Artificial Intelligence, Intersectionality and Social Justice

Topic Description

Large language models (LLMs) and other AI systems are now widely used to generate, evaluate and moderate content, but they can reproduce and amplify social inequalities in ways that are often subtle. Recent work shows that LLMs can hold covert biases, for example against speakers of particular dialects, even when they appear unbiased on the surface. It also shows that judgements of what counts as harmful depend on who is doing the judging, and that current models and benchmarks are skewed towards Western, English-speaking contexts. Intersectional harms such as misogynoir (hate directed at Black women) are especially hard to capture, because they come from the combination of identities rather than any single one, and because fairness research has tended to reduce intersectionality to comparing metrics across demographic subgroups. The aim of this project is to investigate computational approaches that better detect, evaluate and mitigate these socially and culturally situated harms. This may include intersectional hate-speech detection, the role of annotator identity and disagreement in human and LLM-based evaluation, or bias and safety in multilingual language settings. It could lead to new models, datasets or evaluation frameworks contributing to responsible, ethical, impactful uses of AI.

Skills Required:

A good degree (2:1 or above) in Computing, Computer Science, Artificial Intelligence, Data Science or a related discipline, with experience in machine learning and/or natural language processing. An interest in intersectionality, social justice or online harms would be advantageous.

Background Reading:

Crenshaw, K. (1989). Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics. University of Chicago Legal Forum, 1989(1), 139–167.

Kwarteng, J., Coppolino Perfumi, S., Farrell, T., Third, A. & Fernandez, M. (2022). Misogynoir: Challenges in Detecting Intersectional Hate. Social Network Analysis and Mining, 12, 166. https://doi.org/10.1007/s13278-022-00993-7

Kwarteng, J., Farrell, T., Third, A. & Fernandez, M. (2023). Annotators' Perspectives: Exploring the Influence of Identity on Interpreting Misogynoir. In: Proceedings of ASONAM 2023.

Sap, M., Swayamdipta, S., Vianna, L., Zhou, X., Choi, Y. & Smith, N. A. (2022). Annotators with Attitudes: How Annotator Beliefs and Identities Bias Toxic Language Detection. Proceedings of NAACL 2022. https://doi.org/10.18653/v1/2022.naacl-main.431

Blodgett, S. L., Barocas, S., Daumé III, H. & Wallach, H. (2020). Language (Technology) is Power: A Critical Survey of "Bias" in NLP. Proceedings of ACL 2020. https://aclanthology.org/2020.acl-main.485/

Hofmann, V., Kalluri, P. R., Jurafsky, D. & King, S. (2024). AI Generates Covertly Racist Decisions about People Based on Their Dialect. Nature, 633, 147–154. https://doi.org/10.1038/s41586-024-07856-5

Kirk, H. R., Whitefield, A., Röttger, P., et al. (2024). The PRISM Alignment Dataset: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Models. NeurIPS 2024 Datasets and Benchmarks Track. https://doi.org/10.52202/079017-3342

Gallegos, I. O., Rossi, R. A., Barrow, J., et al. (2024). Bias and Fairness in Large Language Models: A Survey. Computational Linguistics, 50(3). https://doi.org/10.1162/coli_a_00524

Adilazuarda, M. F., Mukherjee, S., Lavania, P., et al. (2024). Towards Measuring and Modeling "Culture" in LLMs: A Survey. Proceedings of EMNLP 2024, 15763–15784. https://doi.org/10.18653/v1/2024.emnlp-main.882

Ovalle, A., Subramonian, A., Gautam, V., Gee, G. & Chang, K.-W. (2023). Factoring the Matrix of Domination: A Critical Review and Reimagination of Intersectionality in AI Fairness. Proceedings of AIES 2023, 496–511. https://doi.org/10.1145/3600211.3604705

Muhammad, S. H., Abdulmumin, I., Ayele, A. A., et al. (2025). AfriHate: A Multilingual Collection of Hate Speech and Abusive Language Datasets for African Languages. Proceedings of NAACL 2025, 1854–1871. https://doi.org/10.18653/v1/2025.naacl-long.92

Contact