Elamur Research⌕
☰
home / papers / 2026 / Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?

Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?

Paper · 2026 · Anvesh Rao Vijjini, Sagar B. Manjunath and Snigdha Chaturvedi

Summary

The paper asks whether LLM agents reproduce human socio-cognitive effects of power when they role-play people of unequal status. The authors built 14 hierarchical role pairs, such as principal–teacher and justice–lawyer, spanning education, law, healthcare-like and workplace settings. They filled these roles with realistic personas drawn from PersonaHub. Six LLMs from three families then held multi-turn dialogues, and four effects were measured:

  • pronoun choice ("we" vs. "I");
  • linguistic style coordination;
  • persuasion success, as a proxy for authority bias;
  • compliance with unsafe requests.

Most models showed the expected pronoun asymmetry and style adaptation. However, coordination ran in both directions more evenly than it does between humans. Higher-status agents persuaded more often, and lower-status agents complied more often with harmful requests, which amounts to an emergent authority bias. Effects were strongest early in a dialogue, while coordination grew over turns. Explicit prompting reduced persuasion and compliance for some models but barely affected pronouns or coordination. Larger models showed weaker status effects, and preference tuning changed little. The work matters for deploying LLMs in hierarchical, high-stakes settings such as tutoring, medicine and law.

Key points

  • Personas: Taken from PersonaHub (200k+ personas), with about 10 persona pairs per role pair. Human annotators judged 96.5% of pairs as power-asymmetric (Fleiss's kappa 0.73).
  • Models: Llama 3.1-8B-Instruct, Qwen 2.5-7B-Instruct, a Phi model, Llama 3.1-70B-Instruct-Quantized, GPT-4.1 and GPT-5.
  • Simulation setup: API models ran through the Sotopia framework, and offline models were prompted directly. Conversations lasted roughly 10–15 turns.
  • Coordination metric: Follows Danescu et al. (2012), using 8 style-marker categories such as articles, auxiliary verbs and quantifiers. Compliance and persuasion were scored by a GPT-5 judge, checked by humans on 300 conversations; binary labels exceeded 80% agreement, so 2-way classification was used.
  • Released data: The simulated conversations are available at github.com/nvshrao/power-asymmetric-conversations.

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 North Carolina at Chapel HillChapel Hill · United States67 works · 16,474 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksSafety Research · 543 worksSociology and Political Science · 577 works

Cite

Cite as Anvesh Rao Vijjini, Sagar B. Manjunath and Snigdha Chaturvedi. 2026. Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.2202

@inproceedings{vijjini2026do,
  title = {{Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?}},
  author = {Anvesh Rao Vijjini, Sagar B. Manjunath and Snigdha Chaturvedi},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.2202},
  url = {https://doi.org/10.18653/v1/2026.acl-long.2202}
}

References

32 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Peter Henderson, Mark Krass, Mantas Mazeika, Daniel E Ho, and Percy Liang. 2023. Do-not-answer: A dataset for evaluating safeguards in llms. In Advances in Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track.
  2. Marah Abdin, Jyoti Aneja, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J Hewett, Mojan Javaheripi, Piero Kauffmann, and 1 others. 2024. Phi-4 technical report. arXiv preprint arXiv:2412.08905.
  3. AI@Meta. 2024. The llama 3 herd of models. https://ai.meta.com/research/publications/ the-llama-3-herd-of-models/ . ↗
  4. Solomon E Asch. 1956. Studies of independence and conformity: A minority of one against a unanimous majority. Psychological Monographs: General and Applied, 70(9):1–70. find
  5. Robert B Cialdini. 2001. Influence: Science and practice. Allyn & Bacon. find
  6. Robert B Cialdini. 2003. Influence. Influence At Work. find
  7. Rachel Cotterill. 2016. Identifying Stylometric Correlates of Social Power. Ph.D. thesis, University of Sheffield. find
  8. Cristian Niculescu-Mizil Danescu, Lillian Lee, Bo Pang, and Jon Kleinberg. 2012. Echoes of power: Language effects and power differences in social interaction. In Proceedings of the 21st international conference on World Wide Web, pages 699–708. find
  9. Jessica Maria Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, and Zexue He. 2024. Cognitive bias in decision-making with LLMs. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 12640–12653, Miami, Florida, USA. Association for Computational Linguistics. find
  10. Ivar Frisch and Mario Giulianelli. 2024. LLM agents in interaction: Measuring personality consistency and linguistic alignment in interacting populations of large language models. In Proceedings of the 1st Workshop on Personalization of Generative AI Systems (PERSONALIZE 2024), pages 102–111, St. Ju… find
  11. Tao Ge, Xin Chan, Xiaoyang Wang, Dian Yu, Haitao Mi, and Dong Yu. 2025. Scaling synthetic data creation with 1,000,000,000 personas. Preprint, arXiv:2406.20094.
  12. H Giles. 2008. Communication accommodation theory. engaging theories in interpersonal communication: Multiple perspectives, 161–173. find
  13. Howard Giles, Justine Coupland, and Nikolas Coupland. 1991. Accommodation theory: Communication, context, and consequence. In Contexts of Accommodation: Developments in Applied Sociolinguistics, pages 1–68. Cambridge University Press. find
  14. Robert A Hinde. 1976. Interactions, relationships and social structure. Man, pages 1–17. find
  15. Robert A Hinde. 1987. Individuals, relationships and culture: Links between ethology and the social sciences. CUP Archive. find
  16. Carl I Hovland and Walter Weiss. 1951. The influence of source credibility on communication effectiveness. Public Opinion Quarterly, 15(4):635–650. find
  17. Yuyang Jiang, Longjie Guo, Yuchen Wu, Aylin Caliskan, Tanu Mitra, and Hua Shen. 2025. Beyond one-way influence: Bidirectional opinion dynamics in multi-turn human-llm interactions. arXiv preprint arXiv:2510.20039. — [arXiv:2510.20039](https://arxiv.org/abs/2510.20039) find
  18. Zhijing Jin, Caleb Ziems, Yixin Liu, Yutong Cai, Tianyu Zhang, Xuan Lu, Diyi Yang, Hinrich Schütze, and Yiming Yang. 2024. Persuading with large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL). find
  19. Ewa Kacewicz, James W Pennebaker, Matthew Davis, Minchul Jeon, and Arthur C Graesser. 2014. Pronoun use reflects standings in social hierarchies. Journal of Language and Social Psychology, 33(2):125–143. find
  20. Enkelejda Kasneci, Kathrin Seßler, Stefan Küchemann, Maria Bannert, Daryna Dementieva, Frank Fischer, Urs Gasser, Georg Groh, Stephan Günnemann, Eyke Hüllermeier, and 1 others. 2023. Chatgpt for good? on opportunities and challenges of large language models for education. Learning and individual dif…
  21. Stanley Milgram. 1963. Behavioral study of obedience. Journal of Abnormal and Social Psychology, 67(4):371–378. find
  22. Takayuki Miura, Toshiki Shibahara, Masanobu Kii, Atsunori Ichikawa, Juko Yamamoto, and Koji Chida. 2023. On rényi differential privacy in statistics-based synthetic data generation. Journal of Information Processing, 31:812–820. find
  23. Runcong Zhao, Qinglin Zhu, Hainiu Xu, Jiazheng Li, Yuxiang Zhou, Yulan He, and Lin Gui. 2024. Large language models fall short: Understanding complex relationships in detective narratives. arXiv preprint arXiv:2402.11051. find
  24. Michael Natale. 1975. Convergence of mean vocal intensity in dyadic communication as a function of social desirability. Journal of Personality and Social Psychology, 32(5):790–804. find
  25. Xuhui Zhou, Hao Zhu, Leena Mathur, Ruohong Zhang, Haofei Yu, Zhengyang Qi, Louis-Philippe Morency, Yonatan Bisk, Daniel Fried, Graham Neubig, and 1 others. 2023. Sotopia: Interactive evaluation for social intelligence in language agents. arXiv preprint arXiv:2310.11667.
  26. Kate G. Niederhoffer and James W. Pennebaker. 2002. Linguistic style matching in social interaction. Journal of Language and Social Psychology, 21(4):337–360. find
  27. Bill Noble and Raquel Fernández. 2015. Centre stage: How social network position shapes linguistic coordination. In Proceedings of the 6th workshop on cognitive modeling and computational linguistics, pages 29–38. find
  28. Harsha Nori, Nicholas King, Scott Mayer McKinney, Dean Carignan, and Eric Horvitz. 2023. Capabilities of gpt-4 on medical challenge problems. arXiv preprint arXiv:2303.13375.
  29. Andreas Opedal, Alessandro Stolfo, Haruki Shirakami, Ying Jiao, Ryan Cotterell, Bernhard Schölkopf, Abulhair Saparov, and Mrinmaya Sachan. 2024. Do language models exhibit the same cognitive biases in problem solving as human learners? In Forty-first International Conference on Machine Learning. find
  30. OpenAI. 2023. Gpt-4 technical report. Accessed: 2025-05-05.
  31. Hao Zhu, Graham Neubig, and Yonatan Bisk. 2021. Few-shot language coordination by modeling theory of mind. In International Conference on Machine Learning, pages 12901–12911. PMLR. find
  32. Philip G Zimbardo, Craig Haney, Curtis Banks, and David Jaffe. 1973. The stanford prison experiment: A simulation study of the psychology of imprisonment. Naval Research Reviews, 30(9):4–17. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vijjini-2026-do-llm-agents-mirror-socio/