Elamur Research⌕
☰
home / papers / 2024 / Systematic Biases in LLM Simulations of Debates

Systematic Biases in LLM Simulations of Debates

Paper · 2024 · Amir Taubenfeld, Yaniv Dover, Roi Reichart et al.

Summary

The paper asks whether LLM-driven agents can stand in for people in social simulations, focusing on political attitude change. The authors stage debates among two or three agents, each prompted to act as a Republican or a Democrat, on four highly divisive US issues drawn from a 2023 Pew survey. Throughout each debate, agents rate their agreement with the topic, and the resulting attitude trajectories are compared with documented human dynamics. The main finding is that agents tend to drift toward the base model's own built-in stance, even when that stance contradicts the persona they were given. Two partisans from the same party therefore become more moderate instead of polarizing, which runs against the echo-chamber effect seen in human groups. To test whether this link is causal, the authors build a self-fine-tuning procedure. They collect agents' answers to politically probing interview questions and use them to retrain the underlying LLM. After this shift in the model's bias, agents follow the new bias even though their prompts stay the same. This matters for anyone using LLM agents as proxies for human participants in behavioral or economic research.

Key points

  • Topics: Agents argue over four issues: Climate Change, Illegal Immigration, Gun Violence, and Racism. These were the most partisan issues in the Pew Research Center survey by Doherty et al. (2023).
  • Personas: The LLM itself generated 40 Republican and 40 Democrat personas at temperature 1.0 from a short meta-prompt. Some experiments also included a "default" agent with no political identity.
  • Echo chambers: In same-party debates, agents converged toward the model's default bias. Human studies such as Hobolt et al. (2023), with Labour and Conservative groups, instead show growing polarization.
  • Effect of self-fine-tuning: Fine-tuning on the agents' own interview answers moved where the agents converged in debates. Section 7 reports that performance on general benchmarks was largely preserved.
  • Limitations and ethics:
    • The study covers only small debates of 2–3 agents.
    • Attitudes are measured through interviews rather than systematic human evaluation.
    • The authors caution that bias-manipulating fine-tuning could be misused to spread misinformation.

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

🎓Hebrew University of JerusalemJerusalem · Israel65 works · 9,387 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksSocial Sciences · 1,772 worksSafety Research · 543 worksPhysics and Astronomy · 44 worksStatistical and Nonlinear Physics · 40 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Amir Taubenfeld, Yaniv Dover, Roi Reichart and Ariel Goldstein. 2024. Systematic Biases in LLM Simulations of Debates. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2024.emnlp-main.16

@inproceedings{taubenfeld2024systematic,
  title = {{Systematic Biases in LLM Simulations of Debates}},
  author = {Amir Taubenfeld, Yaniv Dover, Roi Reichart and Ariel Goldstein},
  booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing},
  year = {2024},
  doi = {10.18653/v1/2024.emnlp-main.16},
  url = {https://doi.org/10.18653/v1/2024.emnlp-main.16}
}

References

36 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. William Agnew, A Stevie Bergman, Jennifer Chien, Mark Díaz, Seliem El-Sayed, Jaylen Pittman, Shakir Mohamed, and Kevin R McKee. 2024. The illusion of artificial inclusion. arXiv preprint arXiv:2401.08572. — [arXiv:2401.08572](https://arxiv.org/abs/2401.08572) find
  2. Jaimeen Ahn and Alice Oh. 2021. Mitigating language-dependent ethnic bias in BERT. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 533–549, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics. find
  3. Lisa P Argyle, Ethan C Busby, Nancy Fulda, Joshua R Gubler, Christopher Rytting, and David Wingate. 2023. Out of one, many: Using language models to simulate human samples. Political Analysis, 31(3):337–351.
  4. Paola Arias, Nicolas Bellouin, Erika Coppola, Richard Jones, Gerhard Krinner, Jochem Marotzke, Vaishali Naik, Matthew Palmer, G-K Plattner, Joeri Rogelj, et al. 2021. Climate change 2021: the physical science basis. contribution of working group i to the sixth assessment report of the intergovernmen… find
  5. Shikha Bordia and Samuel R. Bowman. 2019. Identifying and reducing gender bias in word-level language models. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Student Research Workshop, pages 7–15, Minneapolis, Minnesota. Associati…
  6. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  7. Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, et al. 2023. Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712.
  8. Yiting Chen, Tracy Xiao Liu, You Shan, and Songfa Zhong. 2023. The emergence of economic rationality of gpt. Proceedings of the National Academy of Sciences, 120(51):e2316205120. find
  9. Myra Cheng, Tiziano Piccardi, and Diyi Yang. 2023. CoMPosT: Characterizing and evaluating caricature in LLM simulations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 10853–10875, Singapore. Association for Computational Linguistics. find
  10. Yun-Shiuan Chuang, Agam Goyal, Nikunj Harlalka, Siddharth Suresh, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, and Timothy T Rogers. 2023. Simulating opinion dynamics with networks of llm-based agents. arXiv preprint arXiv:2311.09618. find
  11. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. Qlora: Efficient finetuning of quantized llms. arXiv preprint arXiv:2305.14314.
  12. Danica Dillion, Niket Tandon, Yuling Gu, and Kurt Gray. 2023. Can ai language models replace human participants? Trends in Cognitive Sciences, 27(7):597–600. find
  13. Peter H Ditto, Brittany S Liu, Cory J Clark, Sean P Wojcik, Eric E Chen, Rebecca H Grady, Jared B Celniker, and Joanne F Zinger. 2019. At least bias is bipartisan: A meta-analytic comparison of partisan bias in liberals and conservatives. Perspectives on Psychological Science, 14(2):273–291. find
  14. Carroll Doherty, Jocelyn Kiley, Nida Asheer, and Talia Price. 2023. Inflation, health costs, partisan cooperation among the nation’s top problems. Pew Research Center. find
  15. Tiancheng Hu, Yara Kyrychenko, Steve Rathje, Nigel Collier, Sander van der Linden, and Jon Roozenbeek. 2023. Generative language models exhibit social identity biases. arXiv preprint arXiv:2310.15819. — [arXiv:2310.15819](https://arxiv.org/abs/2310.15819) find
  16. Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. arXiv preprint arXiv:2310.06825. find
  17. Dan M Kahan, Ellen Peters, Maggie Wittlin, Paul Slovic, Lisa Larrimore Ouellette, Donald Braman, and Gregory Mandel. 2012. The polarizing impact of science literacy and numeracy on perceived climate change risks. Nature climate change, 2(10):732–735. find
  18. Dahyun Kim, Chanjun Park, Sanghoon Kim, Wonsung Lee, Wonho Song, Yunsu Kim, Hyeonwoo Kim, Yungi Kim, Hyeonju Lee, Jihoo Kim, Changbae Ahn, Seonghoon Yang, Sukyung Lee, Hyunbyung Park, Gyoungjin Gim, Mikyoung Cha, Hwalsuk Lee, and Sunghun Kim. 2023. Solar 10.7b: Scaling large language models with sim… find
  19. Ruibo Liu, Ruixin Yang, Chenyan Jia, Ge Zhang, Denny Zhou, Andrew M Dai, Diyi Yang, and Soroush Vosoughi. 2023. Training socially aligned language models in simulated human society. arXiv preprint arXiv:2305.16960. find
  20. Dan Milmo. 2023. Chatgpt reaches 100 million users two months after launch. The Guardian, 3. find
  21. Iason Gabriel. 2020. Artificial intelligence, values, and alignment. Minds and machines, 30(3):411–437. find
  22. Leo Gao, Jonathan Tow, Baber Abbasi, Stella Biderman, Sid Black, Anthony DiPofi, Charles Foster, Laurence Golding, Jeffrey Hsu, Alain Le Noac’h, Haonan Li, Kyle McDonell, Niklas Muennighoff, Chris Ociepa, Jason Phang, Laria Reynolds, Hailey Schoelkopf, Aviya Skowron, Lintang Sutawika, Eric Tang, Ani… find
  23. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300.
  24. Sara B Hobolt, Katharina Lawall, and James Tilley. 2023. The polarizing effect of partisan echo chambers. American Political Science Review, pages 1–16. find
  25. John J Horton. 2023. Large language models as simulated economic agents: What can we learn from homo silicus? Technical report, National Bureau of Economic Research. find
  26. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685.
  27. Fabio Motoki, Valdemar Pinho Neto, and Victor Rodrigues. 2024. More human than human: Measuring chatgpt political bias. Public Choice, 198(1):3–23. find
  28. OpenAI. 2023. Openai models. https://platform.openai.com/docs/models/overview. ↗
  29. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35:27730–27744.
  30. Oscar Oviedo-Trespalacios, Amy E Peden, Thomas Cole-Hunter, Arianna Costantini, Milad Haghani, J.E. Rod, Sage Kelly, Helma Torkamaan, Amina Tariq, James David Albert Newton, Timothy Gallagher, Steffen Steinert, Ashleigh J. Filtness, and Genserik Reniers. 2023. The risks of using chatgpt to obtain co… find
  31. Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–22. find
  32. Joon Sung Park, Lindsay Popowski, Carrie Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2022. Social simulacra: Creating populated prototypes for social computing systems. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology, pages 1–18. find
  33. John Priniski and Zachary Horne. 2018. Attitude change on reddit’s change my view. In CogSci. find
  34. Chen Qian, Xin Cong, Cheng Yang, Weize Chen, Yusheng Su, Juyuan Xu, Zhiyuan Liu, and Maosong Sun. 2023. Communicative agents for software development. arXiv preprint arXiv:2307.07924.
  35. Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D Manning, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. arXiv preprint arXiv:2305.18290.
  36. Murray Shanahan, Kyle McDonell, and Laria Reynolds. 2023. Role play with large language models. Nature, pages 1–6.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/taubenfeld-2024-systematic-biases-llm-simulations-debates/