CoChat

Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT

Open access
Hagendorff, Thilo, Fabi, Sarah, Kosinski, Michal · 2023 · Nature Computational Science

The DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index.

Abstract

We design a battery of semantic illusions and cognitive reflection tests, aimed to elicit intuitive yet erroneous responses. We administer these tasks, traditionally used to study reasoning and decision-making in humans, to OpenAI’s generative pre-trained transformer model family. The results show that as the models expand in size and linguistic proficiency they increasingly display human-like intuitive system 1 thinking and associated cognitive errors. This pattern shifts notably with the introduction of ChatGPT models, which tend to respond correctly, avoiding the traps embedded in the tasks. Both ChatGPT-3.5 and 4 utilize the input–output context window to engage in chain-of-thought reasoning, reminiscent of how people use notepads to support their system 2 thinking. Yet, they remain accurate even when prevented from engaging in chain-of-thought reasoning, indicating that their system-1-like next-word generation processes are more accurate than those of older models. Our findings highlight the value of applying psychological methodologies to study large language models, as this can uncover previously undetected emergent characteristics.

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.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s43588-023-00527-x

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Human and LLM performance on the CRT tasks. a, Exemplary responses to one of the CRT tasks, categorized as correct, intuitive (but incorrect) and atypical (that is, all other incorrect responses). Within each category, the responses that were preceded by written chain-of-thought reasoning were additionally labeled as ‘chain-of-thought responses’. b, Human and LLM performance on 150 CRT tasks. c, LLMs’ responses when instructed to engage or prevented from engaging in chain-of-thought reasoning. The data source file includes 95% confidence intervals. Source data
Fig. 1 Human and LLM performance on the CRT tasks. a, Exemplary responses to one of the CRT tasks, categorized as correct, intuitive (but incorrect) and atypical (that is, all other incorrect responses). Within each category, the responses that were preceded by written chain-of-thought reasoning were additionally labeled as ‘chain-of-thought responses’. b, Human and LLM performance on 150 CRT tasks. c, LLMs’ responses when instructed to engage or prevented from engaging in chain-of-thought reasoning. The data source file includes 95% confidence intervals. Source dataCC BY · PubMed Central
Fig. 2 Human and LLM performance on semantic illusions. a. Exemplary responses to one of the semantic illusions, categorized as correct, intuitive and atypical. b, Human and LLM performance on 50 semantic illusions. c, GPT-3-davinci-003’s responses when instructed to examine the task’s assumptions. The data source file includes 95% confidence intervals. Source data
Fig. 2 Human and LLM performance on semantic illusions. a. Exemplary responses to one of the semantic illusions, categorized as correct, intuitive and atypical. b, Human and LLM performance on 50 semantic illusions. c, GPT-3-davinci-003’s responses when instructed to examine the task’s assumptions. The data source file includes 95% confidence intervals. Source dataCC BY · PubMed Central
Extended Data Fig. 1 Learning curves. Change in the fraction of GPT-3-davinci-003’s correct responses against the number of training examples that the task was prefixed with. Error bars represent 95% confidence intervals. Source data
Extended Data Fig. 1 Learning curves. Change in the fraction of GPT-3-davinci-003’s correct responses against the number of training examples that the task was prefixed with. Error bars represent 95% confidence intervals. Source dataCC BY · PubMed Central

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.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

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

Show all 21 references

Research works printed but not linked: 5

  1. Sloman, S. A. Psychol. Bull. 119, 3–22 (1996). · DOI · Paper page
  2. Radford, A., Narasimhan, K., Salimans, T. & Sutskever, I. Improving language understanding by generative pre-training. Preprint at https://paperswithcode.com/paper/improving-language-understanding-by (2018).
  3. Radford, A. et al. Language models are unsupervised multitask learners. Preprint at https://paperswithcode.com/paper/language-models-are-unsupervised-multitask (2019).
  4. Jones, E. & Steinhardt, J. Capturing failures of large language models via human cognitive biases. 36th Conference on Neural Information Processing Systems (2022).
  5. Nye, M., Tessler, M. H., Tenenbaum, J. B. & Lake, B. M. Improving coherence and consistency in neural sequence models with dual-system, neuro-symbolic reasoning. 35th Conference on Neural Information Processing Systems (2021).

Cited by

See all 328

Metadata

Authors
Hagendorff, T., Fabi, S., & Kosinski, M.
Affiliations
  • University of Stuttgart (DE)
  • University of California San Diego (US)
Venue
Nature Computational Science
Published
2023-10-05
Volume
3
Issue
10
Pages
833-838
ISSN
2662-8457 (ISSN-L)
OpenAlex
W4387378202
Cited by
243
References
16
Citations by year
2026: 87 · 2025: 78 · 2024: 71 · 2023: 7
FWCI
22.91
Type
journal article
Language
en
Open access
Licence
Locations
Nature Computational Science (publishedVersion, cc-by)
Funders
  • Ministerium für Wissenschaft, Forschung und Kunst Baden-Württemberg (Az. 33-7533-9-19/54/5)
  • Interchange Forum for Reflecting on Intelligent Systems (IRIS) at the University of Stuttgart
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
  • Social Sciences › Decision Sciences › General Decision Sciences › Decision-Making and Behavioral Economics
Keywords
large language models, ChatGPT, intuitive thinking, cognitive biases, System 2 thinking, chain-of-thought reasoning, semantic illusions, cognitive reflection, model size, context window, emergent behavior, human-like reasoning

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT. https://app.cochat.ai/papers/10.1038/s43588-023-00527-x
The paper
Hagendorff, T., Fabi, S., & Kosinski, M. (2023). Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT. Nature Computational Science. https://doi.org/10.1038/s43588-023-00527-x

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.