Supporting Human-AI Collaboration in Auditing LLMs with LLMs
Open accessThe DOI matches its Crossref record; 24 of the 24 references it cites were found in a registry or an index.
Abstract
Large language models (LLMs) are increasingly becoming all-powerful and pervasive via deployment in sociotechnical systems. Yet these language models, be it for classification or generation, have been shown to be biased, behave irresponsibly, causing harm to people at scale. It is crucial to audit these language models rigorously before deployment. Existing auditing tools use either or both humans and AI to find failures. In this work, we draw upon literature in human-AI collaboration and sensemaking, and interview research experts in safe and fair AI, to build upon the auditing tool: AdaTest [36], which is powered by a generative LLM. Through the design process we highlight the importance of sensemaking and human-AI communication to leverage complementary strengths of humans and generative models in collaborative auditing. To evaluate the effectiveness of AdaTest++, the augmented tool, we conduct user studies with participants auditing two commercial language models: OpenAI’s GPT-3 and Azure’s sentiment analysis model. Qualitative analysis shows that AdaTest++ effectively leverages human strengths such as schematization, hypothesis testing. Further, with our tool, users identified a variety of failures modes, covering 26 different topics over 2 tasks, that have been shown in formal audits and also those previously under-reported.
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
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676
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 1×
“Raji et al. (2020) introduce a framework for algorithm auditing to be applied throughout the algorithm’s internal development lifecycle.”
MethodPrinciples of mixed-initiative user interfaces
Horvitz, E. (1999) · Proceedings of the SIGCHI conference on Human factors in computing systems the CHI is the limit - CHI '99 · doi:10.1145/302979.303030 · Cited by 1,490
Why Johnny Can’t Prompt: How Non-AI Experts Try (and Fail) to Design LLM Prompts
Zamfirescu-Pereira, J. D., Wong, R. Y., Hartmann, B., & Yang, Q. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581388 · Cited by 1,274 · cited 3×
“This may be explained by users’ lack of knowledge and confidence in the abilities of LLMs, and further exacerbated by the brittleness of prompt-based interactions (Zamfirescu-Pereira et al., 2023).”
BackgroundMethodInfluentialMcDonald, 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 derived from an iterative process (McDonald et al., 2019).”
BackgroundExplainable artificial intelligence: A survey
Dosilovic, F. K., Brcic, M., & Hlupic, N. (2018) · 2018 41st International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO) · doi:10.23919/mipro.2018.8400040 · Cited by 1,147
Power to the People: The Role of Humans in Interactive Machine Learning
Amershi, S., Cakmak, M., Knox, W. B., & Kulesza, T. (2014) · AI Magazine · doi:10.1609/aimag.v35i4.2513 · Cited by 1,084
Self-citationLanguage (Technology) is Power: A Critical Survey of “Bias” in NLP
Blodgett, S. L., Barocas, S., Daumé III, H., & Wallach, H. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.485 · Cited by 1,050
Beyond accuracy: behavioral testing of NLP models with CheckList
Ribeiro, M. T., Wu, T., Guestrin, C., & Singh, S. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.442 · Cited by 890
Self-citationRaji, 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
Red Teaming Language Models with Language Models
Perez, E., Huang, S., Song, F., Cai, T., Ring, R., Aslanides, J., Glaese, A., McAleese, N., & Irving, G. (2022) · Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2022.emnlp-main.225 · Cited by 363
Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction
Shelby, R., Rismani, S., Henne, K., Moon, A., Rostamzadeh, N., Nicholas, P., Yilla-Akbari, N., Gallegos, J., Smart, A., Garcia, E., et al. (2023) · Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3600211.3604673 · Cited by 279
Dynabench: Rethinking Benchmarking in NLP
Kiela, D., Bartolo, M., Nie, Y., Kaushik, D., Geiger, A., Wu, Z., Vidgen, B., Prasad, G., Singh, A., Ringshia, P., et al. (2021) · Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/2021.naacl-main.324 · Cited by 254
Problematic 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 3×
“Recent works (DeVos et al., 2022; Cabrera et al., 2022; Shen et al., 2021) have operationalized this model to analyse human-driven auditing.”
BackgroundGoldstein, J. A., Sastry, G., Musser, M., DiResta, R., Gentzel, M., & Sedova, K. (2023) · arXiv · doi:10.48550/arxiv.2301.04246 · Cited by 152
Beat 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×
“In response, past research has considered crowdsourcing (Kiela et al., 2021b; Kaushik et al., 2021; Attenberg et al., 2015) and end-user bug reporting (Lam et al., 2022) to audit algorithms.”
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
DeVos, 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
“I’m sorry to hear that”: Finding new biases in language models with a holistic descriptor dataset
Smith, E. M., Hall, M., Kambadur, M., Presani, E., & Williams, A. (2022) · Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2022.emnlp-main.625 · Cited by 97 · cited 1×
“Participants were building on previous work that identified 13 axes of identity-based bias (Smith et al., 2022) such as ability, age, body type, etc.”
MethodCapturing Failures of Large Language Models Via Human Cognitive Biases
Jones, E., & Steinhardt, J. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-0856 · Cited by 48
Capturing Failures of Large Language Models via Human Cognitive Biases
Jones, E., & Steinhardt, J. (2022) · arXiv · doi:10.48550/arxiv.2202.12299 · Cited by 43
Cited by
The Metacognitive Demands and Opportunities of Generative AI
Tankelevitch, L., Kewenig, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., & Rintel, S. (2024) · Proceedings of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642902 · Cited by 493
“Future work should characterize the role of metacognitive flexibility in adapting one’s workflows across different GenAI interfaces and interaction modes.”
BackgroundBias and ethics of AI systems applied in auditing - A systematic review
Murikah, W., Nthenge, J. K., & Musyoka, F. M. (2024) · Scientific African · doi:10.1016/j.sciaf.2024.e02281 · Cited by 148
Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Casper, S., Davies, X., Shi, C., Gilbert, T. K., Scheurer, J., Rando, J., Freedman, R., Korbak, T., Lindner, D., Freire, P., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.15217 · Cited by 101
“Red-teaming can also be complemented by interpretability techniques [Rastogi et al., 2023, Räuker et al., 2023], especially for purposes of identifying adversarial inputs [Ziegler et al., 2022, Casper et al., 2023c,a] or anomalous inputs [Pang et al., 2021].”
BackgroundFarsight: Fostering Responsible AI Awareness During AI Application Prototyping
Wang, Z. J., Kulkarni, C., Wilcox, L., Terry, M., & Madaio, M. (2024) · Proceedings of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642335 · Cited by 55
Beyond the Chat: Executable and Verifiable Text-Editing with LLMs
Laban, P., Vig, J., Hearst, M., Xiong, C., & Wu, C.-S. (2024) · Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3654777.3676419 · Cited by 31
“In turn, recent work has included these methodologies into systems, allowing users to receive feedback on the factuality of generated text, allowing users to handle AI-generated text with more caution [27, 35, 37, 38, 60, 77].”
Metadata
- Authors
- Rastogi, C., Tulio Ribeiro, M., King, N., Nori, H., & Amershi, S.
- Venue
- Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society
- Publisher
- ACM
- Published
- 2023-08-08
- Pages
- 913-926
- ISSN
- 3065-8365 (ISSN-L)
- OpenAlex
- W4386246835
- Cited by
- 78
- References
- 24
- FWCI
- 20.12
- Type
- conference paper
- Language
- en
- Open access
- hybrid
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.