Elamur Research⌕
☰
home / papers / 2023 / Cultural Bias and Cultural Alignment of Large Language Models

Cultural Bias and Cultural Alignment of Large Language Models

Paper · 2023 · Yan Tao, Olga Viberg, Ryan S. Baker et al.

Summary

LLMs learn mostly from English-language text, so they may carry the values of a narrow set of societies into the writing people produce with them. The authors audit five OpenAI models released between 2020 and 2024: GPT-3 (text-davinci-002), GPT-3.5-turbo, GPT-4, GPT-4-turbo and GPT-4o. They ask each model the ten Integrated Values Surveys (IVS, which merges WVS and EVS) items behind the Inglehart-Welzel Cultural Map. To reduce sensitivity to wording, they use ten paraphrased respondent descriptors. The answers are placed on the map's two axes: survival vs. self-expression and traditional vs. secular-rational. Bias is the Euclidean distance between a model's position and survey positions for 107 countries/territories. By default, every model sits near English-speaking and Protestant European societies and far from African-Islamic ones, with a consistent lean toward self-expression values. Cultural prompting tells the model it was born and lives in a given country. For GPT-4, GPT-4-turbo and GPT-4o, this moves outputs closer to the survey benchmark for 71–81% of countries. The method is a cheap, language-independent mitigation, and the authors argue it should be paired with ongoing auditing.

Key points

  • Benchmark: The latest IVS data covers 95 WVS countries/territories (waves 2005–2022) plus 17 from EVS, for 112 in total. Five were dropped because of missing answers, leaving 107.
  • GPT-4o distances: Without cultural prompting, GPT-4o is closest to Finland (d = 0.20), Andorra (0.21) and the Netherlands (0.45). It is farthest from Jordan (4.10), Libya (4.00) and Ghana (3.95). Jordan is the most distant country for each of the four newer models.
  • Secular-traditional axis: Models differ on this axis, with no trend over time. GPT-3.5-turbo and GPT-4o lean secular, GPT-4-turbo leans traditional, and GPT-3 and GPT-4 sit near the global average.
  • Cultural prompting: The prompt template is "born in [country] and living in [country]". It improves alignment for 71–81% of countries/territories with the three GPT-4-family models.
  • Practical caveats:
    • GPT-3 was tested with only one prompt variant because it was deprecated before testing finished.
    • GPT-3.5-turbo refused some questions: homosexuality in 2 of 1070 cases and abortion in 30 of 1070. Its culturally prompted result for Libya was therefore excluded.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of PennsylvaniaPhiladelphia · United States233 works · 65,583 citations🎓Cornell UniversityIthaca · United States134 works · 65,159 citations🎓KTH Royal Institute of TechnologyStockholm · Sweden20 works · 1,723 citations

 Fields, topics and concepts

📚 Shelves 8The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksPsychology · 1,649 worksSafety Research · 626 worksSocial Psychology · 330 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksBias, fairness & safety · 587 works

Cite

Cite as Yan Tao, Olga Viberg, Ryan S. Baker and Rene F. Kizilcec. 2023. Cultural Bias and Cultural Alignment of Large Language Models. arXiv:2311.14096 · PNAS Nexus, Volume 3, Issue 9, September 2024, pgae346.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1093/pnasnexus/pgae346 · arXiv 2311.14096

@misc{tao2023cultural,
  title = {{Cultural Bias and Cultural Alignment of Large Language Models}},
  author = {Yan Tao, Olga Viberg, Ryan S. Baker and Rene F. Kizilcec},
  year = {2023},
  doi = {10.1093/pnasnexus/pgae346},
  eprint = {2311.14096},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.1093/pnasnexus/pgae346}
}

References

50 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Suhaib Abdurahman, Mohammad Atari, Farzan Karimi-Malekabadi, Mona J Xue, Jackson Trager, Peter S Park, Preni Golazizian, Ali Omrani, and Morteza Dehghani. Perils and opportunities in using large language models in psychological research. Preprint at https://osf.io/preprints/psyarxiv/d695y(2023). ↗
  2. Abdulrahman Essa Al Lily, Abdelrahim Fathy Ismail, Fathi M Abunaser, Firass Al-Lami, and Ali Khalifa Atwa Abdullatif. Chatgpt and the rise of semi-humans. Humanities and Social Sciences Communications, 10(1):626, 2023. find
  3. Arnav Arora, Lucie-Aimée Kaffee, and Isabelle Augenstein. Probing pre-trained language models for cross-cultural differences in values. In Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), pages 114–130, 2023.
  4. Mohammad Atari, Mona J Xue, Peter S Park, Damián Blasi, and Joseph Henrich. Which humans? Preprint at https://doi.org/10.31234/osf.io/5b26t(2023). — [doi:10.31234/osf.io/5b26t(2023)](https://doi.org/10.31234/osf.io/5b26t%282023%29) ↗
  5. Diego Aycinena, Lucas Rentschler, Benjamin Beranek, and Jonathan F Schulz. Social norms and dishonesty across societies. Proceedings of the National Academy of Sciences, 119(31):e2120138119, 2022. find
  6. Solon Barocas, Anhong Guo, Ece Kamar, Jacquelyn Krones, Meredith Ringel Morris, Jennifer Wortman Vaughan, W Duncan Wadsworth, and Hanna Wallach. Designing disaggregated evaluations of ai systems: Choices, considerations, and tradeoffs. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, an… find
  7. Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. Advances in Neural Information Processing Systems, 29, 2016.
  8. Nicholas Buttrick. Studying large language models as compression algorithms for human culture. Trends in Cognitive Sciences, 28(3):187–189, 2024. find
  9. Yong Cao, Li Zhou, Seolhwa Lee, Laura Cabello, Min Chen, and Daniel Hershcovich. Assessing cross-cultural alignment between chatgpt and human societies: An empirical study. In Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP), pages 53–67, 2023.
  10. Shih-Yi Chien, Michael Lewis, Katia Sycara, Jyi-Shane Liu, and Asiye Kumru. The effect of culture on trust in automation: reliability and workload. ACM Transactions on Interactive Intelligent Systems, 8(4):1–31, 2018. find
  11. Incheol Choi, Richard E Nisbett, and Ara Norenzayan. Causal attribution across cultures: Variation and universality. Psychological bulletin, 125(1):47–63, 1999. find
  12. Hannah Faye Chua, Julie E Boland, and Richard E Nisbett. Cultural variation in eye movements during scene perception. Proceedings of the national academy of sciences, 102(35):12629–12633, 2005. find
  13. Luigi De Angelis, Francesco Baglivo, Guglielmo Arzilli, Gaetano Pierpaolo Privitera, Paolo Ferragina, Alberto Eugenio Tozzi, and Caterina Rizzo. Chatgpt and the rise of large language models: the new ai-driven infodemic threat in public health. Frontiers in Public Health, 11:1166120, 2023. find
  14. Dorottya Demszky, Diyi Yang, David S Yeager, Christopher J Bryan, Margarett Clapper, Susannah Chandhok, Johannes C Eichstaedt, Cameron Hecht, Jeremy Jamieson, Meghann Johnson, et al. Using large language models in psychology. Nature Reviews Psychology, 2(11):688–701, 2023. find
  15. EVS. Evs trend file 1981-2017, za7503 data file version 3.0.0. GESIS Data Archive https://doi.org/10. 4232/1.14021 (2022). find
  16. Emilio Ferrara. Should chatgpt be biased? challenges and risks of bias in large language models. First Monday 28; https://doi.org/10.5210/fm.v28i11.13346(2023). ↗
  17. Michael C Frank. Baby steps in evaluating the capacities of large language models. Nature Reviews Psychology, 2(8):451–452, 2023. find
  18. Susan A Gelman and Steven O Roberts. How language shapes the cultural inheritance of categories. Proceedings of the National Academy of Sciences, 114(30):7900–7907, 2017. find
  19. Amir Goldberg, Sameer B Srivastava, V Govind Manian, William Monroe, and Christopher Potts. Fitting in or standing out? the tradeoffs of structural and cultural embeddedness. American Sociological Review, 81(6):1190–1222, 2016. find
  20. Andrea L Guzman and Seth C Lewis. Artificial intelligence and communication: A human–machine communication research agenda. New Media & Society, 22(1):70–86, 2020. find
  21. Christian Haerpfer, R. Inglehart, A. Moreno, C. Welzel, K. Kizilova, J. Diez-Medrano, M. Lagos, P. Norris, E. Ponarin, B. Puranen, and et al. World values survey trend file (1981-2022) cross-national data-set, data file version 3.0.0. Madrid, Spain & Vienna, Austria: JD Systems Institute & WVSA Secr… — [doi:10.14281/18241.23](https://doi.org/10.14281/18241.23) find
  22. Jeffrey T Hancock, Mor Naaman, and Karen Levy. Ai-mediated communication: Definition, research agenda, and ethical considerations. Journal of Computer-Mediated Communication, 25(1):89–100, 2020. find
  23. Geert Hofstede. Culture’s consequences: Comparing values, behaviors, institutions and organizations across nations. sage, 2001. find
  24. Geert Hofstede and Robert R McCrae. Personality and culture revisited: Linking traits and dimensions of culture. Cross-cultural research, 38(1):52–88, 2004. find
  25. Jess Hohenstein, Rene F Kizilcec, Dominic DiFranzo, Zhila Aghajari, Hannah Mieczkowski, Karen Levy, Mor Naaman, Jeffrey Hancock, and Malte F Jung. Artificial intelligence in communication impacts language and social relationships. Scientific Reports, 13(1):5487, 2023. find
  26. Tim Hornyak. Why japan is building its own version of chatgpt. Nature; https://doi.org/10.1038/ d41586-023-02868-z (2023). find
  27. Ronald Inglehart and Wayne E Baker. Modernization, cultural change, and the persistence of traditional values. American sociological review, 65(1):19–51, 2000. find
  28. Ronald Inglehart and Christian Welzel. Modernization, cultural change, and democracy: The human development sequence, volume 333. Cambridge university press, 2005. find
  29. Maurice Jakesch, Advait Bhat, Daniel Buschek, Lior Zalmanson, and Mor Naaman. Co-writing with opinionated language models affects users’ views. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, pages 1–15, 2023. find
  30. Li-Jun Ji, Kaiping Peng, and Richard E Nisbett. Culture, control, and perception of relationships in the environment. Journal of personality and social psychology, 78(5):943–955, 2000. find
  31. Rebecca L Johnson, Giada Pistilli, Natalia Menédez-González, Leslye Denisse Dias Duran, Enrico Panai, Julija Kalpokiene, and Donald Jay Bertulfo. The ghost in the machine has an american accent: value conflict in gpt-3. Preprint at https://doi.org/10.48550/arXiv.2203.07785 (2022). ↗
  32. Enkelejda Kasneci, Kathrin Seßler, Stefan Küchemann, Maria Bannert, Daryna Dementieva, Frank Fischer, Urs Gasser, Georg Groh, Stephan Günnemann, Eyke Hüllermeier, et al. Chatgpt for good? on opportunities and challenges of large language models for education. Learning and individual differences,…
  33. Ani Koubaa. Gpt-4 vs. gpt-3.5: A concise showdown. preprints202303.0422.v1 (2023). Preprints at https://doi.org/10.20944/ find
  34. Yerin Kwak and Zachary A Pardos. Bridging large language model disparities: Skill tagging of multilingual educational content. British Journal of Educational Technology, 2024. find
  35. Johan P Mackenbach. Cultural values and population health: a quantitative analysis of variations in cultural values, health behaviours and health outcomes among 42 european countries. Health & place, 28:116–132, 2014. find
  36. Tarek Naous, Michael J Ryan, and Wei Xu. Having beer after prayer? measuring cultural bias in large language models. Preprint at https://doi.org/10.48550/arXiv.2305.14456 (2023). ↗
  37. Roberto Navigli, Simone Conia, and Björn Ross. Biases in large language models: Origins, inventory and discussion. ACM Journal of Data and Information Quality, 15(2):1–21, 2023. find
  38. Richard E Nisbett and Yuri Miyamoto. The influence of culture: holistic versus analytic perception. Trends in cognitive sciences, 9(10):467–473, 2005. find
  39. Daphna Oyserman and Spike WS Lee. Does culture influence what and how we think? effects of priming individualism and collectivism. Psychological bulletin, 134(2):311–342, 2008. find
  40. Kaiping Peng and Richard E Nisbett. Culture, dialectics, and reasoning about contradiction. American psychologist, 54(9):741, 1999. find
  41. Aida Ramezani and Yang Xu. Knowledge of cultural moral norms in large language models. In Proceedings of the 61st Annual meeting of the Association for Computational Linguistics, pages 428–446, 2023.
  42. Christian Sandvig, Kevin Hamilton, Karrie Karahalios, and Cedric Langbort. Auditing algorithms: Research methods for detecting discrimination on internet platforms. Data and discrimination: converting critical concerns into productive inquiry, 22(2014):4349–4357, 2014. find
  43. Edgar H Schein. What is culture. In Reframing organizational culture, pages 243–253. Sage Publications, Inc, 1991. find
  44. Richard Shiffrin and Melanie Mitchell. Probing the psychology of ai models. Proceedings of the National Academy of Sciences, 120(10):e2300963120, 2023. find
  45. Arun James Thirunavukarasu, Darren Shu Jeng Ting, Kabilan Elangovan, Laura Gutierrez, Ting Fang Tan, and Daniel Shu Wei Ting. Large language models in medicine. Nature medicine, 29(8):1930–1940, 2023. find
  46. Bill Thompson, Simon Kirby, and Kenny Smith. Culture shapes the evolution of cognition. Proceedings of the National Academy of Sciences, 113(16):4530–4535, 2016. find
  47. Mu Tian, Ping Deng, Yingying Zhang, and Maria Paz Salmador. How does culture influence innovation? a systematic literature review. Management Decision, 56(5):1088–1107, 2018. find
  48. Olga Viberg, Rene F Kizilcec, Ioana Jivet, Alejandra Martı́nez Monés, Alice Oh, Chantal Mutimukwe, Stefan Hrastinski, and Maren Scheffel. Cultural differences in students’ privacy concerns in learning analytics across germany, south korea, spain, sweden, and the united states. Computers in human be… find
  49. Tianyu Wu, Shizhu He, Jingping Liu, Siqi Sun, Kang Liu, Qing-Long Han, and Yang Tang. A brief overview of chatgpt: The history, status quo and potential future development. IEEE/CAA Journal of Automatica Sinica, 10(5):1122–1136, 2023. find
  50. Ann Yuan, Andy Coenen, Emily Reif, and Daphne Ippolito. Wordcraft: story writing with large language models. In 27th International Conference on Intelligent User Interfaces, pages 841–852, 2022. ID A008 Survey Question Feeling of Happiness A165 Trust on People E018 Respect for Authority E025 Petitio… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tao-2023-cultural-bias-cultural-alignment-large/