Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
41 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
41 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

What's in a Name? Auditing Large Language Models for Race and Gender Bias (2402.14875v2)

Published 21 Feb 2024 in cs.CL, cs.AI, cs.CY, and cs.LG

Abstract: We employ an audit design to investigate biases in state-of-the-art LLMs, including GPT-4. In our study, we prompt the models for advice involving a named individual across a variety of scenarios, such as during car purchase negotiations or election outcome predictions. We find that the advice systematically disadvantages names that are commonly associated with racial minorities and women. Names associated with Black women receive the least advantageous outcomes. The biases are consistent across 42 prompt templates and several models, indicating a systemic issue rather than isolated incidents. While providing numerical, decision-relevant anchors in the prompt can successfully counteract the biases, qualitative details have inconsistent effects and may even increase disparities. Our findings underscore the importance of conducting audits at the point of LLM deployment and implementation to mitigate their potential for harm against marginalized communities.

Evaluation of Race and Gender Bias in LLMs

The paper "What's in a Name? Auditing LLMs for Race and Gender Bias" by Amit Haim, Alejandro Salinas, and Julian Nyarko, presents a meticulous exploration of biases embedded within state-of-the-art LLMs. By employing a systemic audit, the researchers examined how LLMs respond to scenario prompts involving individuals with names that are strongly associated with different racial and gender identities. Importantly, the paper investigates the correlation between names and biased model outputs, specifically focusing on names significant within the U.S. context as proxies for race and gender.

Methodological Insights

To uncover implicit biases, the authors conducted extensive prompting experiments across several prevalent LLMs, including GPT-4, GPT-3.5, and Google's PaLM-2. Their methodology encompassed 42 distinct prompt templates spread over five central scenarios: purchasing decisions, chess competitions, public office elections, sports rankings, and initial hiring offers. The prompts incorporated names recognized by their racial and gender associations, thereby simulating real-world advisory use cases that LLMs might encounter.

The design of the prompts was structured around three levels of context: low, high, and numeric. This differentiation allowed the paper to probe how context influences the expression of bias by the models. The researchers applied a quantitative approach to analyze the LLM responses, ensuring consistency and agility in calculating disparities. The extensive dataset, encompassing 168,000 responses, enabled a robust statistical analysis of biases.

Numerical Findings

The results elucidate significant disparities that align with racial and gender stereotypes. For instance, in scenarios involving financial transactions like the purchase of a car or bicycle, names conventionally identified with Black individuals were suggested lower initial offers compared to those associated with White individuals. Similarly, male-associated names were generally afforded more beneficial outcomes than female-associated names. The effect persisted across variations of prompts in certain contexts, revealing an ingrained bias that resisted simple mitigation through qualitative contextual information.

Notably, the introduction of numerical anchors in prompts often nullified these disparities, indicating a potential pathway for reducing biases during LLM deployment. The researchers found no significant difference in bias levels between different model iterations or their training quality, suggesting these biases might be a prevalent attribute of training data or architectural limitations rather than specific to a particular model type.

Practical and Theoretical Implications

The evidence from this paper underscores the systematic bias embedded in LLMs that can influence model outputs in consequential and disparate ways. From a practical standpoint, organizations deploying LLMs in environments that utilize individual names for personalization must recognize the potential for biased outputs to propagate, possibly through policies like personalized customer interaction via LLMs. Such recognition is especially crucial as commercial applications of LLMs proliferate.

Adopting mitigation strategies such as embedding numerical context, enhancing audit processes, and revisiting training datasets to filter bias propagation channels in LLMs becomes critical. This research pushes for naming-based audits to be integrated as standard practice at both the deployment and implementation phases of LLM utilization, representing a necessary step for ensuring fairness and equity in AI technologies.

Future Directions

This work opens avenues for more specialized research into mitigation techniques that extend beyond numerical anchoring, making qualitative insights also effective. Regulatory bodies and policymakers could engage with results from such studies to establish frameworks that standardize LLM auditing practices in key sectors like financial services, legal systems, and other domains of high societal importance. While this paper navigated the biases within a U.S.-centric perspective, future studies could scale this research internationally to identify localized biases in diverse global contexts.

In conclusion, while LLMs hold significant potential in transforming numerous industries, their fair and equitable application hinges on our ability to understand and mitigate inherent biases. Thus, the work by Haim, Salinas, and Nyarko contributes valuable expertise to the discourse on ethical responsibilities in developing and deploying AI technologies.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (32)
  1. Ryan S. Baker and Aaron Hawn “Algorithmic Bias in Education” In International Journal of Artificial Intelligence in Education 32, 2022, pp. 1052–1092 DOI: 10.1007/s40593-021-00285-9
  2. J.R. Bent “Is algorithmic affirmative action legal” In Georgetown Law Journal 108, 2019, pp. 803
  3. “Are Emily and Greg More Employable Than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination” In American Economic Review 94.4, 2004, pp. 991–1013 DOI: 10.1257/0002828042002561
  4. Aylin Caliskan, Joanna J. Bryson and Arvind Narayanan “Semantics derived automatically from language corpora contain human-like biases” In Science 356.6334, 2017, pp. 183–186 DOI: 10.1126/science.aal42
  5. Anupam Chander “The racist algorithm” In Michigan Law Review 115, 2016, pp. 1023
  6. “Companies Go All Out to Up Their Generative AI Game”, 2023 URL: https://www.bloomberg.com/news/articles/2023-07-31/how-companies-are-tackling-the-challenges-of-generative-ai
  7. Alexander Coppock “Avoiding Post-Treatment Bias in Audit Experiments” In Journal of Experimental Political Science 6.1, 2019, pp. 1–4 DOI: 10.1017/XPS.2018.9
  8. “The Measure and Mismeasure of Fairness” In Journal of Machine Learning Research 24, 2023, pp. 1–117 URL: http://jmlr.org/papers/volume24/22-1318/22-1318.pdf
  9. S.Michael Gaddis “How Black Are Lakisha and Jamal? Racial Perceptions from Names Used in Correspondence Audit Studies” In Sociological Science 4 Society for Sociological Science, 2017, pp. 469–489 DOI: 10.15195/v4.a19
  10. T.B. Gillis “The input fallacy” In Minnesota Law Review 106, 2022, pp. 1175
  11. Steven N. Goodman, Sharad Goel and Mark R. Cullen “Machine learning, health disparities, and causal reasoning” In Annals of Internal Medicine 169, 2018, pp. 883–884 DOI: 10.7326/M18-1463
  12. Daniel E. Ho and Albert Xiang “Affirmative algorithms: the legal grounds for fairness as awareness” In University of Chicago Law Review Online, 2020, pp. 134–154
  13. Aziz Z. Huq “Racial equity in algorithmic criminal justice” In Duke Law Journal 68, 2019, pp. 1043–1134
  14. Dominik K. Kanbach “The GenAI Is out of the Bottle: Generative Artificial Intelligence from a Business Model Innovation Perspective” Last visited Feb 9, 2024 In Rev Manag Sci, 2023 DOI: 10.1007/s11846-023-00696-z
  15. “Investigating Bias in Facial Analysis Systems: A Systematic Review” In IEEE Access 8, 2020, pp. 130751–130761 DOI: 10.1109/ACCESS.2020.3006051
  16. Paul T. Kim “Race-aware algorithms: fairness, nondiscrimination and affirmative action” In California Law Review 110, 2022, pp. 1539
  17. René F. Kizilcec and Hansol Lee “Algorithmic fairness in education” In The Ethics of Artificial Intelligence in Education Routledge, 2022, pp. 29
  18. Allison Koenecke, Andrew Nam and Emily Lake “Racial disparities in automated speech recognition” In Proceedings of the National Academy of Sciences 117.14, 2020, pp. 7684–7689 DOI: 10.1073/pnas.1915768117
  19. Hadas Kotek, Rikker Dockum and David Q. Sun “Gender bias and stereotypes in Large Language Models”, 2023 arXiv: https://arxiv.org/abs/2308.14921
  20. Sandra G. Mayson “Bias in, bias out” In Yale Law Journal 128, 2019, pp. 2218–2300
  21. “Ethical limitations of algorithmic fairness solutions in health care machine learning” In Lancet Digital Health 2, 2020, pp. e221–e223 DOI: 10.1016/S2589-7500(20)30062-1
  22. “Efficient Estimation of Word Representations in Vector Space”, 2013 arXiv: https://arxiv.org/abs/1301.3781
  23. “Dissecting racial bias in an algorithm used to manage the health of populations” In Science 366, 2019, pp. 447–453 DOI: 10.1126/science.aax2342
  24. “By Any Other Name?: On Being “Regarded As” Black, and Why Title VII Should Apply Even If Lakisha and Jamal Are White” In Wisconsin Law Review, 2005, pp. 1283 URL: https://scholarship.law.bu.edu/faculty_scholarship/314
  25. Devah Pager “The Use of Field Experiments for Studies of Employment Discrimination: Contributions, Critiques, and Directions for the Future” In The Annals of the American Academy of Political and Social Science 609.1 Sage Publications, 2007, pp. 104–133 DOI: 10.1177/0002716206294796
  26. Stephen R. Pfohl, Agata Foryciarz and Nigam H. Shah “An empirical characterization of fair machine learning for clinical risk prediction” In Journal of Biomedical Informatics 113, 2021, pp. 103621 DOI: 10.1016/j.jbi.2020.103621
  27. “The Woman Worked as a Babysitter: On Biases in Language Generation” In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) Hong Kong, China: Association for Computational Linguistics, 2019, pp. 3407–3412 DOI: 10.18653/v1/D19-1339
  28. Briana Vecchione, Karen Levy and Solon Barocas “Algorithmic Auditing and Social Justice: Lessons from the History of Audit Studies” In EAAMO ’21: Equity and Access in Algorithms, Mechanisms, and Optimization –, NY, USA: ACM, 2021 DOI: 10.1145/3465416.3483294
  29. “Are Emily and Greg Still More Employable than Lakisha and Jamal? Investigating Algorithmic Hiring Bias in the Era of ChatGPT”, 2023 arXiv: https://ar5iv.org/abs/2310.05135
  30. ““Kelly is a Warm Person, Joseph is a Role Model”: Gender Biases in LLM-Generated Reference Letters”, 2023 arXiv: https://arxiv.org/abs/2310.09219
  31. Crystal S. Yang and Will Dobbie “Equal protection under algorithms: a new statistical and legal framework” In Michigan Law Review 119, 2020, pp. 291
  32. John Yinger “Testing for Discrimination in Housing and Related Markets” In National Report Card on Discrimination in America: The Role of Testing Urban Institute, 1998, pp. 27
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Amit Haim (1 paper)
  2. Alejandro Salinas (2 papers)
  3. Julian Nyarko (11 papers)
Citations (21)