CoChat

Exploring creativity in human-AI co-creation: a comparative study across design experience

Open access
Wang, Nan, Kim, Hyunsuk, Peng, Junfeng, Wang, Jiayi · 2025 · Frontiers in Computer Science

The DOI matches its Crossref record; 37 of the 37 references it cites were found in a registry or an index.

Abstract

As generative artificial intelligence (GAI) becomes increasingly integrated into the design domain, research has begun to explore how it can be meaningfully incorporated into traditional design practices, fostering the development of more collaborative design processes. This study proposes a Human–AI Co-Creative Design Process (HAI-CDP) model and evaluates its impact on designers’ creativity through a comparative experimental design. The results indicate that the HAI-CDP substantially improves creative performance over the traditional design process. For novice designers, its primary value lies in facilitating idea generation, whereas for experienced designers, it contributes more to elevating the quality and refinement of creative outcomes. Although the Human–AI Co-Creative Design Process lowers the entry barrier to creative engagement, the findings also reaffirm that design experience remains a critical factor shaping creative output.

Overview

CoChat Summary

TL;DR

This study proposes a Human–AI Co-Creative Design Process (HAI-CDP) model, built from interviews with 13 professional designers and the Double Diamond framework, and tests it against a Traditional Creative Design Process (TCDP) in a 2 × 2 experiment with 42 designers. The HAI-CDP significantly improved creative performance on every measured dimension, but its benefits differed by experience: novices gained most in idea generation, while experienced designers gained most in quality and refinement.

Key findings

  • HAI-CDP outperformed TCDP across all five creativity dimensions. For novices, significant gains appeared in creative support capability (t = 8.645, p < 0.001), quality (t = 13.974, p < 0.001), refinement (t = 19.481, p < 0.001), and number of ideas (t = 8.919, p < 0.001); for experienced designers, the largest gains were quality (t = 21.359, p < 0.001) and refinement (t = 21.855, p < 0.001), with novelty also significant (t = 17.426, p < 0.001).
  • Experience effects persisted under HAI-CDP: experienced designers still scored significantly higher on novelty (t = 6.526, p < 0.001), quality (t = 6.145, p < 0.001), refinement (t = 13.099, p < 0.001), and number of ideas (t = 6.460, p < 0.001), while the experience gap in perceived creative support disappeared (t = −0.591, p > 0.05).
  • Gap-reduction analysis showed HAI-CDP improved novices' creative support capability (Gap Reduction = +133.33%) but widened the experience gap in novelty (−55.09%), quality (−14.54%), and number of ideas (−78.95%).
  • Two-way ANOVA found significant main effects of design process on all dimensions (e.g., Refinement F(1, 80) = 619, p < 0.001) and a significant process × experience interaction only for number of ideas generated (F(1, 80) = 5.56, p = 0.021); no significant interactions for novelty, quality, refinement, or creative support. Reliability was Cronbach's alpha = 0.878.

Methods

  • Stage 1: semi-structured interviews with 13 professional designers (all with >3 years of project experience) in visual design and animation, analyzed via three-level coding, yielding four stages (concept definition, visual exploration, design development, implementation integration) and a Four-Diamond extension of the Double Diamond model.
  • Stage 2: a 2 × 2 factorial experiment with 42 designers (21 novices — second-year animation/visual design undergraduates, mean age 19.67, SD = 0.73; 21 experienced — >3 years' experience, mean age 28.38, SD = 1.36). Each participant completed a 48-h online character design task ("Mechanical Beings of a Future World") under both TCDP and HAI-CDP, always in that order.
  • Five dependent variables (perceived creative support capability, novelty, quality, refinement, number of ideas); novelty, quality, and refinement rated on a 5-point Likert scale by three blind expert raters (>10 years' experience), averaged; analyzed with paired-sample t-tests, independent-samples t-tests, and two-way ANOVA in SPSS 27.

Limitations

  • Stated by the authors: the participant pool was limited to animation and visual design students and designers; the relatively small sample size may constrain generalizability; and the subjective nature of creative assessments may have influenced outcomes.
  • Evident in the text: all participants completed TCDP before HAI-CDP, so order and carryover effects were not counterbalanced (the authors justify this as protecting the human-only baseline, but it remains a confound); the paper also notes a correction has been applied to the article.

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

Integrity signals

Registry

Crossref · 10.3389/fcomp.2025.1672735

Cites

37 of 37 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

0 Valid · 8 Not valid · 4 Not checked · 74 Own results and interpretations

29 more claims

Create a free account to see all 86 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 33
  • Probably not a substantial claim. 28

Create a free account to see these claims.

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.

  • ConsistentF(1, 80) = 170reported p < .001 · recomputed p < .001

    “F(1,80) = 170, p < .001”

  • ConsistentF(1, 80) = 448reported p < .001 · recomputed p < .001

    “F(1, 80) = 448, p < .001”

  • ConsistentF(1, 80) = 377reported p < .001 · recomputed p < .001

    “F(1, 80) = 377, p < .001”

  • ConsistentF(1, 80) = 619reported p < .001 · recomputed p < .001

    “F (1, 80) = 619, p < .001”

  • ConsistentF(1, 80) = 178reported p < .001 · recomputed p < .001

    “F(1, 80) = 178, p < .001”

5 of 15 · the rest aren’t loaded

10 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.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 37 references

Research works printed but not linked: 1

  1. 2005, The ‘double diamond’ design process model

Cited by

Metadata

Authors
Wang, N., Kim, H., Peng, J., & Wang, J.
Affiliations
Hongik University (KR)
Corresponding author
Kim, Hyunsuk
Venue
Frontiers in Computer Science
Published
2025-09-30
Volume
7
ISSN
2624-9898 (ISSN-L)
OpenAlex
W4414677733
Cited by
59
References
32
Citations by year
2026: 55 · 2025: 4
FWCI
16.98
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Computer Science (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Physical Sciences › Engineering › Mechanical Engineering › Design Education and Practice
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Digital Media and Visual Art
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Creativity in Education and Neuroscience
Keywords
human-AI co-creation, generative AI, design creativity, co-design, design experience, novice designers, idea generation, creative performance, design process model

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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

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

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.