How Computers See Gender

Abstract

Investigations of facial analysis (FA) technologies-such as facial detection and facial recognition-have been central to discussions about Artificial Intelligence's (AI) impact on human beings. Research on automatic gender recognition, the classification of gender by FA technologies, has raised potential concerns around issues of racial and gender bias. In this study, we augment past work with empirical data by conducting a systematic analysis of how gender classification and gender labeling in computer vision services operate when faced with gender diversity. We sought to understand how gender is concretely conceptualized and encoded into commercial facial analysis and image labeling technologies available today. We then conducted a two-phase study: (1) a system analysis of ten commercial FA and image labeling services and (2) an evaluation of five services using a custom dataset of diverse genders using self-labeled Instagram images. Our analysis highlights how gender is codified into both classifiers and data standards. We found that FA services performed consistently worse on transgender individuals and were universally unable to classify non-binary genders. In contrast, image labeling often presented multiple gendered concepts. We also found that user perceptions about gender performance and identity contradict the way gender performance is encoded into the computer vision infrastructure. We discuss our findings from three perspectives of gender identity (self-identity, gender performativity, and demographic identity) and how these perspectives interact across three layers: the classification infrastructure, the third-party applications that make use of that infrastructure, and the individuals who interact with that software. We employ Bowker and Star's concepts of "torque" and "residuality" to further discuss the social implications of gender classification. We conclude by outlining opportunities for creating more inclusive classification infrastructures and datasets, as well as with implications for policy.CCS Concepts: • Social and professional topics → User characteristics; Gender.

Between Subjectivity and Imposition

“…This is relevant for state management and policy, i.e., to pinpoint places where intervention or allocation of resources is needed. However, the tendency of classification practices towards the erasure of residual categories [16] can cause tension and even be harmful for individuals who remain unseen or misclassified by data-driven systems [19,71].…”

“…These systems are often expected to minimize human intervention in decision-making and thus be neutral and value-free [23,51,73]. However, previous research has shown that they may contain biases that lead to discrimination and exclusion in several domains such as credit [37], the job market [70], facial recognition systems [19,45,71], algorithmic filtering [4,62], and even advertisement [1]. Critical academic work has furthermore discussed the politics involved in data-driven systems [27,30,56] and highlighted the importance of investigating the capitalistic logics woven into them [20,26,81].…”

Gender Bias in the News: A Scalable Topic Modelling and Visualization Framework

“…Our language-based analyses rely on names to assign gender to individuals and are not well suited to study such forms of diversity. Additionally, potential harms from assigning specific labels to individuals need to be considered ( Scheuerman et al, 2019 ; Blodgett et al, 2020 ). Nevertheless, studies of diversity in the news are valuable.…”

The Ethics of Emotion in AI Systems

“…A large segment of the discipline of "affective computing," involving tracking and analyzing a variety of bio-signals like heart rate, is grounded in Basic Emotion Theory [22,83,84,86]. Models powering facial analysis technologies are designed around Ekman's thesis regarding the legibility of basic emotions through facial expression [35], and analyze large databases of human faces [18], using various AI techniques to estimate emotional expression (along with other characteristics like age and gender) [97]. These systems are increasingly commercially in areas such as human resource management, advertising [69] and education [11].…”