Wolfgang Spahn

the sonification of the computing process in an analog neural network, 2019

The first song a computer intoned was in the 1970s when an IBM 704 at the Bell Labs sang‘ Daisy Bell’. The same song was sung by the artificial intelligence HAL-9000 in the film “2001: A Space Odyssey” by Stanley Kubrick.
But why should an AI sang human songs?
Won’t they prefer their own beats and melodies?
The sonic structures produced by the eternal oscillation of analog neural networks is the more possible soundtrack when the technological singularity is reached and AI takes over. At that point in the future, when technology becomes uncontrollable - like the mathematischen John von Neumann predicted in the 1950s - the remains of the human civilization will probably listen to some kind of “Echo State Network”

The installation raises the question: can one hear the firing of neurons? Does their action potential have pitch and timbre? How do patterns and structures of a neural network look and sound? And what will happen if the network listen to these sound and pattern, too - and feeds this information back into the network?

At the core of the sound installation “Echo State Network” by Wolfgang Spahn is an artificial analog neural network that generates sonic pattern and audible structures. The circuit of the neuron is an adaption of an early neural model originally designed by Japanese mathematician Shun'ichi Amari to explain the human heart beats. Thus the factor time and feedback are added to the network. Compared to Facebook and Google networks this neural systems is much more complex and manifold in its appearance. One can say this analog artificial network mirrors far more the chaotic human nature than the digital AI implementations.

As part of his Analog Computer Confetti the artist created an analogue electric circuit that mimic a model of a neuron. More than 150 of the Confetti Neurons will form an analogue version of a so called “Echo State Network”. And because the model for that neuron was developed to explain our heart beat this network will oscillate, pulse and create most complex pattern one can see and listen to.

An Echo State Network is a particular kind of model of a recurrent neural network. At it‘s core is a complex, chaotic and back coupling neural network, similar to a Liquid-State Machine. At the networks input layer a couple of sensors will react to sound and light. Brightness and frequencies will be processed and fed back in the Echo State Network. A couple of firing neurons that oscillates in a audible frequency will form the output layer. Indicator LEDs shows the action in the network as well as 8 speakers make these activity hearable at 8 different nodes of the network. Thus the visitor can witness the activity in the neural network and can get an idea of the complexity of these processes.

“Echo State Network” was shown at Alt-Space Loop as part of the exhibition Sound Effects Seoul Korea, 2019.
With support of Goethe Institute Korea.

… and many thanks to Martin Kuentz for helping to solder one million neurons.