{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/9753b8b67b3444a1a51f18cb3a6f7a81\" frameborder=\"0\" width=\"1280\" height=\"960\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":960,"width":1280,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":960,"thumbnail_width":1280,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/9753b8b67b3444a1a51f18cb3a6f7a81-1600940023578.gif","duration":313,"title":"&quot;To See is to Stereotype&quot; - CSCW 2020 - Barlas, Kyriakou, Guest, Kleanthous, Otterbacher","description":"PREPRINT:&nbsp;https://zenodo.org/record/4028263\"To \"See\" is to Stereotype: Image Tagging Algorithms, Gender Recognition, and the Accuracy-Fairness Trade-off\" (Paper 6033) - Pinar Barlas, Kyriakos Kyriakou, Olivia Guest, Styliani Kleanthous, Jahna Otterbacher -&nbsp;CSCW 2020Abstract:Machine-learned computer vision algorithms for tagging images are increasingly used by developers and&nbsp;researchers, having become popularized as easy-to-use “cognitive services.” Yet these tools struggle with&nbsp;gender recognition, particularly when processing images of women, people of color and non-binary individuals.&nbsp;Socio-technical researchers have cited data bias as a key problem; training datasets often over-represent&nbsp;images of people and contexts that convey social stereotypes. The social psychology literature explains&nbsp;that people learn social stereotypes, in part, by observing others in particular roles and contexts, and can&nbsp;inadvertently learn to associate gender with scenes, occupations and activities. Thus, we study the extent to&nbsp;which image tagging algorithms mimic this phenomenon. We design a controlled experiment, to examine&nbsp;the interdependence between algorithmic recognition of context and the depicted person’s gender. In the&nbsp;spirit of auditing to understand machine behaviors, we create a highly controlled dataset of people images,&nbsp;imposed on gender-stereotyped backgrounds. Our methodology is reproducible and our code publicly available.&nbsp;Evaluating five proprietary algorithms, we find that in three, gender inference is hindered when a background&nbsp;is introduced. Of the two that “see” both backgrounds and gender, it is the one whose output is most consistent&nbsp;with human stereotyping processes that is superior in recognizing gender. We discuss the accuracy – fairness&nbsp;trade-off, as well as the importance of auditing black boxes in better understanding this double-edged sword."}