and software used in the actual installation.

After the completion of the hardware installation, the modelling software became its own end, and a host of new features were introduced. Individual nodes can now move about, drawing functions can be limited, and the sound can be filtered. Both potential and actual network connections are shown, and signals along the connections are visible.