Elamur Research⌕
☰
home / papers / 2026 / Can Large Language Models Infer Human Actions and Motives? Evaluation in Social Prediction and Inspection Games

Can Large Language Models Infer Human Actions and Motives? Evaluation in Social Prediction and Inspection Games

Paper · 2026 · Kaleen Shrestha, Abhinav Gupta, Harish Dukkipati et al.

Summary

The paper asks whether current LLMs can anticipate a person's choices in strategic settings by working out the hidden motives behind them, and whether those inferred motives carry over to a new situation. The authors adapt two validated cognitive-science paradigms from van Baar et al. (2022). In the Social Prediction Game (SPG), an LLM observes a simulated Player A over 16 rounds of randomly drawn two-player economic games and predicts each move. Player A follows either a motive documented in people (greedy, risk-averse) or an artificial inverse of one. In the Inspection Game (IG), the LLM then acts as an Employer deciding whether to pay to inspect that same player, which shows whether it actually learned the motive. Open- and closed-source models were compared with human data and a statistical baseline. The strongest models reached near-human accuracy and, like people, did better on human-like motives. However, their accuracy did not rise across rounds, and they failed to transfer motives to the IG. The authors therefore warn against treating LLMs as substitutes for humans in behavior modeling.

Key points

  • The SPG uses four canonical games (Prisoner's Dilemma, Stag Hunt, Harmony Game, Snowdrift Game), with fixed payoffs R = 10 and P = 5. The S and T values vary to define each game.
  • There are four simulated motives. Greedy and risk-averse come from human choice data (Poncela-Casasnovas et al., 2016), while inverse greedy and inverse risk-averse are artificial controls.
  • The IG lasts 16 rounds and uses a staircase design in which the inspection cost shifts by $5 after each decision. Willingness to pay is taken as the average cost over the final 5 rounds, and the LLM also states the probability that the Employee is working.
  • Unlike humans and the statistical model, LLMs showed no learning over rounds. Despite GPT 4.1's human-level SPG accuracy, the models did not generalize the motive to the IG. Smaller open models (3–13B) often broke the required response format.
  • The authors release code, simulation data and evaluation scripts at github.com/interaction-lab/social-decision-modeling.

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 Southern CaliforniaLos Angeles · United States162 works · 35,836 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 worksDecision Sciences · 209 worksManagement Science and Operations Research · 108 works

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

Cite

Cite as Kaleen Shrestha, Abhinav Gupta, Harish Dukkipati, Zhonghao Shi and Maja Mataric. 2026. Can Large Language Models Infer Human Actions and Motives? Evaluation in Social Prediction and Inspection Games. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1641

@inproceedings{shrestha2026can,
  title = {{Can Large Language Models Infer Human Actions and Motives? Evaluation in Social Prediction and Inspection Games}},
  author = {Kaleen Shrestha, Abhinav Gupta, Harish Dukkipati, Zhonghao Shi and Maja Mataric},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1641},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1641}
}

References

25 references, 6 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. 2024. Perils and opportunities in using large language models in psychological research. PNAS nexus, 3(7):pgae245. find
  2. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  3. Gati Aher, Rosa I. Arriaga, and Adam Tauman Kalai. 2023. Using large language models to simulate multiple humans and replicate human subject studies. arXiv preprint arXiv:2208.10264. find
  4. Anthropic. 2025. Claude sonnet 4. Technical report, Anthropic. find
  5. Rudolf Avenhaus, Bernhard Von Stengel, and Shmuel Zamir. 2002. Inspection games. Handbook of game theory with economic applications, 3:1947–1987. find
  6. Nicolas Bougie and Narimawa Watanabe. 2025. Citysim: Modeling urban behaviors and city dynamics with large-scale llm-driven agent simulation. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 215–229. find
  7. Zhuang Chen, Jincenzi Wu, Jinfeng Zhou, Bosi Wen, Guanqun Bi, Gongyao Jiang, Yaru Cao, Mengting Hu, Yunghwei Lai, Zexuan Xiong, and 1 others. 2024. Tombench: Benchmarking theory of mind in large language models. arXiv preprint arXiv:2402.15052.
  8. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  9. Caoyun Fan, Jindou Chen, Yaohui Jin, and Hao He. 2024. Can large language models serve as rational players in game theory? a systematic analysis. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 17960–17967. find
  10. Nicoló Fontana, Francesco Pierri, and Luca Maria Aiello. 2024. Nicer than humans: How do large language models behave in the prisoner’s dilemma? arXiv preprint arXiv:2406.13605. — [arXiv:2406.13605](https://arxiv.org/abs/2406.13605) find
  11. Nicoló Fontana, Francesco Pierri, and Luca Maria Aiello. 2025. Nicer than humans: How do large language models behave in the prisoner’s dilemma? In Proceedings of the International AAAI Conference on Web and Social Media, volume 19, pages 522–535. find
  12. Groq. 2025. Groq cloud. Technical report, Groq. find
  13. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  14. Jingru Jia, Zehua Yuan, Junhao Pan, Paul E McNamara, and Deming Chen. 2024. Decision-making behavior evaluation framework for llms under uncertain context. Advances in neural information processing systems, 37:113360–113382. find
  15. Dongjun Kang, Joonsuk Park, Yohan Jo, and JinYeong Bak. 2023. From values to opinions: Predicting human behaviors and stances using value-injected large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 15539–15559, Singapore. Associati… find
  16. Akaash Kolluri, Shengguang Wu, Joon Sung Park, and Michael S Bernstein. 2025. Finetuning llms for human behavior prediction in social science experiments. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 30084–30099. find
  17. John O Ledyard and 1 others. 1994. Public goods: A survey of experimental research. Division of the Humanities and Social Sciences, California Inst. of Technology. find
  18. Xiaoou Liu, Tiejin Chen, Longchao Da, Chacha Chen, Zhen Lin, and Hua Wei. 2025. Uncertainty quantification and confidence calibration in large language models: A survey. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2, pages 6107–6117.
  19. Meta. 2025a. Llama 3.3. Technical report, Meta. find
  20. Meta. 2025b. The llama 4 herd: The beginning of a new era of natively multimodal ai innovation. Technical report, Meta. find
  21. Allen Nie, Yuhui Zhang, Atharva Amdekar, Chris Piech, Tatsunori Hashimoto, and Tobias Gerstenberg. 2023. Moca: Measuring human-language model alignment on causal and moral judgment tasks.
  22. 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
  23. Julia Poncela-Casasnovas, Mario Gutiérrez-Roig, Carlos Gracia-Lázaro, Julian Vicens, Jesús Gómez-Gardeñes, Josep Perelló, Yamir Moreno, Jordi Duch, and Angel Sánchez. 2016. Humans display a reduced set of consistent behavioral phenotypes in dyadic games. Science advances, 2(8):e1600451. find
  24. Karl Sigmund. 2010. The calculus of selfishness. Princeton University Press. find
  25. Ala N Tak, Amin Banayeeanzade, Anahita Bolourani, Fatemeh Bahrani, Ashutosh Chaubey, Sai Praneeth Karimireddy, Norbert Schwarz, and Jonathan Gratch. 2026. Sparks of rationality: Do reasoning llms align with human judgment and choice? arXiv preprint arXiv:2601.22329. — [arXiv:2601.22329](https://arxiv.org/abs/2601.22329) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shrestha-2026-can-large-language-models-infer/