Embracing Data Noise

How can systems be designed and created with and for noise?

Noise is a helpful concept in fields like data science, machine learning, and artificial intelligence. It can help make data manageable, for example by allowing “noisy” data points to be removed so the data can be streamlined to fit a computational structure. But there is always a tension when people’s reality must be represented for a computational system to deal with: Computer systems operate with explicit definitions and discrete structures, but human existence does not lend itself to strict boundaries or clear definitions. This presents us with choices that involve noise. For example, what input will we be expecting and what remaining potential input will, consequently, be noise that lies outside what we plan for the system to acknowledge? What constitutes valid input, and what are the consequences of deciding that something is “invalid”?

This paper discusses examples that involve conceptualization, acceptance, and use of noise; including what may be gained from viewing seemingly undesirable output as noise with potential. It ends with reflections on what it could mean to embrace noise when designing computer systems.