End-User Audits: A System Empowering Communities to Lead Large-Scale Investigations of Harmful Algorithmic Behavior
The DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.
Abstract
Because algorithm audits are conducted by technical experts, audits are necessarily limited to the hypotheses that experts think to test. End users hold the promise to expand this purview, as they inhabit spaces and witness algorithmic impacts that auditors do not. In pursuit of this goal, we propose end-user audits-system-scale audits led by non-technical users-and present an approach that scaffolds end users in hypothesis generation, evidence identification, and results communication. Today, performing a system-scale audit requires substantial user effort to label thousands of system outputs, so we introduce a collaborative filtering technique that leverages the algorithmic system's own disaggregated training data to project from a small number of end user labels onto the full test set. Our end-user auditing tool, IndieLabel, employs these predicted labels so that users can rapidly explore where their opinions diverge from the algorithmic system's outputs. By highlighting topic areas where the system is under-performing for the user and surfacing sets of likely error cases, the tool guides the user in authoring an audit report. In an evaluation of end-user audits on a popular comment toxicity model with 17 non-technical participants, participants both replicated issues that formal audits had previously identified and also raised previously underreported issues such as under-flagging on veiled forms of hate that perpetuate stigma and over-flagging of slurs that have been reclaimed by marginalized communities.
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
Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
Reimers, N., & Gurevych, I. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1410 · Cited by 12,778 · cited 1×
“We used Sentence Transformers [60] and BERTopic [27] to generate topic clusters in an unsupervised manner.”
MethodConstructing Grounded Theory. A Practical Guide Through Qualitative Analysis
(2006) · QMiP Bulletin · doi:10.53841/bpsqmip.2006.1.2.36 · Cited by 11,828
Crenshaw, K. (2018) · Feminist Legal Theory · doi:10.4324/9780429500480-5 · Cited by 9,510
Dissecting racial bias in an algorithm used to manage the health of populations
Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019) · Science · doi:10.1126/science.aax2342 · Cited by 7,137 · cited 2×
“Given that algorithmic systems are often opaque and commercially owned, audit-based approaches have proven to be incredibly impactful, revealing harmful algorithmic behavior in domains such as housing [2, 21], employment [13, 14, 72], healthcare [55], search [49, 54, 61], and technologies such as facial recognition [9…”
BackgroundA survey on semi-supervised learning
van Engelen, J. E., & Hoos, H. H. (2020) · Machine Learning · doi:10.1007/s10994-019-05855-6 · Cited by 2,666
Raji, I. D., Smart, A., White, R. N., Mitchell, M., Gebru, T., Hutchinson, B., Smith-Loud, J., Theron, D., & Barnes, P. (2020) · Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency · doi:10.1145/3351095.3372873 · Cited by 2,245 · cited 2×
“If system developers host their own end-user audits, users could directly share their reports with the developers in a closed setting; this would bear similarities to a first- or second-party audit [59, 72].”
BackgroundMcDonald, N., Schoenebeck, S., & Forte, A. (2019) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3359174 · Cited by 1,206 · cited 1×
“Here, our goal was to summarize the high-level themes that emerged from our participants, so the codes were our process, not our product [48].”
BackgroundGorwa, R., Binns, R., & Katzenbach, C. (2020) · Big Data & Society · doi:10.1177/2053951719897945 · Cited by 1,097 · cited 1×
“A core issue with algorithmic moderation is that language is complex and highly context-dependent [25], so systems can fail when, for instance, models incorrectly learn to associate markers of dialects like African American English with higher toxicity scores [64], or users adversarially mask harmful comments with min…”
BackgroundRacial Discrimination in the Sharing Economy: Evidence from a Field Experiment
Edelman, B. G., Luca, M., & Svirsky, D. (2015) · SSRN Electronic Journal · doi:10.2139/ssrn.2701902 · Cited by 999
The Risk of Racial Bias in Hate Speech Detection
Sap, M., Card, D., Gabriel, S., Choi, Y., & Smith, N. A. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1163 · Cited by 817 · cited 2×
“Another major documented issue is that the Perspective model displays false positive bias for AAE (African American English) dialect examples [64] and for a host of demographic identity terms that were over-represented among the toxic comments in the training dataset [20].”
BackgroundImproving Fairness in Machine Learning Systems: What Do Industry Practitioners Need?
Holstein, K., Wortman Vaughan, J., Daumé, H., Dudik, M., & Wallach, H. (2019) · Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3290605.3300830 · Cited by 777 · cited 4×
“In situations where users are largely aligned in their views, the case for mitigation is clear, which can help to combat the struggles that system developers often face in deciding where to focus their efforts and how to trade off against various fairness concerns [33].”
BackgroundInfluentialRacial disparities in automated speech recognition
Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., Toups, C., Rickford, J. R., Jurafsky, D., & Goel, S. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1915768117 · Cited by 765
Measuring and Mitigating Unintended Bias in Text Classification
Dixon, L., Li, J., Sorensen, J., Thain, N., & Vasserman, L. (2018) · Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3278721.3278729 · Cited by 674 · cited 1×
“Another major documented issue is that the Perspective model displays false positive bias for AAE (African American English) dialect examples [64] and for a host of demographic identity terms that were over-represented among the toxic comments in the training dataset [20].”
BackgroundRaji, I. D., & Buolamwini, J. (2019) · Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3306618.3314244 · Cited by 651
Where fairness fails: data, algorithms, and the limits of antidiscrimination discourse
Hoffmann, A. L. (2019) · Information, Communication & Society · doi:10.1080/1369118x.2019.1573912 · Cited by 535 · cited 1×
“In the context of algorithmic fairness analyses, research has noted that dominant single-axis approaches (accounting for a single demographic attribute such as gender or race) may be inadequate to measure and counter discrimination [32, 39].”
BackgroundMeasuring price discrimination and steering on e-commerce web sites
Hannak, A., Soeller, G., Lazer, D., Mislove, A., & Wilson, C. (2014) · Proceedings of the 2014 Conference on Internet Measurement Conference · doi:10.1145/2663716.2663744 · Cited by 405 · cited 1×
“One line of crowdsourced audits involves passive data collection from opted-in users, for example, through browser extensions or automated scripts that gather data from users’ machines [17, 30, 51, 61, 67].”
BackgroundTowards a critical race methodology in algorithmic fairness
Hanna, A., Denton, R., Smart, A., & Smith-Loud, J. (2020) · Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency · doi:10.1145/3351095.3372826 · Cited by 295 · cited 1×
“Group-based fairness approaches tend to abstract away the complexities and hierarchies of identity (especially for categories like race that are socially constructed and contested) and tend to treat oppressed groups as interchangeable [29].”
BackgroundWho Audits the Auditors? Recommendations from a field scan of the algorithmic auditing ecosystem
Costanza-Chock, S., Raji, I. D., & Buolamwini, J. (2022) · 2022 ACM Conference on Fairness Accountability and Transparency · doi:10.1145/3531146.3533213 · Cited by 277 · cited 4×
“Moreover, stakeholders and marginalized groups are likely underrepresented among expert auditors [16].”
BackgroundInfluentialAuditing Partisan Audience Bias within Google Search
Robertson, R. E., Jiang, S., Joseph, K., Friedland, L., Lazer, D., & Wilson, C. (2018) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3274417 · Cited by 209
The algorithm audit: Scoring the algorithms that score us
Brown, S., Davidovic, J., & Hasan, A. (2021) · Big Data & Society · doi:10.1177/2053951720983865 · Cited by 204 · cited 1×
“While there is broad acknowledgement that stakeholder involvement should be a critical part of algorithm audits, such involvement is rare in practice [8, 16].”
BackgroundProblematic Machine Behavior: A Systematic Literature Review of Algorithm Audits
Bandy, J. (2021) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3449148 · Cited by 193
Raji, I. D., Smart, A., White, R. N., Mitchell, M., Gebru, T., Hutchinson, B., Smith-Loud, J., Theron, D., & Barnes, P. (2020) · arXiv · doi:10.48550/arxiv.2001.00973 · Cited by 179
Shen, H., DeVos, A., Eslami, M., & Holstein, K. (2021) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3479577 · Cited by 171 · cited 5×
“End users, on the other hand, do possess this rich situated knowledge of the particular impacts that algorithmic systems have on their own communities, and they already identify problematic behaviors of algorithmic systems through their everyday interactions with such systems [66].”
BackgroundResultInfluentialAssessing the Fairness of AI Systems: AI Practitioners' Processes, Challenges, and Needs for Support
Madaio, M., Egede, L., Subramonyam, H., Wortman Vaughan, J., & Wallach, H. (2022) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3512899 · Cited by 156
Investigating the Impact of Gender on Rank in Resume Search Engines
Chen, L., Ma, R., Hannák, A., & Wilson, C. (2018) · Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3173574.3174225 · Cited by 155 · cited 2×
“Given that algorithmic systems are often opaque and commercially owned, audit-based approaches have proven to be incredibly impactful, revealing harmful algorithmic behavior in domains such as housing [2, 21], employment [13, 14, 72], healthcare [55], search [49, 54, 61], and technologies such as facial recognition [9…”
BackgroundBuilding and Auditing Fair Algorithms: A Case Study in Candidate Screening
Wilson, C., Ghosh, A., Jiang, S., Mislove, A., Baker, L., Szary, J., Trindel, K., & Polli, F. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445928 · Cited by 133 · cited 4×
“Given that algorithmic systems are often opaque and commercially owned, audit-based approaches have proven to be incredibly impactful, revealing harmful algorithmic behavior in domains such as housing [2, 21], employment [13, 14, 72], healthcare [55], search [49, 54, 61], and technologies such as facial recognition [9…”
BackgroundInfluentialBeat the Machine: Challenging Humans to Find a Predictive Model's “Unknown Unknowns”
Attenberg, J., Ipeirotis, P., & Provost, F. (2015) · Journal of Data and Information Quality · doi:10.1145/2700832 · Cited by 103 · cited 1×
“Similarly, Beat the Machine introduced game-like incentives to encourage users to brainstorm examples that would trigger “unknown unknown” errors in predictive models [3].”
BackgroundThe Disagreement Deconvolution: Bringing Machine Learning Performance Metrics In Line With Reality
Gordon, M. L., Zhou, K., Patel, K., Hashimoto, T., & Bernstein, M. S. (2021) · Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3411764.3445423 · Cited by 101
Self-citationDeVos, A., Dhabalia, A., Shen, H., Holstein, K., & Eslami, M. (2022) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3491102.3517441 · Cited by 98
Crowd-assisted search for price discrimination in e-commerce: first results
Mikians, J., Gyarmati, L., Erramilli, V., & Laoutaris, N. (2013) · Proceedings of the ninth ACM conference on Emerging networking experiments and technologies · doi:10.1145/2535372.2535415 · Cited by 82
Smith, C. E., Yu, B., Srivastava, A., Halfaker, A., Terveen, L., & Zhu, H. (2020) · Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3313831.3376783 · Cited by 77
Understanding international perceptions of the severity of harmful content online
Jiang, J. A., Scheuerman, M. K., Fiesler, C., & Brubaker, J. R. (2021) · PLOS ONE · doi:10.1371/journal.pone.0256762 · Cited by 69
Designing Toxic Content Classification for a Diversity of Perspectives
Kumar, D., Kelley, P. G., Consolvo, S., Mason, J., Bursztein, E., Durumeric, Z., Thomas, K., & Bailey, M. (2021) · arXiv · doi:10.48550/arxiv.2106.04511 · Cited by 28
Cited by
Deng, W. H., Guo, B., Devrio, A., Shen, H., Eslami, M., & Holstein, K. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581026 · Cited by 103
Supporting Human-AI Collaboration in Auditing LLMs with LLMs
Rastogi, C., Ribeiro, M. T., King, N., Nori, H., & Amershi, S. (2023) · arXiv · doi:10.48550/arxiv.2304.09991 · Cited by 79
The Foundation Model Transparency Index
Bommasani, R., Klyman, K., Longpre, S., Kapoor, S., Maslej, N., Xiong, B., Zhang, D., & Liang, P. (2023) · arXiv · doi:10.48550/arxiv.2310.12941 · Cited by 46
Turri, V., Morrison, K., Robinson, K.-M., Abidi, C., Perer, A., Forlizzi, J., & Dzombak, R. (2024) · The 2024 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3630106.3658985 · Cited by 10
Human-Centered Evaluation and Auditing of Language Models
Liu, Y. L., Deng, W. H., Lam, M. S., Eslami, M., Kim, J., Liao, Q. V., Xu, W., Novikova, J., & Xiao, Z. (2025) · Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706599.3706729 · Cited by 7
“…the perspectives and lived experiences of diverse stakeholders in AI evaluation and auditing, there’s growing interest among both HCI and AI researchers in directly involving these stakeholders in the evaluation and auditing of LLMs (as well as other AI systems) [3, 13, 23, 26, 41, 42, 45].”
Self-citation
Metadata
- Authors
- Lam, M. S., Gordon, M. L., Metaxa, D., Hancock, J. T., Landay, J. A., & Bernstein, M. S.
- Venue
- Proceedings of the ACM on Human-Computer Interaction
- Published
- 2022-11-07
- Volume
- 6
- Issue
- CSCW2
- Pages
- 1-34
- ISSN
- 2573-0142 (ISSN-L)
- OpenAlex
- W4309618902
- Cited by
- 77
- References
- 33
- FWCI
- 15.07
- Type
- journal article
- Language
- en
- Open access
- closed
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.