Algorithmic Expertise: Explanation and Repair in the Making of Similarity in Art

February 12, 2019

Sarah Sachs, Columbia University, Department of Sociology

Research Presentation: Algorithmic Expertise: Explanation and Repair in the Making of Similarity in Art

Today’s firms face an unprecedented number of opportunities to capitalize off of “big data.” Increasing volumes and types of analyzable data have driven demand for the labor necessary to support the technologies used to analyze and process this data. The repair work required by data analytic technologies and algorithmic systems is expanding the work practices and roles necessary to bring such technologies to market. My work is concerned with the forms of expertise that accompany this repair work and how they shape, and are shaped by, existing structures of organization and work.

I focus on the data classification practices of a team of knowledge workers—art experts—whose responsibility is to make art image data legible to a machine and to make a similarity matching algorithm’s output legible to those with knowledge about art. In short, they are to create and maintain the algorithm’s, and thus the firm’s, cultural credibility.

In this talk, I will present findings from an ethnography of this team. I will briefly describe the interactional process that grounds the team’s data classification and annotation practices and discuss how these nontechnical experts repair divergent output from the algorithm. I will show how, in doing so, the team draws on evolving practical theories about how the algorithm works. I will then theorize how, over time, some team members develop ease in deploying theories in practice, and therefore become recognized for their algorithmic expertise. Finally, I will briefly discuss some implications of this form of expertise, particularly for collaboration and conflict on the team.