Noise and Information

Lafrance notes that technology has often been used to metaphorically embody complex bodily functions. Machine learning has been subjected to these types of analogies since its inception several decades ago, as neural networks are modeled based on the structure of the brain, consisting of numerous neurons which collectively function to propagate information. Although the comparisons between neural networks and humans have some merit, there are some significant differences. The most significant, which is mentioned in Lafrance’s article, is the fact that neural network architectures can be erased and retrained, a process that is not natural to the human brain.

In general, technological processes and entities are frequently used to make sense of phenomenon in the real world. For example, the notion that our bodies can be quantified by streams of information is a direct consequence of the prolific rise of the digitization of physical phenomenon. However, is it really reasonable to assume that it is possible to quantify a human being using collected information? At its core, these types of thoughts are based on the assumption that collected information is noiseless, and that the captured representations accurately embody physical phenomenon. It is unfortunate that information stored in a binary sequence is sometimes mistakenly thought to possess clinical accuracy, as the collection process is not without its biases.

Transforming any physical phenomenon into a binary sequence requires quantization at some level. Admittedly, if a sufficient number of bits are used to represent the material being recorded the quantization may not be that noticeable. Although collecting information from a physical source is sometimes understood as the removal of superfluous noise from the source, it could also be construed as the addition of noise, as the physical reality is distorted. The image below shows the digitization of an audio waveform at two different bit-depths, 8 and 16. When looking at the image below, it is immediately clear that both the 8 and 16-bit versions of the original waveform do not preserve some of the characteristics of the original waveform. In a very clear way, the quantization of an analog signal, adds noise to the signal.

As a result, making a distinction between information and noise can be difficult, which is thoroughly discussed in Clarke’s chapter. What some consider to be noise, can be an important source of information, a perspective voiced by Clarke in the chapter title ‘Information’. Clarke states that “The noise of timbre does not physically corrupt but, rather, informatically enhances the sound it inhabits (see Kahn 2002).”

So perhaps, in an increasingly digital world, it is necessary to evaluate the way we see the world, as there are biases that emerge when we conceptualize the world from the perspective of technology. In many circumstances, there is a profound beauty in the disruption of patterns, unexpectedness and what some may consider meaningless noise. When we conflate binary representations with information, we risk losing a sense of other invaluable sources of information that are not always captured by the digitization process. And these additional sources of information can be quite valuable, informing artistic exploration, like the role of feedback in 1960’s rock music, which exposed the world to an array of evocative sounds.