Canons of Algorithmic Inference: Feminist Theoretical Virtues in Machine Learning by Gabbrielle M. Johnson (New York University)
Date: 3:30pm PST November 15, 2019 Location: J.R. Howard Hall
J.R. Howard Hall
As inductive decision-making procedures, the inferences made by machine learning programs are subject to underdetermination by evidence and bear inductive risk. Previous attempts to address these challenges have been guided by the presumption that machine learning processes can and should be formally objective. In doing so, the influence of values has been restricted to data and decision outcomes, thereby ignoring internal value-laden choice points. In this paper, I argue that these efforts rest on a mistake: the resources required to respond to these challenges render machine learning processes essentially value-laden, and thus sanction ethical and socio-historical interventions throughout their production, use, and evaluation. I demonstrate these points in the case of recidivism algorithms, arguing that the adoption of feminist theoretical virtues supports the use of false positive equality as a measure of fairness in order to stymie the ongoing harm to the black community within the criminal justice system.