Algorithmic Gender Prediction Is Illegitimate, but Gender Imputation Can Yield Valid Measurements

Abstract

Machine learning ethics researchers, critical HCI scholars, and interdisciplinary academics have argued that algorithmically predicting gender is wrong. At the same time, others in the same research communities rely on predicted gender labels to study gender disparities and develop algorithmic fairness techniques. Taking both of these perspectives in good faith pushes us to believe that algorithmic gender prediction simultaneously cannot be trusted and at the same time can identify genuine gender disparities. How do we reconcile these two seemingly contradictory intuitions? We draw a distinction between two questions: whether algorithmic gender prediction is wrong, being illegitimate and thereby contributing to harm; and whether algorithmic gender prediction is wrong, being invalid and thereby producing unusable measurements. Our analysis translates different arguments against gender prediction into these terms of legitimacy and validity and shows how gender imputation applied for fairness purposes can be both illegitimate and still yield valid disparity measurements. We clarify this bind by drawing upon transfeminist literature to distinguish two forms of sexism: sexism that targets women and femininity, and sexism that targets transgender and nonbinary people. While gender imputation can produce valid measurements for the former, it is illegitimate and harmful for the latter. We argue that practitioners should deploy gender imputation only when it would achieve anti-discrimination benefits that cannot be achieved through other reasonable means, while harms are minimized to the extent possible. We examine how this tension plays out through three case studies: auditing gender bias in generative image models, measuring gender disparities in film, and imputing gender from personal names. By disentangling both questions of legitimacy from questions of validity, as well as these two different forms of sexism, we show how debates over gender prediction have often conflated distinct concerns, obscuring both the settings in which gender imputation can support fairness efforts and the harms towards transgender and nonbinary people that it fundamentally cannot capture. We conclude by recommending the development of more inclusive methods that address all kinds of sexism.