Elamur Research⌕
☰
home / papers / 2024 / Comparing Humans and Large Language Models on an Experimental Protocol Inventory for Theory of Mind Evaluation (EPITOME)

Comparing Humans and Large Language Models on an Experimental Protocol Inventory for Theory of Mind Evaluation (EPITOME)

Paper · 2024 · Cameron R. Jones, Sean Trott and Benjamin Bergen

Summary

Whether large language models show Theory of Mind (ToM) is disputed. Earlier evaluations often relied on single tasks and gave conflicting results, and human ToM measures correlate weakly with one another. Jones, Trott and Bergen build EPITOME, an inventory of six established psychology experiments. Together they cover belief tracking, deeply nested mental states, emotion inference, non-literal language, and pragmatic inference that depends on what a speaker knows. Using identical materials, the authors collect human baselines for every item and compare them with five LLMs. The main focus is GPT-3 text-davinci-002, a strong model trained without RLHF, so the comparison isolates what language statistics alone can supply. Besides accuracy, they test whether models respond to manipulated mental-state variables. They also test whether these variables still explain human answers once model predictions are controlled for. Results are mixed. The model is sensitive to mental states and matches humans on several tasks, but it errs systematically on others, especially when pragmatic conclusions hinge on inferred knowledge. This suggests distributional learning alone may not explain human ToM.

Key points

  • EPITOME's six tasks are the False Belief Task (FB), Recursive Mindreading (RM, up to 7 embedding levels), the Short Story Task (ShS), the Strange Stories Task (StS), the Indirect Request Task (IR), and Scalar Implicature (SI).
  • Four of the six studies were preregistered. Human data for all items, plus code and materials, are publicly released to allow item-level comparison.
  • The main model is GPT-3 text-davinci-002, chosen because it lacks RLHF. Five LLMs in total are compared, partly to examine effects of model scale.
  • A "distributional baseline" analysis tests whether mental-state variables still account for variation in human answers that LLM predictions leave unexplained.
  • LLMs did comparatively better on non-mental control questions (in RM and ShS). Their weakest area was pragmatic reasoning based on speaker knowledge.

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 California San DiegoSan Diego · United States73 works · 51,948 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 worksPsychology · 1,552 worksDevelopmental and Educational Psychology · 499 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Cameron R. Jones, Sean Trott and Benjamin Bergen. 2024. Comparing Humans and Large Language Models on an Experimental Protocol Inventory for Theory of Mind Evaluation (EPITOME). Transactions of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/tacl_a_00674

@article{jones2024comparing,
  title = {{Comparing Humans and Large Language Models on an Experimental Protocol Inventory for Theory of Mind Evaluation (EPITOME)}},
  author = {Cameron R. Jones, Sean Trott and Benjamin Bergen},
  journal = {Transactions of the Association for Computational Linguistics},
  year = {2024},
  doi = {10.1162/tacl\_a\_00674},
  url = {https://doi.org/10.1162/tacl_a_00674}
}

References

66 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D. Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chr… — Language Models are Few-Shot Learners
  2. Ian A. Apperly. 2012. What is ‘‘theory of mind’’? Concepts, cognitive processes and individual differences. Quarterly Journal of Experimental Tyler A. Chang and Benjamin K. Bergen. 2023. Language model behavior: A comprehensive survey. \#\# Acknowledgments We would like to thank Owen Pi, Alice Zhang,…
  3. Jill G. de Villiers and Peter A. de Villiers. 2014. The role of language in theory of mind development. Topics in Language Disorders, 34(4):313–328. https://doi.org/10.1097/TLD.0000000000000037 https://doi.org/10.1111/tops.12007, PubMed: 23335578 — [doi:10.1097/tld.0000000000000037](https://doi.org/10.1097/tld.0000000000000037) ↗
  4. Herbert P. Grice. 1975. Logic and conversation. In Speech Acts, pages 41–58. Brill. https://doi.org/10.1163/9789004368811 003 ↗
  5. Daniel C. Dennett. 1978. Beliefs about beliefs [P&W, SR&B]. Behavioral and Brain Sciences, 1(4):568–570. https://doi.org/10.1017/S0140525X00076664 — [doi:10.1017/s0140525x00076664](https://doi.org/10.1017/s0140525x00076664) ↗
  6. Thilo Hagendorff. 2023. Machine psychology: Investigating emergent capabilities and behavior in large language models using psychological methods. find
  7. Sahraoui Dhelim, Huansheng Ning, Fadi Farha, Liming Chen, Luigi Atzori, and Mahmoud Daneshmand. 2021. IoT-enabled social relation ships meet artificial social intelligence. IEEE Internet of Things Journal, 8(24):17817–17828. https://doi.org/10.1109/JIOT.2021 .3081556 — [doi:10.1109/jiot.2021](https://doi.org/10.1109/jiot.2021) ↗
  8. Courtney Melinda Hale and Helen Tager-Flusberg. 2003. The influence of language on theory of mind: A training study. Developmental Science, 6(3):346–359. https://doi.org/10.1111/1467-7687.00289, PubMed: 16467908 ↗
  9. Marta Halina. 2015. There is no special problem of mindreading in nonhuman animals. Philosophy of Science, 82(3):473–490. https://doi.org/10.1086/681627 — [doi:10.1086/681627](https://doi.org/10.1086/681627) ↗
  10. David Dodell-Feder, Sarah Hope Lincoln, Joseph P. Coulson, and Christine I. Hooker. 2013. Using fiction to assess mental state understanding: A new task for assessing theory of mind in adults. PLOS ONE, 8(11):e81279. https://doi.org/10.1371/journal .pone.0081279, PubMed: 24244736 — [doi:10.1371/journal](https://doi.org/10.1371/journal) ↗
  11. J. R. Firth. 1957. A Synopsis of Linguistic Theory. Blackwell, Oxford. find
  12. Chris D. Frith and Uta Frith. 2012. Mechanisms of social cognition. Annual Review of Psychology, 63(1):287–313. https://doi.org/10 .1146/annurev-psych-120710-100449, PubMed: 21838544 find
  13. Kanishk Gandhi, Jan-Philipp Fränken, Tobias Gerstenberg, and Noah D. Goodman. 2023. Understanding social reasoning in language models with language models. find
  14. Francesca G. E. Happé. 1994. An advanced test of theory of mind: Understanding of story characters’ thoughts and feelings by able autistic, mentally handicapped, and normal children and adults. Journal of Autism and Developmental Disorders, 24(2):129–154. https://doi.org/10.1007/BF02172093, PubMed… — [doi:10.1007/bf02172093](https://doi.org/10.1007/bf02172093) ↗
  15. Paul L. Harris. 2005. Conversation, pretense, and theory of mind. In Why Language Matters for Theory of Mind, pages 70–83. Oxford University Press, New York, NY, US. https://doi.org/10.1093/acprof:oso/9780195159912 .003.0004 — [doi:10.1093/acprof:oso/9780195159912](https://doi.org/10.1093/acprof:oso/9780195159912) ↗
  16. Zellig S. Harris. 1954. Distributional structure. Word, 10(2–3):146–162. https://doi.org/10.1080/00437956.1954.11659520 ↗
  17. Morton Ann Gernsbacher and Melanie Yergeau. 2019. Empirical failures of the claim that autistic people lack a theory of mind. Archives of Scientific Psychology, 7(1):102–118. https://doi.org/10.1037/arc0000067, PubMed: 31938672 — [doi:10.1037/arc0000067](https://doi.org/10.1037/arc0000067) ↗
  18. Elizabeth O. Hayward and Bruce D. Homer. 2017. Reliability and validity of advanced theory-of-mind measures in middle childhood and adolescence. British Journal of Developmental Psychology, 35(3):454–462. https://doi.org/10.1111/bjdp.12186, PubMed: 28464376 — [doi:10.1111/bjdp.12186](https://doi.org/10.1111/bjdp.12186) ↗
  19. Shahriar Golchin and Mihai Surdeanu. 2023. Time travel in LLMs: Tracing data contamination in large language models.
  20. Cecilia Heyes. 2014. Submentalizing: I am not really reading your mind. Perspectives on Psychological Science, 9(2):131–143. https://doi.org/10.1177/1745691613518076, PubMed: 26173251 — [doi:10.1177/1745691613518076](https://doi.org/10.1177/1745691613518076) ↗
  21. Noah D. Goodman and Andreas Stuhlmüller. 2013. Knowledge and implicature: Modeling language understanding as social cognition. Topics in Cognitive Science, 5(1):173–184. find
  22. Ari Holtzman, Peter West, Vered Shwartz, Yejin Choi, and Luke Zettlemoyer. 2022. Surface Form Competition: Why the Highest Probability Answer Isn’t Always Right. https://doi.org/10.18653/v1/2021.emnlp-main.564 viduals will act according to false beliefs. Science (New York, N.Y.), 354(6308):110–114.… ↗
  23. Christelle Langley, Bogdan Ionut Cirstea, Fabio Cuzzolin, and Barbara J. Sahakian. 2022. Theory of mind and preference learning at the interface of cognitive science, neuroscience, and AI: A review. Frontiers in Artificial Intelligence, 5. https://doi.org/10.3389/frai.2022.778852, PubMed: 35493614 ↗
  24. Jennifer Hu, Sammy Floyd, Olessia Jouravlev, Evelina Fedorenko, and Edward Gibson. 2022. A fine-grained comparison of pragmatic language understanding in humans and language models.
  25. Claire Hughes, Sara R. Jaffee, Francesca Happé, Alan Taylor, Avshalom Caspi, and Terrie E. Moffitt. 2005. Origins of individual differences in theory of mind: From nature to nurture? Child Development, 76(2):356–370. https://doi.org/10.1111/j.1467-8624 .2005.00850 a.x, PubMed: 15784087 — [doi:10.1111/j.1467-8624](https://doi.org/10.1111/j.1467-8624) ↗
  26. Matthew Le, Y-Lan Boureau, and Maximilian Nickel. 2019. Revisiting the evaluation of theory of mind through question answering. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-I… find
  27. Steven Johnson and Nikita Iziev. 2022. AI is mastering language. Should we trust what it says. The New York Times. find
  28. Cameron R. Jones, Tyler A. Chang, Seana Coulson, James A. Michaelov, Sean Trott, and Benjamin Bergen. 2022. Distrubutional semantics still can’t account for affordances. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 44. find
  29. James A. Michaelov, Seana Coulson, and Benjamin K. Bergen. 2022. So cloze yet so far: N400 amplitude is better predicted by distributional information than human predictability judgements. IEEE Transactions on Cognitive and Developmental Systems. https://doi.org/10.1109/TCDS.2022.3176783 — [doi:10.1109/tcds.2022.3176783](https://doi.org/10.1109/tcds.2022.3176783) ↗
  30. Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. 2020. Scaling laws for neural language models. arXiv:2001.08361 [cs, stat].
  31. Melanie Mitchell and David C. Krakauer. 2023. The debate over understanding in AI’s large language models. Proceedings of the National Academy of Sciences, 120(13):e2215907120. https://doi.org/10.1073/pnas.2215907120, PubMed: 36943882 — [doi:10.1073/pnas.2215907120](https://doi.org/10.1073/pnas.2215907120) ↗
  32. Hyunwoo Kim, Melanie Sclar, Xuhui Zhou, Ronan Le Bras, Gunhee Kim, Yejin Choi, and Maarten Sap. 2023. FANToM: A benchmark for stress-testing machine theory of mind in interactions. find
  33. Shima Rahimi Moghaddam and Christopher J. Honey. 2023. Boosting theory-of-mind performance in large language models via prompting. find
  34. Michal Kosinski. 2023. Theory of mind may have spontaneously emerged in large language models. find
  35. Aida Nematzadeh, Kaylee Burns, Erin Grant, Alison Gopnik, and Thomas L. Griffiths. 2018. Evaluating theory of mind in question answering. arXiv:1808.09352 [cs]. https://doi.org/10.18653/v1/D18-1261 ↗
  36. Christopher Krupenye and Josep Call. 2019. Theory of mind in animals: Current and future directions. WIREs Cognitive Science, 10(6):e1503. https://doi.org/10.1002/wcs.1503, PubMed: 31099977 ↗
  37. Timothy Niven and Hung-Yu Kao 2020. Probing neural network comprehension of natural language arguments. In ACL 2019-57th Annual Meeting of the Association for Computational Linguistics, pages 4658–4664. https://doi.org/10.18653/v1/P19-1459 ↗
  38. Christopher Krupenye, Fumihiro Kano, Satoshi Hirata, Josep Call, and Michael Tomasello. 2016. Great apes anticipate that other indi-Cathleen O’Grady, Christian Kliesch, Kenny Smith, and Thomas C. Scott-Phillips. 2015. The ease and extent of recursive mindreading, across implicit and explicit tasks.… — [doi:10.1016/j.evolhumbehav.2015](https://doi.org/10.1016/j.evolhumbehav.2015) find
  39. Artificial Intelligence, 4. https://doi.org/10.3389/frai.2021.682578, PubMed: 34557662 — [doi:10.3389/frai.2021.682578](https://doi.org/10.3389/frai.2021.682578) ↗
  40. Maarten Sap, Ronan LeBras, Daniel Fried, and Yejin Choi. 2022. Neural theory-of-mind? On the limits of social intelligence in large LMs. find
  41. Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. 2019. Social IQa: Commonsense reasoning about social interactions. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processi… find
  42. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Leike, and Ryan Lowe. 2022. Tr… — Training language models to follow instructions with human feedback
  43. Martin Schrimpf, Idan Asher Blank, Greta Tuckute, Carina Kauf, Eghbal A. Hosseini, Nancy Kanwisher, Joshua B. Tenenbaum, and Evelina Fedorenko. 2021. The neural architecture of language: Integrative modeling converges on predictive processing. Proceedings of the National Academy of Sciences, 118(45)… — [doi:10.1073/pnas.2105646118](https://doi.org/10.1073/pnas.2105646118) find
  44. Derek C. Penn and Daniel J. Povinelli. 2007. On the lack of evidence that non-human animals possess anything remotely resembling a ‘theory of mind’. Philosophical Transactions of the Royal Society B: Biological Sciences, 362(1480):731–744. https://doi.org/10 .1098/rstb.2006.2023, PubMed: 17264056 find
  45. Josef Perner, Susan R. Leekam, and Heinz Wimmer. 1987. Three-year-olds’ difficulty with false belief: The case for a conceptual deficit. British Journal of Developmental Psychology, 5(2):125–137. https://doi.org/10 .1111/j.2044-835X.1987.tb01048.x Eric Schwitzgebel. 2013. A dispositional approach to… — [doi:10.1057/9781137026521_5](https://doi.org/10.1057/9781137026521_5) find
  46. David Premack and Guy Woodruff. 1978. Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences, 1(4):515–526. https://doi.org/10.1017/S0140525X00076512 — [doi:10.1017/s0140525x00076512](https://doi.org/10.1017/s0140525x00076512) ↗
  47. John R. Searle. 1980. Minds, brains, and programs. Behavioral and Brain Sciences, 3(3):417–424. https://doi.org/10.1017/S0140525X00005756 ↗
  48. Neil Rabinowitz, Frank Perbet, Francis Song, Chiyuan Zhang, S. M. Ali Eslami, and Matthew Botvinick. 2018. Machine theory of mind. In Proceedings of the 35th International Conference on Machine Learning, pages 4218–4227. PMLR. find
  49. Natalie Sebanz, Harold Bekkering, and Günther Knoblich. 2006. Joint action: Bodies and minds moving together. Trends in Cognitive Sciences, 10(2):70–76. https://doi.org/10.1016/j .tics.2005.12.009, PubMed: 16406326 — [doi:10.1016/j](https://doi.org/10.1016/j) find
  50. Inioluwa Deborah Raji, Emily M. Bender, Amandalynne Paullada, Emily Denton, and Alex Hanna. 2021. AI and the everything in the whole wide world benchmark. arXiv preprint arXiv:2111.15366. find
  51. Natalie Shapira, Mosh Levy, Seyed Hossein Alavi, Xuhui Zhou, Yejin Choi, Yoav Goldberg, Maarten Sap, and Vered Shwartz. 2023. Clever Hans or neural theory of mind? Stress testing social reasoning in large language models. find
  52. Magnus Sahlgren and Fredrik Carlsson. 2021. The singleton fallacy: Why current critiques of language models miss the point. Frontiers in Ashish Sharma, Inna W. Lin, Adam S. Miner, David C. Atkins, and Tim Althoff. 2021. find
  53. Towards facilitating empathic conversations in online mental health support: A reinforcement learning approach. In Proceedings of the Web Conference 2021, pages 194–205. https://doi.org/10.1145/3442381 .3450097 — [doi:10.1145/3442381](https://doi.org/10.1145/3442381) ↗
  54. Henry Shevlin. under review. Uncanny believers: Chatbots, beliefs, and folk psychology. https://henryshevlin.com/wp-content/uploads/2021/11/Uncanny-Believers .pdf. ↗
  55. Dan Sperber and Deirdre Wilson. 2002. Pragmatics, modularity and mind-reading. Mind & Language, 17(1–2):3–23. https://doi.org/10.1111/1468-0017.00186 a personalized persuasive dialogue system for social good. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics… ↗
  56. Alex Warstadt and Samuel R. Bowman. 2022. What artificial neural networks can tell us about human language acquisition. https://doi.org/10.1201/9781003205388-2 ↗
  57. Henry M. Wellman, David Cross, and Julanne Watson. 2001. Meta-analysis of theory-of-mind development: The truth about false belief. Child Development, 72(3):655–684. https://doi.org/10.1111/1467-8624.00304, PubMed: 11405571 — [doi:10.1111/1467-8624.00304](https://doi.org/10.1111/1467-8624.00304) ↗
  58. Luca Surian, Stefania Caldi, and Dan Sperber. 2007. Attribution of beliefs by 13-month-old infants. Psychological Science, 18(7):580–586. https://doi.org/10.1111/j.1467-9280 .2007.01943.x, PubMed: 17614865 — [doi:10.1111/j.1467-9280](https://doi.org/10.1111/j.1467-9280) ↗
  59. Heinz Wimmer and Josef Perner. 1983. Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13(1):103–128. https://doi.org/10.1016 /0010-0277(83)90004-5, PubMed: 6681741 find
  60. Michael Tomasello, Malinda Carpenter, Josep Call, Tanya Behne, and Henrike Moll. 2005. Understanding and sharing intentions: The origins of cultural cognition. Behavioral and Brain Sciences, 28(5):675–691. https://doi.org/10.1017/S0140525X05000129, PubMed: 16262930 — [doi:10.1017/s0140525x05000129](https://doi.org/10.1017/s0140525x05000129) ↗
  61. Blaise Agüera y Arcas. 2022. Do Large Language Models Understand Us? Daedalus, 151(2):183–197. https://doi.org/10.1162/daed a 01909 ↗
  62. Sean Trott and Benjamin Bergen. 2018. Individual differences in mentalizing capacity predict indirect request comprehension. Discourse Processes, 56(8):675–707. https://doi.org/10.1080/0163853X.2018.1548219 — [doi:10.1080/0163853x.2018.1548219](https://doi.org/10.1080/0163853x.2018.1548219) ↗
  63. Sean Trott and Benjamin Bergen. 2020. When do comprehenders mentalize for pragmatic inference? Discourse Processes, 57(10):900–920. https://doi.org/10.1080/0163853X .2020.1822709 — [doi:10.1080/0163853x](https://doi.org/10.1080/0163853x) ↗
  64. Sean Trott, Cameron Jones, Tyler Chang, James Michaelov, and Benjamin Bergen. 2023. Do large language models know what humans know? Cognitive Science, 47(7):e13309. https://doi.org/10.1111/cogs.13309, PubMed: 37401923 — [doi:10.1111/cogs.13309](https://doi.org/10.1111/cogs.13309) ↗
  65. Zheng Zhang, Leon Bergen, Alexander Paunov, Rachel Ryskin, and Edward Gibson. 2023. Scalar implicature is sensitive to contextual alternatives. Cognitive Science, 47(2):e13238. https://doi.org/10.1111/cogs.13238, PubMed: 36739521 — [doi:10.1111/cogs.13238](https://doi.org/10.1111/cogs.13238) ↗
  66. Hao Zhu, Graham Neubig, and Yonatan Bisk. 2021. Few-shot language coordination by modeling theory of mind. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jones-2024-comparing-humans-large-language-models/