The authors explore the implications of model multiplicity – the phenomenon in the development of machine learning models that produces several model specifications for a given task that differ in various ways but deliver equal accuracy.

This paper examines concerns about big data’s disparate impact risk from the perspective of American antidiscrimination law, more specifically, through Title VII’s prohibition of discrimination in employment. The paper also calls out the legal and political difficulties of addressing and remedying this type of discrimination, in particular, the tension between the two major theories underlying antidiscrimination law: anticlassification and antisubordination.