Pith. sign in

REVIEW 6 major objections 6 minor 49 references

Pencils to Pixels: A Systematic Study of Creative Drawings across Children, Adults and AI

T0 review · 6 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Who draws most creatively depends on who does the judging.

desk verdict Useful dataset and a real expert-vs-automated disagreement finding, but the cross-group style and automated-score claims are provisional until rendering artifacts are ruled out. read the letter →

arxiv 2502.05999 v1 pith:34ENQ2QE submitted 2025-02-09 cs.HC cs.AIcs.CV

Surabhi S Nath , Guiomar del Cuvillo y Schröder , Claire E. Stevenson This is my paper · ORCID
classification cs.HCcs.AIcs.CV
keywords visualcreativitydrawingschildrenadultsDall-ecomputationalmeasuresMTCIexpertvsautomatedscoring
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether visual creativity can be measured computationally across children, adults, and text-to-image AI, using a newly curated set of 1,338 drawings made on the Multi-trial Creative Ideation (MTCI) shape-completion task. It separates drawings into style (ink density, ink distribution, number of components and lines) and content (semantic themes, caption uniqueness, distance from the base shape), then compares the three groups and builds regression models of creativity ratings. The core empirical result is that every group is distinguishable: children draw more separate components, AI draws denser, more line-heavy images, and adults cover the widest conceptual range. The paper's most consequential finding is that expert human raters and two automated scoring tools disagree substantially (ICC 0.48, Spearman 0.36) about which drawings are creative, with experts ranking schematic children highest and automated tools ranking AI highest. The authors present this as the first framework for studying human and artificial creativity beyond text.

What carries the argument

The central object is the 1,338-drawing MTCI dataset, gathered from two child age groups, adults, and a text-to-image model under three prompts, all preprocessed to align size, background, and line thickness. The framework that carries the argument has four style measures (ink density, ink fraction inside the stimulus, number of connected components, number of detected straight lines), content measures built from image embeddings and model-generated captions (distance from the base shape, distance to ten nearest neighbors, hierarchical theme clusters), expert concept annotations, and four creativity scorers: two expert raters and two automated tools. Linear mixed-effects models then map style and content features onto expert versus automated scores, exposing which features each rating method values.

What would settle it

Recompute the four style measures after rendering all 1,338 drawings through a common pipeline with identical stroke texture, width, and resolution; if the group differences in ink density or line count vanish, the paper's style-based creativity comparisons collapse.

Watch

Extended reading notes

Core claim

The paper claims that a content/style decomposition plus computational measures can quantify visual creativity across intelligent agents, and that the answer to "who is more creative" depends on who does the scoring. Human experts and automated tools both rate reliably within their own kind, but they diverge sharply between kinds: experts favor drawings that use the stimulus shape and remain interpretable, while automated tools reward high ink density and image-space novelty. Consequently, on the same 1,338 drawings, experts rated schematic children most creative and AI least creative, whereas automated tools rated AI most creative and adults least creative. The paper also documents group-level differences in style and content, including children's many separate components, AI's dense and line-dominated output, and adults' broader conceptual diversity, and it offers simple regression models that predict expert and automated scores from style and content features.

Load-bearing premise

The comparison assumes that after cropping, resizing, binarizing, eroding pencil spots, and dilating lines, any remaining style differences reflect creativity rather than source-specific rendering artifacts; only line thickness is explicitly validated.

Editorial extensions

If this is right

  • Automated creativity scores should not be treated as a neutral ground truth for comparing humans and AI, because they systematically rank AI above humans while expert raters do the opposite.
  • Style differences are measurable and group-specific, so any fair comparison of drawing creativity across sources must control for rendering artifacts before interpreting style as a creativity signal.
  • Expert scoring depends on utility signals such as stimulus use and interpretability, which the automated tools ignore, meaning automated models capture only part of the creativity construct.
  • Within children, flexibility rises sharply from the pre-schematic (4-6 years) to schematic (7-9 years) stage, and within AI, content-based prompting raises flexibility and conceptual diversity.
  • The complete absence of overlap between the top-rated drawings by experts and by automated tools indicates the two rating approaches use very different strategies.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the evaluator-dependence result holds, then any claim that one group is 'more creative' must specify the rating method, and cross-agent creativity rankings lose meaning without that qualifier.
  • The disagreement between expert and automated scores may partly reflect that the automated tools were trained on adult drawings, which the paper's correlation pattern (highest agreement for adults, lowest for AI) supports but does not definitively prove.
  • A direct extension the authors do not pursue is applying the same style/content decomposition to other creative outputs, such as music notation or 3D sketches, to test whether the evaluator-dependence pattern generalizes beyond drawings.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

6 major / 6 minor

Summary. The paper curates a new dataset of 1,338 drawings on the MTCI creative-drawing task — 444 by children (pre-schematic and schematic), 444 by adults, and 450 by DALL-E v2 under three prompts — and proposes computational style measures (ink density, ink-inside-stimulus fraction, number of components, number of Hough lines) and content measures (CLIP distance to the base stimulus, caption-embedding 10-nearest-neighbor distance, expert-annotated categories). Each drawing was scored by two expert raters and two automated tools (AuDrA and OSC-figural). The main findings are: (1) group differences in style and content, with children drawing more components, AI drawings having higher ink density and more straight lines, and adults showing the highest conceptual diversity; (2) only moderate agreement between expert and automated creativity ratings (ICC 0.48, Spearman 0.36), with experts rating schematic children highest and automated tools rating AI highest; and (3) linear mixed-effects models in which expert scores are driven by interpretability, utility, and text-space novelty, whereas automated scores are driven by ink density and image-space novelty. The paper claims to provide the first framework for studying human and artificial creativity beyond the textual modality.

Significance. The dataset and the release of scripts are the clearest strengths: 1,338 drawings with expert annotations, captions, and four creativity scores per drawing, on a standardized task, made publicly available for reuse. The study also models several good practices: it reports inter-rater reliabilities (ICC 0.82 for experts, 0.90 for automated tools), validates line-thickness parity after preprocessing, checks GPT4o captions (83% accuracy), and uses cross-validated regression with BIC-based model selection rather than zero-order correlations only. If the rendering-invariance and shared-method-variance concerns raised below are addressed, the framework would be a useful benchmark for visual-creativity research, and the documented expert-automated disagreement is a genuine, field-relevant finding for automated creativity assessment. The interpretive claims about which cues drive expert versus automated judgments currently outrun the supporting evidence and need to be re-anchored.

major comments (6)
  1. [Methods (Preprocessing); Results Sec. 1 (Style)] Only line-thickness parity is validated after preprocessing (Kruskal-Wallis p > 0.1 in Methods), yet the two headline style results — highest ink density in AI (Fig. 2a) and highest straight-line count in AI (Fig. 2d) — come from binarized rasters and a Hough skeleton whose group-specific pixel statistics are otherwise unverified. Children's graphite scans contain shading and grain, while DALL-E outputs are anti-aliased digital line art, so black-pixel counts, component segmentation, and Hough-line detection can differ by source even for identical content. A robustness analysis over binarization thresholds, kernel sizes, and Hough parameters — or control drawings made on paper and digitally from the same templates — is needed before these differences can be attributed to creative choices.
  2. [Results Sec. 3 and Table 2; Methods (Annotation; Creativity Scoring)] The expert LME is partly circular: the predictors 'used stim' and 'hard to interpret' come from the same expert annotation and utility-scoring pipeline that produced the target mean expert creativity score (Methods: Annotation, Creativity Scoring). The largest coefficient in Table 2b is 'hard to interpret' (-0.77), so the reported fit (R2_test = 0.60) and the conclusion that experts 'value utility' are inflated by shared method variance. The dissociation claim in Section 3 should be re-estimated with a purely computational predictor set or with utility measured independently of the creativity rating.
  3. [Results Sec. 1 (Content); Fig. 3] The abstract-level claim that adult drawings display maximum conceptual diversity rests on point estimates (adults ~50%, children ~30%, AI ~18% of 253 categories) reported without error bars, bootstrap intervals, rarefaction curves, or significance tests. In addition, the AI group's thematic diversity is constrained by the prompt design ('day-to-day object or scene' and 'living figures' in prompts 2 and 3), so the observed ordering may reflect task constraints rather than creative capacity. Bootstrap confidence intervals, a permutation test, and an analysis restricted to prompt 1 would make the claim supportable.
  4. [Methods (Data, AI)] The AI sample is a compliance-filtered selection: roughly two-thirds of generated images were discarded for 'ignoring the stimulus or adding colors,' whereas human non-compliance (e.g., 60% of 'R' stimulus drawings drawn inverted) was retained in the dataset. This asymmetry makes the AI group a selected, instruction-following subset and can bias all downstream group comparisons. The validity criteria (who judged them, with what reliability) should be reported, and the consequences of this filtering for the style, content, and creativity comparisons should be discussed.
  5. [Results Secs. 2-3; Figs. 2/4; Table 2b] Statistical reporting is too sparse to verify the significance claims: Figures 2 and 4 mark groups of asterisks at p < 0.01 without naming the tests or post-hoc procedures, no multiple-comparison correction is reported across the four style measures, and the LME in Table 2a uses (1 | subgroup) but no participant-level random intercept even though each child and adult contributes three non-independent drawings. Additionally, the text of Section 3 that 'dist from stim was the only predictor contributing significantly to both models' is contradicted by Table 2b, where ink density and 10NN text also carry starred coefficients in both columns; this inconsistency, and the asterisk convention itself, should be corrected.
  6. [Methods (Creativity Scoring); Results Sec. 2; Discussion] The headline misalignment is partly built into the instruments: the experts rate 'creativity' on the MTCI rubric while AuDrA and OSC-figural rate 'originality,' so the ICC of 0.48 between mean expert and mean automated scores conflates construct mismatch with evaluator disagreement. The paper does acknowledge that the automated tools were trained on adult drawings, but the Discussion's 'We confirm a self-bias' framing goes further than the data warrant, since the tools are classifiers trained on adult human drawings rather than AI agents with a self, and the within-adult Spearman of 0.66 shows that even in-distribution agreement is only moderate. Comparing automated scores with an originality-specific component of the expert rubric, or explicitly reframing the result as a construct mismatch plus out-of-distribution application, would strengthen the interpretation.
minor comments (6)
  1. [Throughout] Numerous typographical and formatting errors (e.g., 'Numbergof Lines' and 'Fractiongof Ink' in Methods, 'OCS automated scoring' versus 'OSC-figural' in Figure 1, 'gtelarge' for 'gte-large', 'diving' for 'dividing' in Annotation, 'incates' for 'indicates' in Table 1, and the GitHub URL rendered with spaces as 'pencils to pixels') require a careful copy-edit.
  2. [Methods (Preprocessing; Measures)] The binarization threshold, erosion/dilation kernel sizes, Hough transform thresholds, and skeletonization algorithm are not reported; these parameters are necessary to reproduce the style measures.
  3. [Methods (Annotation); Results Sec. 3] The 10NN text measure is computed over GPT4o captions that are 17% incorrect according to a single expert validator; the authors should report whether the semantic-uniqueness and LME results are robust to excluding incorrect captions and note the single-rater limitation.
  4. [Fig. 5] Figure 5 plots expert scores on a 0-4 scale and automated scores on a 0-1 scale without stating the scales in the caption, while Table 1 reports min-max normalized expert scores; the figure should use a consistent scale or state its scales explicitly.
  5. [Abstract; Introduction] The claim of being 'the first framework for studying human and artificial creativity beyond the textual modality' should be scoped relative to Stevenson and Chen (2023), which is already cited, and to prior computational work on drawing and sketch evaluation; a dedicated related-work section would clarify the contribution.
  6. [Methods (Annotation); Results Sec. 3] The GPT4o caption prompt instructs captioning 'hard to interpret' drawings as such, so the 'hard to interpret' feature and the 10NN text measure are not independent: hard-to-interpret drawings receive similar captions and thus artificially low 10NN distances; this dependence should be acknowledged in the interpretation of the expert model.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's empirical comparisons and statistical models are self-contained, with no prediction or derivation reducing by construction to its own inputs.

full rationale

This is an empirical, data-driven study rather than a formal derivation. The central claims—group differences in style and content, and the disagreement between expert and automated creativity scores—are computed directly from images and annotations via explicit external algorithms (binarization, Hough transform, CLIP embeddings, GPT4o captions, AuDrA, OSC-figural). The regression models are descriptive: they fit linear mixed-effects models to already-existing creativity scores and report R^2 and correlations, which is standard statistical explanation rather than a construction of the scores from the predictors. The expert model includes expert-derived predictors such as 'used stim' and 'hard to interpret,' and this raises a legitimate criterion-contamination or validity concern—experts who judge interpretability and utility may also let those judgments inform their creativity ratings—but the paper nowhere defines the creativity score as a sum of these predictors, so the model is not equivalent to its input by definition. Similarly, the automated model's use of ink density to explain automated scores is an empirical characterization of those tools, not a derivation of their outputs. The preprocessing invariance limitations (binarization, erosion, dilation) are threats to measurement validity and generalizability, not circularity. The self-citations for number-of-components and flexibility are method borrowings with independent computational definitions, and no load-bearing argument reduces to an unverified self-citation. Therefore no circular step meets the required evidence standard.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claims rest on pretrained models (CLIP, GPT-4o, gte-large, AuDrA, OSC), expert judgment, and preprocessing choices rather than on fitted theoretical parameters. No new physical or conceptual entities are introduced. The listed free parameters are preprocessing and modeling choices that could affect the comparisons.

free parameters (3)
  • Binarization threshold and morphological kernel sizes = not reported
    Chosen during preprocessing to remove pencil spots and match adult line thickness; affects all style metrics, but only line thickness is statistically checked.
  • Nearest-neighbor count k = 10
    Chosen by hand for the semantic uniqueness measure; no sensitivity analysis is reported.
  • LME standardized beta coefficients = reported in Table 2b, e.g., ink density 0.44 for automated scores
    Fitted to expert and automated creativity scores to support claims about what raters value; standard supervised model fitting rather than a theoretical derivation.
assumptions (5)
  • domain assumption CLIP image embeddings separate semantic content from stylistic rendering.
    Methods/Content states that the distance measures are 'largely influenced by semantic rather than stylistic attributes', citing Udandarao et al. and Rashtchian et al.
  • domain assumption GPT4o-generated captions are accurate content descriptors for young children's and AI drawings.
    Methods/Annotation reports 83% expert validation but still uses the captions for clustering and semantic uniqueness.
  • domain assumption AuDrA and OSC-figural scores are meaningful automated creativity judgments when applied to children and AI.
    Methods/Creativity Scoring and Discussion acknowledge that the tools were trained on adult drawings and generalize poorly, yet the paper still uses their outputs as automated creativity ratings.
  • domain assumption DALL-E inpainting mode with the three prompts is a fair proxy for a human performing the MTCI task.
    Methods/AI: 1,350 images were generated to obtain 450 that followed instructions, and the model's task comprehension is not otherwise validated.
  • domain assumption Preprocessing equalizes source differences so style measures are comparable across pencil, digital pen, and DALL-E outputs.
    Methods/Preprocessing verifies line thickness parity but does not test other rendering artifacts that could drive ink density and line counts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pencils to Pixels: A Systematic Study of Creative Drawings across Children, Adults and AI." pith.science (2026). https://pith.science/paper/34ENQ2QE

@misc{pith2026250205999,
  author       = {Pith},
  title        = {Pith review of: Pencils to Pixels: A Systematic Study of Creative Drawings across Children, Adults and AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/34ENQ2QE}},
  note         = {Machine review of arXiv:2502.05999}
}
read the original abstract

Can we derive computational metrics to quantify visual creativity in drawings across intelligent agents, while accounting for inherent differences in technical skill and style? To answer this, we curate a novel dataset consisting of 1338 drawings by children, adults and AI on a creative drawing task. We characterize two aspects of the drawings -- (1) style and (2) content. For style, we define measures of ink density, ink distribution and number of elements. For content, we use expert-annotated categories to study conceptual diversity, and image and text embeddings to compute distance measures. We compare the style, content and creativity of children, adults and AI drawings and build simple models to predict expert and automated creativity scores. We find significant differences in style and content in the groups -- children's drawings had more components, AI drawings had greater ink density, and adult drawings revealed maximum conceptual diversity. Notably, we highlight a misalignment between creativity judgments obtained through expert and automated ratings and discuss its implications. Through these efforts, our work provides, to the best of our knowledge, the first framework for studying human and artificial creativity beyond the textual modality, and attempts to arrive at the domain-agnostic principles underlying creativity. Our data and scripts are available on GitHub.

Figures

Figures reproduced from arXiv: 2502.05999 by the authors.

Figure 1
Figure 1. Overview of task and framework for studying creative drawing. The three MTCI stimuli G, I, R are shown on the top [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparing style measures of ink density, ink frac [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Conceptual themes present in children, adult and AI drawings. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Comparing drawing flexibility across subgroups. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Top three drawings per subgroup receiving the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 40 canonical work pages

  1. [1]

    write newline

    " write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot := field num.names 'numnames := numnames 'format.num.names := format.num.names na...

  2. [2]

    APACrefauthors \ 2023

    acar2023creativity APACrefauthors Acar, S. APACrefauthors \ 2023 . Creativity assessment, research, and practice in the age of artificial intelligence Creativity assessment, research, and practice in the age of artificial intelligence . Creativity Research Journal 1--7

    work page 2023

  3. [3]

    , Organisciak, P

    acar2023autofigural APACrefauthors Acar, S. , Organisciak, P. \ Dumas, D. APACrefauthors \ 2023 . Automated scoring of figural tests of creativity with computer vision Automated scoring of figural tests of creativity with computer vision . The Journal of Creative Behavior

    work page 2023

  4. [4]

    APACrefauthors \ 2004

    ardila2004cavedrawings APACrefauthors Ardila, A. APACrefauthors \ 2004 . There is not any specific brain area for writing: From cave-paintings to computers There is not any specific brain area for writing: From cave-paintings to computers . International Journal of Psychology 39 1 61--67

    work page 2004

  5. [5]

    , Defranceschi, B

    augustin2011neural APACrefauthors Augustin, M D. , Defranceschi, B. , Fuchs, H K. , Carbon, C C. \ Hutzler, F. APACrefauthors \ 2011 . The neural time course of art perception: An ERP study on the processing of style versus content in art The neural time course of art perception: An erp study on the processing of style versus content in art . Neuropsychol...

    work page 2011

  6. [6]

    , Leder, H

    augustin2008style APACrefauthors Augustin, M D. , Leder, H. , Hutzler, F. \ Carbon, C C. APACrefauthors \ 2008 . Style follows content: On the microgenesis of art perception Style follows content: On the microgenesis of art perception . Acta psychologica 128 1 127--138

    work page 2008

  7. [7]

    APACrefauthors \ 2018

    barbot2018dynamics APACrefauthors Barbot, B. APACrefauthors \ 2018 . The dynamics of creative ideation: Introducing a new assessment paradigm The dynamics of creative ideation: Introducing a new assessment paradigm . Frontiers in psychology 9 2529

    work page 2018

  8. [8]

    g” in creativity? A specialization--differentiation hypothesis Where is the “g

    barbot2015domaingen APACrefauthors Barbot, B. \ Tinio, P P. APACrefauthors \ 2015 . Where is the “g” in creativity? A specialization--differentiation hypothesis Where is the “g” in creativity? a specialization--differentiation hypothesis \ ( 8). Frontiers Media SA

    work page 2015

Show all 49 references
  1. [9]

    \ Johnson, D R

    beaty2021semdis APACrefauthors Beaty, R E. \ Johnson, D R. APACrefauthors \ 2021 . Automating creativity assessment with SemDis: An open platform for computing semantic distance Automating creativity assessment with semdis: An open platform for computing semantic distance . Be...

    2021

  2. [10]

    , Lespinasse, F

    bellemare2024divergent APACrefauthors Bellemare-Pepin, A. , Lespinasse, F. , Th \"o lke, P. , Harel, Y. , Mathewson, K. , Olson, J A. Jerbi, K. APACrefauthors \ 2024 . Divergent Creativity in Humans and Large Language Models Divergent creativity in humans and large language mo...

    2024

  3. [11]

    \ Zhao, Y

    chan2010drawingskill APACrefauthors Chan, D W. \ Zhao, Y. APACrefauthors \ 2010 . The relationship between drawing skill and artistic creativity: do age and artistic involvement make a difference? The relationship between drawing skill and artistic creativity: do age and artis...

    2010

  4. [12]

    \ Marrone, R L

    cropley2022automated APACrefauthors Cropley, D H. \ Marrone, R L. APACrefauthors \ 2022 . Automated scoring of figural creativity using a convolutional neural network. Automated scoring of figural creativity using a convolutional neural network. Psychology of Aesthetics, Creat...

    2022

  5. [13]

    , Benedek, M

    diedrich2015creative APACrefauthors Diedrich, J. , Benedek, M. , Jauk, E. \ Neubauer, A C. APACrefauthors \ 2015 . Are creative ideas novel and useful? Are creative ideas novel and useful? Psychology of aesthetics, creativity, and the arts 9 1 35

    2015

  6. [14]

    , Bainbridge, W A

    fan2023drawing APACrefauthors Fan, J E. , Bainbridge, W A. , Chamberlain, R. \ Wammes, J D. APACrefauthors \ 2023 . Drawing as a versatile cognitive tool Drawing as a versatile cognitive tool . Nature Reviews Psychology 2 9 556--568

    2023

  7. [15]

    , Chi, Z

    hao2024optimizing APACrefauthors Hao, Y. , Chi, Z. , Dong, L. \ Wei, F. APACrefauthors \ 2024 . Optimizing prompts for text-to-image generation Optimizing prompts for text-to-image generation . Advances in Neural Information Processing Systems 36

    2024

  8. [16]

    , Kosoy, E

    hart2022development APACrefauthors Hart, Y. , Kosoy, E. , Liquin, E G. , Leonard, J A. , Mackey, A P. \ Gopnik, A. APACrefauthors \ 2022 . The development of creative search strategies The development of creative search strategies . Cognition 225 105102

    2022

  9. [17]

    APACrefauthors \ 1988

    heard1988children APACrefauthors Heard, D. APACrefauthors \ 1988 . Children's drawing styles Children's drawing styles . Studies in Art Education 29 4 222--231

    1988

  10. [18]

    , Stevenson, C

    ismayilzada2024evaluating APACrefauthors Ismayilzada, M. , Stevenson, C. \ van der Plas, L. APACrefauthors \ 2024 . Evaluating Creative Short Story Generation in Humans and Large Language Models Evaluating creative short story generation in humans and large language models . a...

    2024 arXiv

  11. [19]

    \ Baer, J

    kaufman2012beyond APACrefauthors Kaufman, J C. \ Baer, J. APACrefauthors \ 2012 . Beyond new and appropriate: Who decides what is creative? Beyond new and appropriate: Who decides what is creative? Creativity Research Journal 24 1 83--91

    2012

  12. [20]

    APACrefauthors \ 2019

    kenett2019semdis APACrefauthors Kenett, Y N. APACrefauthors \ 2019 . What can quantitative measures of semantic distance tell us about creativity? What can quantitative measures of semantic distance tell us about creativity? Current Opinion in Behavioral Sciences 27 11--16

    2019

  13. [21]

    , Sanakoyeu, A

    kotovenko2019content APACrefauthors Kotovenko, D. , Sanakoyeu, A. , Lang, S. \ Ommer, B. APACrefauthors \ 2019 . Content and style disentanglement for artistic style transfer Content and style disentanglement for artistic style transfer . Proceedings of the IEEE/CVF internatio...

    2019

  14. [22]

    \ Ewing, R

    latham2018children APACrefauthors Latham, G. \ Ewing, R. APACrefauthors \ 2018 . Children’s images of imagination: The language of drawings Children’s images of imagination: The language of drawings . The Australian Journal of Language and Literacy 41 2 71--81

    2018

  15. [23]

    , Davis, B

    laurito2024ai APACrefauthors Laurito, W. , Davis, B. , Grietzer, P. , Gaven c iak, T. , B \"o hm, A. \ Kulveit, J. APACrefauthors \ 2024 . AI AI Bias: Large Language Models Favor Their Own Generated Content Ai ai bias: Large language models favor their own generated content . ...

    2024 arXiv

  16. [24]

    \ Bus, A G

    levin2003drawingprecedes APACrefauthors Levin, I. \ Bus, A G. APACrefauthors \ 2003 . How is emergent writing based on drawing? Analyses of children's products and their sorting by children and mothers. How is emergent writing based on drawing? analyses of children's products ...

    2003

  17. [25]

    liang2024rich APACrefauthors Liang, Y. , He, J. , Li, G. , Li, P. , Klimovskiy, A. , Carolan, N. others APACrefauthors \ 2024 . Rich human feedback for text-to-image generation Rich human feedback for text-to-image generation . Proceedings of the IEEE/CVF Conference on Compute...

    2024

  18. [26]

    APACrefauthors \ 1957

    lowenfeld1957drawingstages APACrefauthors Lowenfeld, V. APACrefauthors \ 1957 . Creative and mental growth Creative and mental growth . Macmillan

    1957

  19. [27]

    , Park, J

    magni2024humans APACrefauthors Magni, F. , Park, J. \ Chao, M M. APACrefauthors \ 2024 . Humans as creativity gatekeepers: Are we biased against AI creativity? Humans as creativity gatekeepers: Are we biased against ai creativity? Journal of Business and Psychology 39 3 643--656

    2024

  20. [28]

    , Rello, L

    marco2024small APACrefauthors Marco, G. , Rello, L. \ Gonzalo, J. APACrefauthors \ 2024 . Small language models can outperform humans in short creative writing: A study comparing slms with humans and llms Small language models can outperform humans in short creative writing: A...

    2024 arXiv

  21. [29]

    APACrefauthors \ 2023

    marr2023intersection APACrefauthors Marr, B. APACrefauthors \ 2023 . The intersection of AI and human creativity: can machines really be creative? The intersection of ai and human creativity: can machines really be creative? Forbes. Verkkosivu https://www. forbes. com/sites/be...

    2023

  22. [30]

    , Polsley, S

    narvaez2024children APACrefauthors Narvaez, R. , Polsley, S. \ Hammond, T. APACrefauthors \ 2024 . Children's Drawings Speak: Comparing Sketch Features amongst Development Groups Children's drawings speak: Comparing sketch features amongst development groups . Proceedings of t...

    2024

  23. [31]

    , Br \"a ndle, F

    nath2024relating APACrefauthors Nath, S S. , Br \"a ndle, F. , Schulz, E. , Dayan, P. \ Brielmann, A. APACrefauthors \ 2024 . Relating objective complexity, subjective complexity, and beauty in binary pixel patterns. Relating objective complexity, subjective complexity, and be...

    2024

  24. [32]

    , Dayan, P

    nath2024creaprocess APACrefauthors Nath, S S. , Dayan, P. \ Stevenson, C. APACrefauthors \ 2024 . Characterising the Creative Process in Humans and Large Language Models Characterising the creative process in humans and large language models . Proceedings of the 15th Internati...

    2024 arXiv

  25. [33]

    , Linder, R

    oppenlaender2024prompting APACrefauthors Oppenlaender, J. , Linder, R. \ Silvennoinen, J. APACrefauthors \ 2024 . Prompting AI art: An investigation into the creative skill of prompt engineering Prompting ai art: An investigation into the creative skill of prompt engineering ....

    2024

  26. [34]

    \ Horv \'a t, E \'A

    o2024extending APACrefauthors O'Toole, K. \ Horv \'a t, E \'A . APACrefauthors \ 2024 . Extending human creativity with AI Extending human creativity with ai . Journal of Creativity 34 2 100080

    2024

  27. [35]

    , Barbot, B

    patterson2024audra APACrefauthors Patterson, J D. , Barbot, B. , Lloyd-Cox, J. \ Beaty, R E. APACrefauthors \ 2024 . AuDrA: An automated drawing assessment platform for evaluating creativity Audra: An automated drawing assessment platform for evaluating creativity . Behavior R...

    2024

  28. [36]

    , Tsuji, S

    philippsen2022quantifying APACrefauthors Philippsen, A. , Tsuji, S. \ Nagai, Y. APACrefauthors \ 2022 . Quantifying developmental and individual differences in spontaneous drawing completion among children Quantifying developmental and individual differences in spontaneous dra...

    2022

  29. [37]

    \ Sanampudi, S K

    ramesh2022automated APACrefauthors Ramesh, D. \ Sanampudi, S K. APACrefauthors \ 2022 . An automated essay scoring systems: a systematic literature review An automated essay scoring systems: a systematic literature review . Artificial Intelligence Review 55 3 2495--2527

    2022

  30. [38]

    , Herrmann, C

    rashtchian2023substance APACrefauthors Rashtchian, C. , Herrmann, C. , Ferng, C S. , Chakrabarti, A. , Krishnan, D. , Sun, D. Tomkins, A. APACrefauthors \ 2023 . Substance or Style: What Does Your Image Embedding Know? Substance or style: What does your image embedding know? a...

    2023 arXiv

  31. [39]

    \ Acar, S

    runco2012uniqueness APACrefauthors Runco, M A. \ Acar, S. APACrefauthors \ 2012 . Divergent thinking as an indicator of creative potential Divergent thinking as an indicator of creative potential . Creativity research journal 24 1 66--75

    2012

  32. [40]

    \ Taylor, J

    spensley1999development APACrefauthors Spensley, F. \ Taylor, J. APACrefauthors \ 1999 . The development of cognitive flexibility: Evidence from children’s drawings The development of cognitive flexibility: Evidence from children’s drawings . Human Development 42 6 300--324

    1999

  33. [41]

    \ Chen, S

    Stevenson_Chen_2023 APACrefauthors Stevenson, C. \ Chen, S. APACrefauthors \ 2023 Dec . Visual Creativity in AI. Visual creativity in ai. OSF . APACrefURL osf.io/4j5nk APACrefURL

    2023

  34. [42]

    , Powell, L

    thomas2022identifying APACrefauthors Thomas, X. , Powell, L. , Polsley, S. , Ray, S. \ Hammond, T. APACrefauthors \ 2022 . Identifying Features that Characterize Children’s Free-Hand Sketches using Machine Learning Identifying features that characterize children’s free-hand sk...

    2022

  35. [43]

    APACrefauthors \ 1966

    torrance1966TTCT APACrefauthors Torrance, E P. APACrefauthors \ 1966 . Torrance tests of creative thinking Torrance tests of creative thinking . Educational and psychological measurement

    1966

  36. [44]

    , Burg, M F

    udandarao2023visual APACrefauthors Udandarao, V. , Burg, M F. , Albanie, S. \ Bethge, M. APACrefauthors \ 2023 . Visual Data-Type Understanding does not emerge from Scaling Vision-Language Models Visual data-type understanding does not emerge from scaling vision-language model...

    2023

  37. [45]

    APACrefauthors \ 2005

    urban2005TCT-DP APACrefauthors Urban, K K. APACrefauthors \ 2005 . Assessing Creativity: The Test for Creative Thinking-Drawing Production (TCT-DP). Assessing creativity: The test for creative thinking-drawing production (tct-dp). International Education Journal 6 2 272--280

    2005

  38. [46]

    \ Wagemans, J

    van2020order APACrefauthors Van Geert, E. \ Wagemans, J. APACrefauthors \ 2020 . Order, complexity, and aesthetic appreciation. Order, complexity, and aesthetic appreciation. Psychology of aesthetics, creativity, and the arts 14 2 135

    2020

  39. [47]

    , Shaham, T R

    vinker2024sketchagent APACrefauthors Vinker, Y. , Shaham, T R. , Zheng, K. , Zhao, A. , Fan, J E. \ Torralba, A. APACrefauthors \ 2024 . SketchAgent: Language-Driven Sequential Sketch Generation Sketchagent: Language-driven sequential sketch generation . arXiv preprint arXiv:2...

    2024 arXiv

  40. [48]

    , Caldwell, B W

    weaver2019interpreting APACrefauthors Weaver, M B. , Caldwell, B W. \ Sheafer, V. APACrefauthors \ 2019 . Interpreting measures of rarity and novelty: investigating correlations between relative infrequency and perceived ratings Interpreting measures of rarity and novelty: inv...

    2019

  41. [49]

    , Zhang, Y

    zhang2018separating APACrefauthors Zhang, Y. , Zhang, Y. \ Cai, W. APACrefauthors \ 2018 . Separating style and content for generalized style transfer Separating style and content for generalized style transfer . Proceedings of the IEEE conference on computer vision and patter...

    2018

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.