Persons and affiliations

Date

Description

This paper examines the role of visualizations in scientific and artistic experimental practices in machine learning dealing with multi-dimensional data space. Based on my own ethnographic research, I present two ongoing case studies, one of which is situated in an interdisciplinary laboratory of data scientists and biologists in Berlin, Germany, where research in the social structure of honeybee colonies is conducted. My second case study is concerned with the studio practice of a German media artist who uses generative adversarial networks (GANs) to generate visual materials for his video works.
Both cases converge in that in each of them, practitioners seek to generate meaningful information by way of algorithmic operations in high-dimensional vector space (Mackenzie 2017). Both practices are characterized by the central relevance of visualizations of the results of machine learning processes. Here, plots and other media of making visible the structure of data sets feature as media of translation without which sense-making of the data would seem impossible. Data sets and predictions based on them are read through visualization mechanisms using which the relatedness of data becomes intelligible in the first place. In that sense, I will argue based on my field observations, visualizations of vector space are a fundamental element of translational elations-making in algorithmic practices.
The focus of my analysis is on the infrastructure of such translational relations-making. I look at how scientific and artistic knowledge, algorithmic architectures, division of labor, institutionalized expectations, and imaginaries play into the results of the practices discussed, and what is at stake in them.