I.M. (Inga) Luchs, MA

Data Discrimination: Rethinking Systems of Classification Beyond Homophily

Machine learning algorithms permeate our everyday life. They are used to discriminate against spam emails, to sort results in search engines and to recommend content. They find application in the detection of credit card fraud and in predictive policing. My PhD project seeks to investigate the key technical principles of machine learning in order to uncover underlying assumptions and beliefs. Following recent research I contend that, fundamentally, machine learning algorithms operate based on similarity. Consequently, segregation emerges in online networks, resulting not only in the creation of echo chambers but also in discriminatory effects. My research project explicitly deals with this socio-technological condition.

I will conduct a critical review of machine learning literature, as well as an auto-ethnographic participation in online introductory classes and open-source platforms in form of ‘walk-throughs’. The focus will be on classification algorithms and their application in recommender systems. Building on this critical technological understanding, the second part of my project revolves around the question of alternative algorithms: How can we imagine forms of classification that go beyond this homophilic principle? And what if we used this knowledge to design alternative systems that are not discriminating along the lines of race, class, and gender?