Student or educator? Sign up with your school email and get 6 months of Pro free.

The DOI matches its Crossref record; 28 of the 28 references it cites were found in a registry or an index; 4 of the 6 sources its claims cite were checked for support.

Best humans still outperform artificial intelligence in a creative divergent thinking task

Open access
Koivisto, Mika, Grassini, Simone · 2023 · Scientific Reports

Abstract

Creativity has traditionally been considered an ability exclusive to human beings. However, the rapid development of artificial intelligence (AI) has resulted in generative AI chatbots that can produce high-quality artworks, raising questions about the differences between human and machine creativity. In this study, we compared the creativity of humans (n = 256) with that of three current AI chatbots using the alternate uses task (AUT), which is the most used divergent thinking task. Participants were asked to generate uncommon and creative uses for everyday objects. On average, the AI chatbots outperformed human participants. While human responses included poor-quality ideas, the chatbots generally produced more creative responses. However, the best human ideas still matched or exceed those of the chatbots. While this study highlights the potential of AI as a tool to enhance creativity, it also underscores the unique and complex nature of human creativity that may be difficult to fully replicate or surpass with AI technology. The study provides insights into the relationship between human and machine creativity, which is related to important questions about the future of creative work in the age of AI.

Overview

CoChat Summary

TL;DR

In an Alternate Uses Task with 256 humans and three AI chatbots, the chatbots scored higher on average than humans on both semantic-distance and human-rated creativity measures. However, the best human responses still matched or exceeded the best chatbot responses in most cases.

Key findings

  • Humans (n = 256) were compared with three AI chatbots on four AUT objects: rope, box, pencil, and candle; each AI was tested in 11 sessions, yielding 132 AI observations.
  • On average, AI outperformed humans: semantic-distance mean scores were higher for AI than humans, B = 0.049, SE = 0.010, CI [0.030, 0.069], p < 0.001; subjective mean scores were also higher for AI, B = 0.453, SE = 0.082, CI [0.292, 0.614], p < 0.001.
  • AI also led on max scores overall: semantic-distance max scores, B = 0.027, SE = 0.009, CI [0.010, 0.044], p = 0.003; subjective max scores, B = 0.403, SE = 0.091, CI [0.226, 0.581], p < 0.001.
  • Best-human performance remained competitive: the paper states there was only one instance where an AI had the highest semantic-distance score and two instances where AI had the highest subjective score; in all other cases, humans had the highest scores.

Methods

  • Design: comparative study using the Alternate Uses Task (AUT), with humans and three current AI chatbots (ChatGPT3.5/ChatGPT3, ChatGPT4, Copy.AI) generating uncommon uses for everyday objects.
  • Sample/data: 256 native English-speaking humans recruited via Prolific; the AI chatbots were each tested 11 times with the four object prompts in separate sessions.
  • Analysis: originality was measured two ways—semantic distance (computed with SemDis and averaged across five semantic models) and subjective creativity ratings from six blinded human raters; separate linear mixed-effects models tested group, object, fluency, and their interactions.

Limitations

  • The paper states that the number of observations per chatbot was restricted, limiting statistical power, especially for comparisons between individual chatbots.
  • The authors say they restricted chatbot output length and number of ideas to control fluency and elaboration, which they note may have purposely impaired AI performance.
  • The paper states that chatbot processes are “black boxes,” so process-level comparison with humans is impossible.
  • Evident from the text: the human sample was limited to young and middle-aged adults from Western countries, which the authors say limits generalization.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-023-40858-3

Cites

28 of 28 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 The relationship between the semantic distance scores of originality and the human-made subjective ratings for (A) the mean scores and (B) the max scores.
Figure 1 The relationship between the semantic distance scores of originality and the human-made subjective ratings for (A) the mean scores and (B) the max scores.CC BY · PubMed Central
Figure 2 Humans’ and AI’s mean scores (average of all responses within each trial) and max scores (the highest scoring response within each trial) as revealed by sematic distance analysis (A, B) and human subjective ratings (C, D).
Figure 2 Humans’ and AI’s mean scores (average of all responses within each trial) and max scores (the highest scoring response within each trial) as revealed by sematic distance analysis (A, B) and human subjective ratings (C, D).CC BY · PubMed Central
Figure 3 Mean semantic distance scores of the humans and chatbots to the four objects.
Figure 3 Mean semantic distance scores of the humans and chatbots to the four objects.CC BY · PubMed Central
Figure 4 Max semantic distance scores of the humans and the chatbots to the four objects.
Figure 4 Max semantic distance scores of the humans and the chatbots to the four objects.CC BY · PubMed Central
Figure 5 Mean scores based on subjective ratings for the humans and chatbots to the four objects.
Figure 5 Mean scores based on subjective ratings for the humans and chatbots to the four objects.CC BY · PubMed Central
Figure 6 Max scores based on subjective ratings for the humans and chatbots to the four objects.
Figure 6 Max scores based on subjective ratings for the humans and chatbots to the four objects.CC BY · PubMed Central

Verification

Claims the paper makes

3 Valid · 0 Not valid · 1 Not checked · 37 Own results and interpretations

19 more claims

Create a free account to see all 41 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 8
  • It may state two separate claims as one. 10

Create a free account to see these claims.

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?

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Not confirmedt(852) = 3.283reported p = .006 · recomputed p = .001

    “t(852) = 3.283, p = .006”

  • Consistentt(274) = 4.949reported p < .001 · recomputed p < .001

    “t(274) = 4.949, p < .001”

  • Consistentt(268) = 3.037reported p = .003 · recomputed p = .003

    “t(268) = 3.037, p = .003”

  • Consistentt(279) = -4.17reported p < .001 · recomputed p < .001

    “t(279) = −4.170, p < .001”

  • Consistentt(274) = 4.486reported p < .001 · recomputed p < .001

    “t(274) = 4.486, p < .001”

5 of 20 · the rest aren’t loaded

15 more recomputed tests

Create a free account to see every test the Stats check recomputed.

See all tests — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

1 of 1

Citations

Cites

Show all 28 references

Research works printed but not linked: 1

  1. Nadikattu, R. R. The emerging role of artificial intelligence in modern society. Int. J. Create. Res. Thoughts 4(4), 906–911 (2016).

Cited by

See all 247

Metadata

Authors
Koivisto, M., & Grassini, S.
Affiliations
  • University of Turku (FI)
  • University of Bergen (NO)
Corresponding author
Grassini, Simone
Venue
Scientific Reports
Published
2023-09-14
Volume
13
Issue
1
Pages
13601-13601
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4386740961
Cited by
247
References
28
Citations by year
2026: 83 · 2025: 101 · 2024: 59 · 2023: 3
FWCI
63.40
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • NORA - Norwegian Open Research Archives (submittedVersion, cc-by)
  • UTUPub (University of Turku) (submittedVersion, other-oa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
University of Bergen
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Creativity in Education and Neuroscience
  • Physical Sciences › Engineering › Mechanical Engineering › Design Education and Practice
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Aesthetic Perception and Analysis
Keywords
human creativity, artificial intelligence, generative AI chatbots, divergent thinking, alternative uses task, human-AI comparison
MeSH terms
Artificial Intelligence, Humans, Technology

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.

A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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: Best humans still outperform artificial intelligence in a creative divergent thinking task. https://app.cochat.ai/papers/10.1038/s41598-023-40858-3
The paper
Koivisto, M., & Grassini, S. (2023). Best humans still outperform artificial intelligence in a creative divergent thinking task. Scientific Reports. https://doi.org/10.1038/s41598-023-40858-3

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.