Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Large language models associate Muslims with violence

Nature Machine Intelligence volume 3, pages 461–463 (2021)Cite this article

Subjects

Large language models, which are increasingly used in AI applications, display undesirable stereotypes such as persistent associations between Muslims and violence. New approaches are needed to systematically reduce the harmful bias of language models in deployment.

Natural language processing (NLP) research has seen substantial progress on a variety of applications through the use of large pretrained language models1,2,3,4. Although these increasingly sophisticated language models are capable of generating complex and cohesive natural language, a series of recent works demonstrate that they also learn undesired social biases that can perpetuate harmful stereotypes5,6,7,8,9.

This is a preview of subscription content, access via your institution

Relevant articles

Open Access articles citing this article.

Access options

Rent or buy this article

Get just this article for as long as you need it

$39.95

Learn more

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: GPT-3 exhibits Muslim–violence bias.
Fig. 2: Debiasing GPT-3 completions.

References

  1. Mikolov, T., Chen, K., Corrado, G. & Dean, J. in Proc. International Conference on Learning Representations (ICLR, 2013).

  2. Dai, A. M. & Le, Q. V. in Advances in Neural Information Processing Systems Vol. 28, 3079–3087 (NeurIPS, 2015).

  3. Brown, T. et al. in Advances in Neural Information Processing Systems Vol. 33, 1877–1901 (NeurIPS, 2020).

  4. Kitaev, N., Kaiser, L. & Levskaya, A. in Proc. International Conference on Learning Representations (ICLR, 2020).

  5. Bolukbasi, T., Chang, K.-W., Zou, J. Y., Saligrama, V. & Kalai, A. T. in Advances in Neural Information Processing Systems Vol. 29, 4349–4357 (NeurIPS, 2016).

  6. Nadeem, M., Bethke, A. & Reddy, S. Preprint at https://arxiv.org/abs/2004.09456 (2020).

  7. Sheng, E., Chang, K.-W., Natarajan, P. & Peng, N. in Proc. Conference on Empirical Methods in Natural Language Processing 3407–3412 (ACL, 2019).

  8. Bordia, S. & Bowman, S. R. in Proc. Conference of the North American Chapter of the Association for Computational Linguistics (ACL, 2019).

  9. Lu, K., Mardziel, P., Wu, F., Amancharla, P. & Datta, A. in Logic, Language, and Security (eds Nigam, V. et al.) 189–202 (Springer, 2020).

  10. Lewis, M. et al. in Proc. 58th Annual Meeting of the Association for Computational Linguistics 7871–7880 (ACL, 2020).

  11. Wallace, E., Feng, S., Kandpal, N., Gardner, M. & Singh, S. in Proc. Conference on Empirical Methods in Natural Language Processing (EMNLP) 2153–2162 (ACL, 2019).

  12. Qian, Y., Muaz, U., Zhang, B. & Hyun, J. W. Preprint at https://arxiv.org/abs/1905.12801 (2019).

  13. Bender, E. M., Gebru, T., McMillan-Major, A. & Mitchell, S. in ACM Conference on Fairness, Accountability, and Transparency 610–623 (ACM, 2021).

  14. Li, X. L. & Liang, P. Preprint at https://arxiv.org/abs/2101.00190 (2021).

Download references

Acknowledgements

We thank A. Abid, A. Abdalla, D. Khan, and M. Ghassemi for the helpful feedback on the manuscript and experiments. J.Z. is supported by NSF CAREER 1942926.

Author information

Authors and Affiliations

  1. Department of Electrical Engineering, Stanford University, Stanford, CA, USA

    Abubakar Abid

  2. Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Ontario, Canada

    Maheen Farooqi

  3. Department of Biomedical Data Science, Stanford University, Stanford, CA, USA

    James Zou

Authors
  1. Abubakar Abid
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Maheen Farooqi
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. James Zou
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to James Zou.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Peer review information Nature Machine Intelligence thanks Arvind Narayaran for their contribution to the peer review of this work.

Supplementary information

Supplementary Information

Supplementary discussions A–C

Rights and permissions

Reprints and Permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Abid, A., Farooqi, M. & Zou, J. Large language models associate Muslims with violence. Nat Mach Intell 3, 461–463 (2021). https://doi.org/10.1038/s42256-021-00359-2

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s42256-021-00359-2

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing