Elamur Research⌕
☰
home / papers / 2024 / Roleplay-doh: Enabling Domain-Experts to Create LLM-simulated Patients via Eliciting and Adhering to Principles

Roleplay-doh: Enabling Domain-Experts to Create LLM-simulated Patients via Eliciting and Adhering to Principles

Paper · 2024 · Ryan Louie, Ananjan Nandi, William Fang et al.

Summary

Practice partners for novice counselors are hard to build with LLMs. Real therapy transcripts are private, so fine-tuning is mostly impossible. Simple prompting also gives patients that sound unrealistic, missing traits such as casual speech and pushback. Roleplay-doh is an interactive tool that lets experienced counselors shape an AI patient without writing prompts. While chatting with the simulated patient, the expert leaves comments on individual replies. An LLM turns these comments into general natural-language rules called principles, and the patient's next replies follow them along with a persona description.

Pilot tests showed that GPT-4 often ignored or misused these rules. To fix this, the authors added a principle-adherence pipeline with three steps:

  • split complex or situational principles into simple yes/no checks,
  • test whether each check applies to the current turn,
  • self-refine the reply when needed.

In a within-subjects study with 25 counseling experts, patients built with Roleplay-doh were judged more authentic and closer to real cases, both by their creators and by outside counselors. The method has nothing specific to mental health, so it could likely transfer to other expert-guided social simulations.

Key points

  • Experts give three kinds of feedback: "kudos" (keep this behavior), "critique" (avoid this), and "rewrite" (a better reply). GPT-3.5 converts kudos and critiques into principles. GPT-4 infers a principle by comparing an original reply with its rewrite.
  • With plain GPT-4 prompting, about 20% of replies broke principles. The main causes were multipart principles and rules that only apply in certain situations.
  • The adherence pipeline raised response quality and principle following by roughly 30% on the downstream task.
  • Third-party rankings compared the full pipeline with a no-self-refine baseline. For principle following it won 35% of the time and lost 5%. For dialogue consistency it won 35% and lost 10%, the best result among all ablations.
  • Code and data, including the expert-written principles, are released at roleplay-doh.github.io. Known limitations: the tool does not check for conflicting principles, it is text-only, and only one simulated case involved serious mental illness.

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

🎓Stanford UniversityStanford · United States434 works · 314,419 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksMedicine · 241 worksHealth Informatics · 184 works

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

Cite

Cite as Ryan Louie, Ananjan Nandi, William Fang, Cheng Chang, Emma Brunskill and Diyi Yang. 2024. Roleplay-doh: Enabling Domain-Experts to Create LLM-simulated Patients via Eliciting and Adhering to Principles. 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.591

@inproceedings{louie2024roleplay,
  title = {{Roleplay-doh: Enabling Domain-Experts to Create LLM-simulated Patients via Eliciting and Adhering to Principles}},
  author = {Ryan Louie, Ananjan Nandi, William Fang, Cheng Chang, Emma Brunskill and Diyi Yang},
  booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing},
  year = {2024},
  doi = {10.18653/v1/2024.emnlp-main.591},
  url = {https://doi.org/10.18653/v1/2024.emnlp-main.591}
}

References

25 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. 2022. Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073.
  2. Douglas Bates, Martin Mächler, Ben Bolker, and Steve Walker. 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1):1– 48.
  3. Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Michal Podstawski, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Hubert Niewiadomski, Piotr Nyczyk, and Torsten Hoefler. 2024. Graph of thoughts: Solving elaborate problems with large language models. Proceedings of the AAAI Conference… find
  4. Virginia Braun and Victoria Clarke. 2006. Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2):77–101. find
  5. Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Monika Himmelbauer, Tamara Seitz, Charles Seidman, and Henriette Löffler-Stastka. 2018. Standardized patients in psychiatry–the best way to learn clinical skills? BMC medical education, 18:1–6. find
  6. Tae Soo Kim, Yoonjoo Lee, Jamin Shin, Young-Ho Kim, and Juho Kim. 2024. Evallm: Interactive evaluation of large language model prompts on user-defined criteria. In Proceedings of the CHI Conference on Human Factors in Computing Systems, CHI ’24. ACM. find
  7. Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang, and Minlie Huang. 2021. Towards emotional support dialog systems. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Lan…
  8. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, et al. 2024. Self-refine: Iterative refinement with self-feedback. Advances in Neural Information Processing Systems, 36. find
  9. Julia M. Markel, Steven G. Opferman, James A. Landay, and Chris Piech. 2023. Gpteach: Interactive ta training with gpt-based students. In Proceedings of the Tenth ACM Conference on Learning @ Scale, L@S ’23, page 226–236, New York, NY, USA. Association for Computing Machinery. find
  10. Julia Othlinghaus-Wulhorst and H. Ulrich Hoppe. 2020. A technical and conceptual framework for serious role-playing games in the area of social skill training. Frontiers in Computer Science, 2. find
  11. 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, UIST ’23, New York, NY, USA. Ass… find
  12. 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, UIST ’22, New Yo… find
  13. Savvas Petridis, Benjamin D Wedin, James Wexler, Mahima Pushkarna, Aaron Donsbach, Nitesh Goyal, Carrie J Cai, and Michael Terry. 2024. Constitu-tionmaker: Interactively critiquing large language models by converting feedback into principles. In Proceedings of the 29th International Conference on In… find
  14. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2024. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36.
  15. Logan Stapleton, Jordan Taylor, Sarah Fox, Tongshuang Wu, and Haiyi Zhu. 2023. Seeing seeds beyond weeds: Green teaming generative ai for beneficial uses. arXiv preprint arXiv:2306.03097. — [arXiv:2306.03097](https://arxiv.org/abs/2306.03097) find
  16. Michael J Tanana, Christina S Soma, Vivek Srikumar, David C Atkins, and Zac E Imel. 2019. Development and evaluation of clientbot: Patient-like conversational agent to train basic counseling skills. Journal of medical Internet research, 21(7):e12529. find
  17. Ruiyi Wang, Stephanie Milani, Jamie C Chiu, Shaun M Eack, Travis Labrum, Samuel M Murphy, Nev Jones, Kate Hardy, Hong Shen, Fei Fang, et al. 2024. Patient-{\\Psi}: Using large language models to simulate patients for training mental health professionals. arXiv preprint arXiv:2405.19660. find
  18. Diyi Yang, Caleb Ziems, William Held, Omar Shaikh, Michael S Bernstein, and John Mitchell. 2024. Social skill training with large language models. arXiv preprint arXiv:2404.04204. find
  19. Shunyu Yao, Howard Chen, Austin W Hanjie, Runzhe Yang, and Karthik Narasimhan. 2023. Collie: Systematic construction of constrained text generation tasks. arXiv preprint arXiv:2307.08689.
  20. Zheng Yao, Haiyi Zhu, and Robert E. Kraut. 2022. Learning to become a volunteer counselor: Lessons from a peer-to-peer mental health community. Proc. ACM Hum.-Comput. Interact., 6(CSCW2). find
  21. J.D. Zamfirescu-Pereira, Richmond Y. Wong, Bjoern Hartmann, and Qian Yang. 2023. Why johnny can’t prompt: How non-ai experts try (and fail) to design llm prompts. CHI ’23, New York, NY, USA. Association for Computing Machinery. find
  22. Xuhui Zhou, Zhe Su, Tiwalayo Eisape, Hyunwoo Kim, and Maarten Sap. 2024. Is this the real life? is this just fantasy? the misleading success of simulating social interactions with llms. arXiv preprint arXiv:2403.05020. find
  23. Swarnadeep Saha, Omer Levy, Asli Celikyilmaz, Mohit Bansal, Jason Weston, and Xian Li. 2023. Branch-solve-merge improves large language model evaluation and generation. arXiv preprint arXiv:2310.15123. find
  24. Omar Shaikh, Valentino Emil Chai, Michele Gelfand, Diyi Yang, and Michael S. Bernstein. 2024. Rehearsal: Simulating conflict to teach conflict resolution. In Proceedings of the CHI Conference on Human Factors in Computing Systems, CHI ’24, New York, NY, USA. Association for Computing Machinery. find
  25. Weiyan Shi, Emily Dinan, Kurt Shuster, Jason Weston, and Jing Xu. 2022. When life gives you lemons, make cherryade: Converting feedback from bad responses into good labels. arXiv preprint arXiv:2210.15893. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/louie-2024-roleplay-doh-enabling-domain-experts/