EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Moral stereotyping in large language models

Aliah Zewail iD, Alexandra Figueroa iD, Jesse Graham iD, Mohammad Atari iD

Proceedings of the National Academy of Sciences · 10.03.2026 · 123 (10)

DOI: 10.1073/pnas.2519941123 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 04.03.2026. Verfasst von Aliah Zewail und 3 weiteren Autor:innen. Beteiligte Einrichtung: Department of Psychological and Brain Sciences, University of Massachusetts Amherst, Amherst, MA 01003. Hauptthemen laut MEDLINE: Large Language Models, Morals, Stereotyping. Die Arbeit wurde bisher 21-mal zitiert (Crossref) und nennt 87 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Science Foundation (NSF).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Can Large Language Models (LLMs) accurately estimate various societies’ moral values? Here, we query the perceptions of LLMs regarding the moral norms of the “average” person from 48 nations and compare them to a large-scale ( n = 90 , 802 ) survey of six moral values (Care, Equality, Proportionality, Loyalty, Authority, and Purity) from those populations. Our findings indicate that LLMs poorly capture the moral diversity around the globe, systematically overestimating some moral values (particularly Care) and underestimating others (especially Purity). Notably, examining various versions of Generative Pre-trained Transformer (GPT) shows that these LLMs may overestimate the overall moral concerns of some Western countries (e.g., the United States, Canada, and Australia) while underestimating those of non-Western countries (e.g., Nigeria, Morocco, and Indonesia). Our work demonstrates that LLMs are inaccurate generators of cross-cultural estimations in the moral domain; in other words, they stereotype the moral values of non-Western populations in predictable ways. Our results highlight the ethical and epistemic risks of relying on LLMs to estimate the endorsement of moral values around the globe.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Large Language ModelsMoralsStereotypingHumansCross-Cultural ComparisonGenerative Artificial Intelligence

Stichwörter der Autor:innen

AIculturelarge language modelsmorality

Autor:innen und Institutionen

  • Department of Psychological and Brain Sciences, University of Massachusetts Amherst, Amherst, MA 01003.
  • Haas School of Business, University of California, Berkeley, CA 94720.
  • Department of Management, David Eccles School of Business, University of Utah, Salt Lake City, UT 84112.

ORCID-Profile: Aliah Zewail, Alexandra Figueroa, Jesse Graham, Mohammad Atari

Förderung

  • National Science Foundation (NSF)

Zitieren

APA

Zewail, A., Figueroa, A., Graham, J., & Atari, M. (2026). Moral stereotyping in large language models. Proceedings of the National Academy of Sciences, 123(10). https://doi.org/10.1073/pnas.2519941123

Vancouver

Zewail A, Figueroa A, Graham J, Atari M. Moral stereotyping in large language models. Proceedings of the National Academy of Sciences. 2026;123(10). doi:10.1073/pnas.2519941123

BibTeX
@article{Zewail2026,
  author = {Zewail, Aliah and Figueroa, Alexandra and Graham, Jesse and Atari, Mohammad},
  title = {Moral stereotyping in large language models},
  journal = {Proceedings of the National Academy of Sciences},
  year = {2026},
  volume = {123},
  number = {10},
  doi = {10.1073/pnas.2519941123}
}
Kostenlos starten →