HMI DAIS 02 - Lily Hu, Computer Science Harvard University
HMI DAIS 02 - Lily Hu, Computer Science Harvard University
Public online seminar, 9am 16 July 2020 AEST
Lily Hu of Harvard University gave the second HMI Data, AI and Society public seminar.
Lily Hu is a PhD candidate in Applied Mathematics and Philosophy at Harvard University. She works on topics in machine learning theory, algorithmic fairness, and philosophy of (social) science, and political philosophy. Her current work focuses on causal inference methodology in the social sciences and is especially interested in how various statistical frameworks treat and measure the "causal effect" of social categories such as race, and ultimately, how such methods are seen to back normative claims about racial discrimination and inequalities broadly.
She holds an A.B. in Mathematics from Harvard College and taught English and Spanish history in Madrid on a Fulbright Fellowship. Her current work is supported by an NSF Graduate Research Fellowship and a fellowship from the Jain Family Institute. Lily discussed ‘What is "Race" in Algorithmic Discrimination on the Basis of Race?’
Abstract
Machine learning algorithms bring out an under-appreciated problem of discrimination, namely, figuring when a decision made on the basis of a factor correlated with race is a decision made on the basis of race. I argue that adopting a constructivist metaphysics of race can answer this puzzle in a principled manner. First, I suggest that the stubborn statistical correlations between race and various social outcomes that make algorithms such a thorny site of potential racial discrimination, teach us something about the social nature of racial groups and what it means to be raced. On what I call a “thick constructivist” account of race, to be “Black” is to be socially positioned in the way indicated by a certain set of statistical regularities on the basis of a set of superficial phenotypic traits. A thick constructivist sees that acting on the basis of correlations that constitute race qua social position just is acting on the basis of race, because races just are social positions that subject their member individuals to the matrix of privileging and subordinating social relations that define what it is to be raced. Thus, on this view, there may be little conceptual space between acting on certain correlations with race—that is, using machine learning tools to leverage these statistical facts for prediction purposes—and acting on something that is race "itself.” This conclusion has considerable ramifications for our understanding of discrimination, algorithmic and beyond.
HMI Dais recordings can be viewed here.