[Phil Stearns and AANN : install @ Soundwalk 2007]
An Introduction of Sorts:
During my two week residency at STEIM, I was furiously constructing many little electronic sculptures. 50 identical sculptures, piece by painstaking piece. Though I had not named it at the time, I was calling the project the “neural network piece.” As of September 22, 2007, the “neural network piece” stands completed and officially named AANN. AANN is a very uncreative however appropriate name, being an acronym for Analog Artificial Neural Network.
[AANN : install @ Soundwalk 2007]
Description of Project:
AANN is quite possibly the most absurdly meticulous project I’ve dreamed up, worked out, and fully realized. She is a strange very geometric looking jumble of electronic components held together by many meticulously formed pieces of various gauge bare bus wire, interconnected by crisscrossing webs of warm colored insulated wire. The overall form resembles something like a cross between a deep-sea rover, a stripped down spaceship, a high energy laser, and a squid. In fact, there is something life-like to the design of the electronics which comprise AANN. Two ear-like microphones pick up sounds and eye-like photocells detect changes in light, which then trigger a chain reaction of pulses through many organized layers of identical parts. These identical parts are meant to function as biological neurons. Indeed, the overall function of the electronic sculpture is to mimic a small neural cluster. AANN responds to stimuli by lighting up a red LED on each of the neurons with as they fire with the final group of neurons actually producing sound from three small speakers located on her rear. Any attempt to make conversation in AANN’s presence: whether it’s about her or not: will be rudely interrupted by an abrupt tone swooping downwards in pitch like a discontented whine. AANN is loud enough that in the right conditions, she will quite literally begin talking to herself. Attempts to interrupt her when she’s in one of these vicious cycles of self-call and self-response are futile, as oftentimes she’s the only one who can talk herself out of talking to herself. Though there was a mild degree of intelligence in her design, it can’t be said that AANN herself is intelligent. Consisting of only 45 crude neurons, AANN hardly possesses the necessary gray matter requisite for intelligence as we have come to term it. However, 45 Neurons does allow for some interesting generative and dynamical behavioral patterns which emerge in those wonderful moments when she’s intently listening to herself listen to herself. In the end, anthropomorphizing is the best way to convey the complexity of ideas behind the birthing of such a project, which in the next few paragraphs and through some fun illustrations, I will attempt to retrospectively document and account for.
A Longish Story:
The development of AANN started in December 2007 with the idea of creating a network of electronics which would pass around a signal obtained from “the environment” according to some predetermined indeterminate circuit design. Upon receiving that signal, each of the networked parts was to emit a specific pitched tone. The pitches and structure of the network were to be based upon a just intonation pitch lattice like the ones used by many micro tonal composers in obtaining pitch-sets for their compositions.
After a semester of doing “other things”, the network idea was picked up, dusted off, and polished to a new and improved “neural” network piece. I took this to my compositions lessons with David Rosenboom who sent me off researching a few different topics pertaining to analog computing, cognitive science and neural biology. Somewhere in the middle of the middle of all the research, I decided that listening to the “neurons” themselves made more sense then trying to make them do something like trigger pitches as they receive input.
[conceptual sketch of circuit]
There was a point at which reading stopped being useful and the only way to really move forward with the idea was to start designing and building. In early May, I began taking simple behavioral models of neurons and translating them into analog electronic circuits all the while well aware that the exact same task had already been done before by more qualified scientists than I. The reason I went against conventional wisdom, which would have me start with their design and modify it to suit my needs, was that I already felt I had a different idea of how I wanted to approach modeling a neuron. Reinventing the wheel was OK in my book because I felt I was building a different wheel. Frankly it may prove to be an overly complex one, however the behavior of the final circuit I designed was close enough to mimic the behavioral traits that could be derived from the limited data on the bio-electric properties of a neuron that I had at hand. With a working Neuron, It was time to take the project to STEIM.
[conceptual sketch of network path branching, interconnection & structure]
Part of the final crucial stage in developing AANN was figuring out how to put all the neurons together. The connections of a bunch of neurons was what made the network and how those connections are determined influence the final behavior of the entire system. Just how does the arrangement and connectedness of the network effect the behavior? I still can’t answer but it is worth mentioning that at this stage, I had not considered using a computer for modeling. Though it would have been incredibly helpful in determining which arrangements behaved in what sorts of ways, I didn’t find it too appealing to spoil the whole “without the use of a computer” theme the rest of the project – not to mention my recent practice – had been cultivated from. In the end I scribbled in my notebook and on random pieces of paper when I had ideas. Trying this and that arrangement of dots connected by some other arrangement of lines. For a month before I arrived at STEIM, the only work I put in on the project was in doing these sketches.
[sketch working out structure and connection pathways]
In the first week of the STEIM residency, I jumped right in and chatted only briefly with Takuro and Daniel about what it was that I was trying to accomplish. One of the big questions they had for me to ask myself was “why?” Though I didn’t have a proper answer, the smart-ass in the back of my mind was smilingly shouting, “Because!” Of course I held its tongue, but at this point it was difficult to explain the “because” simply because, in order to answer the “why?”, the network had to be made first. The concept, in case it hasn’t become apparent by now, was to build a sculptural electronics piece which exhibited some sort of interactive behavior. How would it be interactive and what sort of behavior was not even thinkable at this stage. I was essentially developing the (new) technology, which would hopefully allow me to create the interactive sculpture, without knowing what the technology would allow me to do. The project was then and probably can still now be viewed in terms of a science experiment. Of course that still evades the ever more pervasive “WHY?” but that’s for a whole other sort of blog entry. The most important thing for me was to build and they even provided me with some assistance in obtaining some crucial parts I was unable to bring to STEIM from the US. So for about two week and on a meager Indian style diet, I worked…
[assembly line of opamps for summing stage]
The outcome of all this building was the completion of most of the larger electronic parts of of each of the 50 neurons I had planned on building. So to clarify, each of the neurons has 4 major parts which perform the following functions: Input, Summing, Threshold Detection, Output. These 4 parts were assembled in sets of 50 while at STEIM. There was still plenty of work left but the most excruciating of the assembly was behind me by the end of the residency. At home the construction continued. The next two photos explain better than words.
[50 neurons before the final structural assembly]
[assembly of neurons into structures which comprised the "layers" of the neural net]
At a point somewhere in the middle of this post, I realized that this has become more of a docu-narrative than a project report. I’m not sure how one goes about ending one of these “docu-narratives” and I’m not sure if I have many words in closing except to give you my account of how things turned out in my mind. The people at STEIM were monstrumentally helpful in providing the atmosphere, the inspiration, and those final few components necessary to get this project off the ground. However, despite all the wonderfulness of STEIM and it’s good people, and all my best efforts, the AANN project has produced little more than some sort of squid baby, which does – aside from getting people to clap their hands and stomp their feet – little more than respond to sound and light by “lighting up like a Christmas Tree” and “shrieking like so many dying seagulls.” Needless to say, I’m thoroughly happy with the results.
More work by Phil Stearns can be found at: www.art-rash.com/pixelform