To "See" is to Stereotype: Image Tagging Algorithms, Gender Recognition, and the Accuracy-Fairness Trade-off
Open accessThe DOI matches its Crossref record; 27 of the 27 references it cites were found in a registry or an index.
Abstract
Machine-learned computer vision algorithms for tagging images are increasingly used by developers and researchers, having become popularized as easy-to-use "cognitive services." Yet these tools struggle with gender recognition, particularly when processing images of women, people of color and non-binary individuals. Socio-technical researchers have cited data bias as a key problem; training datasets often over-represent images of people and contexts that convey social stereotypes. The social psychology literature explains that people learn social stereotypes, in part, by observing others in particular roles and contexts, and can inadvertently learn to associate gender with scenes, occupations and activities. Thus, we study the extent to which image tagging algorithms mimic this phenomenon. We design a controlled experiment, to examine the interdependence between algorithmic recognition of context and the depicted person's gender. In the spirit of auditing to understand machine behaviors, we create a highly controlled dataset of people images, imposed on gender-stereotyped backgrounds. Our methodology is reproducible and our code publicly available. Evaluating five proprietary algorithms, we find that in three, gender inference is hindered when a background is introduced. Of the two that "see" both backgrounds and gender, it is the one whose output is most consistent with human stereotyping processes that is superior in recognizing gender. We discuss the accuracy--fairness trade-off, as well as the importance of auditing black boxes in better understanding this double-edged sword.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Universal dimensions of social cognition: warmth and competence
Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2007) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2006.11.005 · Cited by 4,567 · cited 1×
“For example, it is widely accepted that socially desirable traits for men reflect competence, while traits for women reflect warmth [23, 40, 96].”
BackgroundHow the machine “thinks”: Understanding opacity in machine learning algorithms
Burrell, J. (2016) · Big Data & Society · doi:10.1177/2053951715622512 · Cited by 3,196 · cited 1×
“, interpretable from a technical point of view); users/developers must also have the skills necessary to understand them [16].”
BackgroundEagly, A. H., & Wood, W. (2012) · Handbook of Theories of Social Psychology · doi:10.4135/9781446249222.n49 · Cited by 1,853 · cited 1×
“Socialization facilitates performances, by encouraging individuals of a given gender to develop certain personality traits and skills [35, 96].”
BackgroundThe Chicago face database: a free stimulus set of faces and norming data
Ma, D. S., Correll, J., & Wittenbrink, B. (2015) · Behavior Research Methods · doi:10.3758/s13428-014-0532-5 · Cited by 1,788
Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016) · arXiv · doi:10.48550/arxiv.1607.06520 · Cited by 1,615 · cited 1×
“Co-occurrences reported in text often do not respond to their frequency in the offline world, as people may over-or under-report something unusual or counter-stereotypical (e.g., a “male nurse”) [13].”
BackgroundFairness definitions explained
Verma, S., & Rubin, J. (2018) · Proceedings of the International Workshop on Software Fairness · doi:10.1145/3194770.3194776 · Cited by 1,124 · cited 1×
“However, researchers have emphasized that in the context of socio-technical systems, there is never a single interpretation on “what is fair” [78, 102].”
BackgroundInherent trade-offs in the fair determination of risk scores
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2017) · Schloss Dagstuhl – Leibniz-Zentrum für Informatik · doi:10.4230/lipics.itcs.2017.43 · Cited by 943 · cited 3×
“[63] discuss, as multiple constraints are near impossible to satisfy simultaneously.”
BackgroundInfluentialInherent Trade-Offs in the Fair Determination of Risk Scores
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016) · arXiv · doi:10.48550/arxiv.1609.05807 · Cited by 941
Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints
Zhao, J., Wang, T., Yatskar, M., Ordonez, V., & Chang, K.-W. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1323 · Cited by 829 · cited 1×
“, “shopping” or “cooking” on images of women; “golfing” or “snowboarding” on images depicting men) [50, 108].”
What We Instagram: A First Analysis of Instagram Photo Content and User Types
Hu, Y., Manikonda, L., & Kambhampati, S. (2014) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v8i1.14578 · Cited by 807
Discrimination in online ad delivery
Sweeney, L. (2013) · Communications of the ACM · doi:10.1145/2447976.2447990 · Cited by 783 · cited 1×
“Sweeney [100], in a test for racial bias in Google AdSense, conducted Web searches on names, manipulating them by their racial associations.”
BackgroundAccountability in algorithmic decision making
Diakopoulos, N. (2016) · Communications of the ACM · doi:10.1145/2844110 · Cited by 763
Language use in intergroup contexts: The linguistic intergroup bias.
Maass, A., Salvi, D., Arcuri, L., & Semin, G. R. (1989) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.57.6.981 · Cited by 654
Face Recognition Performance: Role of Demographic Information
Klare, B. F., Burge, M. J., Klontz, J. C., Vorder Bruegge, R. W., & Jain, A. K. (2012) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2012.2214212 · Cited by 516 · cited 1×
“Previous research has shown that such inferences may appear accurate for men with light skin, but that the error rates rise dramatically when analyzing images depicting other social groups, and in particular, women and/or people with dark skin [15, 44, 60, 62].”
BackgroundUnequal Representation and Gender Stereotypes in Image Search Results for Occupations
Kay, M., Matuszek, C., & Munson, S. A. (2015) · Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems · doi:10.1145/2702123.2702520 · Cited by 509 · cited 2×
“Kay et al. [57], Magno et al. [73], and Otterbacher et al. [82] all submitted queries to search engines to study the perpetuation of gender stereotypes in image search engines.”
BackgroundThe Misgendering Machines: Trans/HCI Implications of Automatic Gender Recognition
Keyes, O. (2018) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3274357 · Cited by 505
Sweeney, L. (2013) · Queue · doi:10.1145/2460276.2460278 · Cited by 477
Algorithmic Bias in Autonomous Systems
Danks, D., & London, A. J. (2017) · Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2017/654 · Cited by 472 · cited 1×
“, if developers use a statistically biased estimator in the model [25].”
MethodWomen Also Snowboard: Overcoming Bias in Captioning Models
Hendricks, L. A., Burns, K., Saenko, K., Darrell, T., & Rohrbach, A. (2018) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-01219-9_47 · Cited by 404
UX Design Innovation: Challenges for Working with Machine Learning as a Design Material
Dove, G., Halskov, K., Forlizzi, J., & Zimmerman, J. (2017) · Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3025453.3025739 · Cited by 391 · cited 1×
“For example, researchers studying the use of ML by HCI students [85] and UX designers [31] reported in both groups an enthusiasm for incorporating ML into designs and prototype systems.”
BackgroundOn the apparent conflict between individual and group fairness
Binns, R. (2020) · Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency · doi:10.1145/3351095.3372864 · Cited by 312 · cited 1×
“In contrast, individual fairness concerns the notion of consistency, holding that similar individuals should receive similar treatment by the system [9].”
Bias in Online Freelance Marketplaces: Evidence from TaskRabbit and Fiverr
Hannák, A., Wagner, C., Garcia, D., Mislove, A., Strohmaier, M., & Wilson, C. (2017) · Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing · doi:10.1145/2998181.2998327 · Cited by 286 · cited 1×
“Hannák and colleagues [47] apply both within- and between-platform auditing approaches, in their study of social biases in online marketplaces.”
BackgroundLarge image datasets: A pyrrhic win for computer vision?
Birhane, A., & Prabhu, V. U. (2021) · 2021 IEEE Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv48630.2021.00158 · Cited by 257 · cited 1×
“As such, auditing ITAs, both in terms of their training data [88] as well as their output, ethically evaluates the role such services play.”
BackgroundScheuerman, M. K., Paul, J. M., & Brubaker, J. R. (2019) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3359246 · Cited by 252
Quantifying Search Bias: Investigating Sources of Bias for Political Searches in Social Media
Kulshrestha, J., Eslami, M., Messias, J., Zafar, M. B., Ghosh, S., Gummadi, K. P., & Karahalios, K. (2017) · Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing · doi:10.1145/2998181.2998321 · Cited by 148 · cited 1×
“[66] proposed an auditing technique that gauges information bias on Twitter, by benchmarking its behaviors against those of search engines.”
BackgroundMowshowitz, A., & Kawaguchi, A. (2005) · Information Processing & Management · doi:10.1016/j.ipm.2004.05.005 · Cited by 68 · cited 1×
“Similarly, [76] quantifies search engine bias, by collecting a fair results set, consisting of the results on a given query across multiple engines.”
BackgroundOpening the government’s black boxes: freedom of information and algorithmic accountability
Fink, K. (2018) · Information, Communication & Society · doi:10.1080/1369118x.2017.1330418 · Cited by 61 · cited 1×
“Algorithmic systems and processes are often positioned as “power brokers” that are not always held accountable for their decisions and actions [30, 38].”
Background
Cited by
Algorithmic injustice: A relational ethics approach
Birhane, A. (2021) · Patterns · doi:10.1016/j.patter.2021.100205 · Cited by 707
AI auditing: The Broken Bus on the Road to AI Accountability
Birhane, A., Steed, R., Ojewale, V., Vecchione, B., & Raji, I. D. (2024) · 2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml59370.2024.00037 · Cited by 158
Future Shock: Generative AI and the International AI Policy and Governance Crisis
Leslie, D., & Perini, A. M. (2024) · Harvard Data Science Review · doi:10.1162/99608f92.88b4cc98 · Cited by 33
The Dark Side of Dataset Scaling: Evaluating Racial Classification in Multimodal Models
Birhane, A., Dehdashtian, S., Prabhu, V., & Boddeti, V. (2024) · The 2024 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3630106.3658968 · Cited by 28
Metadata
- Authors
- Barlas, P., Kyriakou, K., Guest, O., Kleanthous, S., & Otterbacher, J.
- Venue
- Proceedings of the ACM on Human-Computer Interaction
- Published
- 2021-01-05
- Volume
- 4
- Issue
- CSCW3
- Pages
- 1-31
- ISSN
- 2573-0142 (ISSN-L)
- OpenAlex
- W3119150429
- Cited by
- 40
- References
- 27
- FWCI
- 8.96
- Type
- journal article
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.