(Ir)rationality and cognitive biases in large language models
Open accessThe DOI matches its Crossref record; 26 of the 26 references it cites were found in a registry or an index.
Abstract
Do large language models (LLMs) display rational reasoning? LLMs have been shown to contain human biases due to the data they have been trained on; whether this is reflected in rational reasoning remains less clear. In this paper, we answer this question by evaluating seven language models using tasks from the cognitive psychology literature. We find that, like humans, LLMs display irrationality in these tasks. However, the way this irrationality is displayed does not reflect that shown by humans. When incorrect answers are given by LLMs to these tasks, they are often incorrect in ways that differ from human-like biases. On top of this, the LLMs reveal an additional layer of irrationality in the significant inconsistency of the responses. Aside from the experimental results, this paper seeks to make a methodological contribution by showing how we can assess and compare different capabilities of these types of models, in this case with respect to rational reasoning.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Tversky, A., & Kahneman, D. (1974) · Science · doi:10.1126/science.185.4157.1124 · Cited by 28,810 · cited 3×
“The cognitive tasks used in this paper were designed to show that human reasoning is not always rational, and that humans often employ heuristics or display cognitive biases [5, 6].”
Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.
Tversky, A., & Kahneman, D. (1983) · Psychological Review · doi:10.1037/0033-295x.90.4.293 · Cited by 3,991
Reasoning the fast and frugal way: Models of bounded rationality.
Gigerenzer, G., & Goldstein, D. G. (1996) · Psychological Review · doi:10.1037/0033-295x.103.4.650 · Cited by 3,504 · cited 1×
“Based primarily on work by Gigerenzer [45, 46], a series of facilitated versions of these tasks were developed, emphasising the impact of context and presentation of the problem.”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3586183.3606763 · Cited by 2,187 · cited 1×
“Park et al. [23] combine large language models with computational interactive agents to simulate human behaviour, both individual and within social settings.”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · arXiv · doi:10.48550/arxiv.2304.03442 · Cited by 2,173
Sparks of Artificial General Intelligence: Early experiments with GPT-4
Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2303.12712 · Cited by 1,526
Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.-F., Breazeal, C., Crandall, J. W., Christakis, N. A., Couzin, I. D., Jackson, M. O., et al. (2019) · Nature · doi:10.1038/s41586-019-1138-y · Cited by 1,364 · cited 1×
“Further to this, some have even discussed the potential of using LLMs as participants in cognitive experiments instead of humans [17], although some see this proposal as too optimistic [18], and others warn against excessive anthropomorphism [19].”
Reflexion: Language Agents with Verbal Reinforcement Learning
Shinn, N., Cassano, F., Berman, E., Gopinath, A., Narasimhan, K., & Yao, S. (2023) · arXiv · doi:10.48550/arxiv.2303.11366 · Cited by 1,087 · cited 1×
“Others have focused on specific qualities of human decision-making that are not possessed by LLMs, namely the ability to reflect and learn from mistakes, and propose an approach using verbal reinforcement to address this limitation [27].”
BackgroundLaMDA: Language Models for Dialog Applications
Thoppilan, R., De Freitas, D., Hall, J., Shazeer, N., Kulshreshtha, A., Cheng, H.-T., Jin, A., Bos, T., Baker, L., Du, Y., et al. (2022) · arXiv · doi:10.48550/arxiv.2201.08239 · Cited by 778 · cited 1×
“5 [40] and GPT-4 [41], Google’s Bard powered by LaMDA [42], Anthropic’s Claude 2 [43], and three versions of Meta’s Llama 2 model: the 7 billion (7b), 13 billion (13b) and 70 billion (70b) parameter versions [44].”
MethodUsing cognitive psychology to understand GPT-3
Binz, M., & Schulz, E. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2218523120 · Cited by 559 · cited 6×
“A number of studies have taken a similar approach, however they do not generally compare across different model types [12, 16, 29–35], or those that do are not evaluating rational reasoning [36].”
InfluentialCan AI language models replace human participants?
Dillion, D., Tandon, N., Gu, Y., & Gray, K. (2023) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2023.04.008 · Cited by 405 · cited 1×
“Further to this, some have even discussed the potential of using LLMs as participants in cognitive experiments instead of humans [17], although some see this proposal as too optimistic [18], and others warn against excessive anthropomorphism [19].”
Emergent analogical reasoning in large language models
Webb, T., Holyoak, K. J., & Lu, H. (2023) · Nature Human Behaviour · doi:10.1038/s41562-023-01659-w · Cited by 404 · cited 1×
“Other studies that find the reasoning of LLMs to outperform that of humans includes Chen et al.’s [33] assessment of the economic rationality of GPT, and Webb et al.’s [34] comparison of GPT-3 and human performance on analogical tasks.”
Large pre-trained language models contain human-like biases of what is right and wrong to do
Schramowski, P., Turan, C., Andersen, N., Rothkopf, C. A., & Kersting, K. (2022) · Nature Machine Intelligence · doi:10.1038/s42256-022-00458-8 · Cited by 291 · cited 3×
“Given the data that they are trained on, LLMs naturally contain human-like biases [24–26].”
Hagendorff, T., Fabi, S., & Kosinski, M. (2023) · Nature Computational Science · doi:10.1038/s43588-023-00527-x · Cited by 243 · cited 4×
“Building on this debate and looking at LLMs being evaluated using human tests, Hagendorff et al. [16] have proposed the creation of a new field of research called machine psychology , which would treat LLMs as participants in psychological experiments.”
InfluentialThe emergence of economic rationality of GPT
Chen, Y., Liu, T. X., Shan, Y., & Zhong, S. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2316205120 · Cited by 164 · cited 1×
“Other studies that find the reasoning of LLMs to outperform that of humans includes Chen et al.’s [33] assessment of the economic rationality of GPT, and Webb et al.’s [34] comparison of GPT-3 and human performance on analogical tasks.”
Performance vs. competence in human–machine comparisons
Firestone, C. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1905334117 · Cited by 150 · cited 1×
“In using methods designed to evaluate human reasoning, it is important to acknowledge the performance vs. competence debate [14].”
Can language models learn from explanations in context?
Lampinen, A., Dasgupta, I., Chan, S., Mathewson, K., Tessler, M., Creswell, A., McClelland, J., Wang, J., & Hill, F. (2022) · Findings of the Association for Computational Linguistics: EMNLP 2022 · doi:10.18653/v1/2022.findings-emnlp.38 · Cited by 142
Large language models show human-like content biases in transmission chain experiments
Acerbi, A., & Stubbersfield, J. M. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2313790120 · Cited by 137
Language Model Behavior: A Comprehensive Survey
Chang, T. A., & Bergen, B. K. (2024) · Computational Linguistics · doi:10.1162/coli_a_00492 · Cited by 105 · cited 1×
“As we will see, these models still exhibit biases and produce information that is not factual [1].”
XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models
Röttger, P., Kirk, H., Vidgen, B., Attanasio, G., Bianchi, F., & Hovy, D. (2024) · Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) · doi:10.18653/v1/2024.naacl-long.301 · Cited by 77
Language models, like humans, show content effects on reasoning tasks
Lampinen, A. K., Dasgupta, I., Chan, S. C. Y., Sheahan, H. R., Creswell, A., Kumaran, D., McClelland, J. L., & Hill, F. (2024) · PNAS Nexus · doi:10.1093/pnasnexus/pgae233 · Cited by 69
Ruis, L., Khan, A., Biderman, S., Hooker, S., Rocktäschel, T., & Grefenstette, E. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-0913 · Cited by 47
Does ChatGPT have Theory of Mind?
Holterman, B., & van Deemter, K. (2023) · arXiv · doi:10.48550/arxiv.2305.14020 · Cited by 35
Lampinen, A. (2024) · Computational Linguistics · doi:10.1162/coli_a_00525 · Cited by 32 · cited 1×
“Lampinen[15] discusses this problem when it comes to language models in particular, highlighting that different approaches must be taking to evaluate cognitive and foundation models.”
BackgroundLampinen, A. K. (2022) · arXiv · doi:10.48550/arxiv.2210.15303 · Cited by 32
XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models
Röttger, P., Kirk, H. R., Vidgen, B., Attanasio, G., Bianchi, F., & Hovy, D. (2023) · arXiv · doi:10.48550/arxiv.2308.01263 · Cited by 6 · cited 1×
“Röttger et al.[47] claim that in some cases we may have gone too far in trying to prevent models from responding to unsafe prompts and refer to exaggerated safety , showing that LLMs refuse to comply with safe requests when they include language that is sensitive or may be included in unsafe…”
Background
Cited by
On the creativity of large language models
Franceschelli, G., & Musolesi, M. (2025) · AI & SOCIETY · doi:10.1007/s00146-024-02127-3 · Cited by 113
Self-citationHagendorff, T., Dasgupta, I., Binz, M., Chan, S. C. Y., Lampinen, A., Wang, J. X., Akata, Z., & Schulz, E. (2023) · arXiv · doi:10.48550/arxiv.2303.13988 · Cited by 68
Inherent and emergent liability issues in LLM-based agentic systems: a principal-agent perspective
Gabison, G. A., & Xian, R. P. (2025) · Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025) · doi:10.18653/v1/2025.realm-1.9 · Cited by 19
Niu, Q., Liu, J., Bi, Z., Feng, P., Peng, B., Chen, K., Li, M., Yan, L. K., Zhang, Y., Yin, C. H., et al. (2024) · arXiv · doi:10.48550/arxiv.2409.02387 · Cited by 16
The bewitching AI: The Illusion of Communication with Large Language Models
Bottazzi Grifoni, E., & Ferrario, R. (2025) · Philosophy & Technology · doi:10.1007/s13347-025-00893-6 · Cited by 15
Metadata
- Authors
- Macmillan-Scott, O., & Musolesi, M.
- Venue
- Royal Society Open Science
- Publisher
- The Royal Society
- Published
- 2024-06
- Volume
- 11
- Issue
- 6
- Pages
- 240255-240255
- ISSN
- 2054-5703 (ISSN-L)
- OpenAlex
- W4399356353
- Cited by
- 47
- References
- 26
- FWCI
- 19.40
- Type
- journal article
- Language
- en
- Open access
- gold
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.