pith. sign in

arxiv: 2607.00239 · v1 · pith:OAJ4S4MOnew · submitted 2026-06-30 · 💻 cs.HC · cs.GR

May (A)I Beautify Your Visualization? Expert Judgments of Acceptable Aesthetic Alterations

Kalina Borkiewicz , Jixian Li , Joshua A. Levine , Katherine E. Isaacs This is my paper

Pith reviewed 2026-07-02 17:10 UTC · model grok-4.3

classification 💻 cs.HC cs.GR
keywords visualization aestheticsexpert surveyAI alterationsdata integrity3D visualizationsperceived acceptabilityhuman vs AIaesthetic modifications
0
0 comments X

The pith

Expert survey finds acceptability of visualization alterations driven by transformation meaning, with AI versions rated lower than human ones regardless of presentation or data level.

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

The paper reports results from an expert survey of 95 visualization researchers, practitioners, and domain scientists rating the acceptability of fifteen alterations to 3D visualizations. These alterations span presentation-level adjustments such as lighting and camera position and data-level modifications such as removing errors or filling gaps, each performed by either humans or AI systems. The findings indicate that perceived acceptability depends on the meaning of the transformation, with some modifications consistently judged more permissible than others in a stable ordering across authorship types. AI-generated changes receive lower ratings than identical human-produced ones. This matters as AI tools make such modifications more accessible, highlighting the need for constraints that reflect distinctions between permissible and sensitive alterations.

Core claim

Through the survey, the authors establish that perceived acceptability of alterations is driven by the transformation's meaning, independent of whether it operates at the presentation or data level. Certain modifications are consistently judged more permissible regardless of human or AI authorship, while the relative ordering remains largely stable, yet AI-generated transformations are rated less acceptable than identical human-produced changes. This reveals a distinction between more permissible and more sensitive alterations.

What carries the argument

Expert survey (N=95) rating fifteen alterations spanning presentation-level and data-level modifications applied by humans or AI.

If this is right

  • Designers should prioritize the meaning of a proposed alteration when evaluating its use in visualizations.
  • AI-assisted visualization tools require guardrails that distinguish between more permissible and more sensitive alterations.
  • The stable ordering of alterations can guide consistent decision-making across human and automated workflows.
  • Both presentation-level and data-level changes need scrutiny based on their interpretive impact rather than their technical category.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Visualization software interfaces could surface warnings or approval steps for alterations that experts rate as more sensitive.
  • The lower acceptability for AI authorship may affect how tools disclose or attribute automated edits to users.
  • Testing whether these expert patterns hold for non-expert viewers or other visualization domains would clarify the scope of the results.

Load-bearing premise

The fifteen alterations selected for the survey represent the range of aesthetic modifications in real visualization practice, and expert self-reported judgments accurately capture concerns about data perception and integrity without significant response bias.

What would settle it

A follow-up study using a different or expanded set of alterations that finds presentation versus data level as the primary driver of acceptability differences, or that finds no difference between human and AI ratings, would challenge the central claims.

Figures

Figures reproduced from arXiv: 2607.00239 by Jixian Li, Joshua A. Levine, Kalina Borkiewicz, Katherine E. Isaacs.

Figure 1
Figure 1. Figure 1: Examples of aesthetic alterations from our study. Top row: a subset of abstract transformation categories. Bottom row: [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The original planetary collision visualization and five AI [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Mean acceptability (±1 SD) for human and AI alterations, [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
read the original abstract

In 3D visualizations of natural phenomena, improving aesthetics can provide measurable benefits, but often involves transformations that affect how the data is perceived. As a growing range of tools - including AI-based methods - make visual design and modification more accessible, it is increasingly important to understand trade offs and concerns when making these changes. We conducted an expert survey (N=95) with visualization researchers, practitioners, and domain scientists, investigating reactions to fifteen alterations spanning presentation-level adjustments (e.g., lighting, camera position) and data-level modifications (e.g., removing errors, filling gaps), applied by both humans and AI systems. Results show differences in perceived acceptability are driven by the transformation's meaning, regardless of whether it operates at the presentation or data level. Additionally, certain modifications were consistently judged as more permissible than others regardless of human or AI authorship. While this relative ordering remains largely stable, AI-generated transformations are consistently rated as less acceptable than identical human-produced changes. These results reveal a distinction between more permissible and more sensitive alterations, and suggest the need for both designers and AI-assisted visualization tools to incorporate constraints and guardrails that reflect these differences.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript reports results from an expert survey (N=95) with visualization researchers, practitioners, and domain scientists. Participants rated the acceptability of 15 aesthetic alterations in 3D visualizations of natural phenomena. Alterations span presentation-level (e.g., lighting, camera) and data-level (e.g., error removal, gap filling) changes, each applied by either humans or AI. Key claims are that perceived acceptability is driven by the meaning of the transformation independent of operational level, that certain modifications show stable relative permissibility rankings across authorship, and that AI versions are consistently rated less acceptable than identical human versions.

Significance. If the methodological gaps are resolved, the work supplies empirical evidence on expert perceptions of visualization aesthetics that can inform design guidelines and guardrails for both human and AI-assisted tools. The reported dissociation between meaning and level, plus the stable ordering with an AI penalty, offers a concrete basis for distinguishing permissible from sensitive alterations in visualization practice.

major comments (3)
  1. [Methods] Methods section: The description provides no information on survey design details, statistical tests, response rates, or controls for order effects and social desirability bias. The central claims about meaning-driven acceptability independent of level and the consistent AI penalty rest on these unexamined choices.
  2. [Methods / Results] The 15 alterations lack a described sampling frame, coverage validation, or pre-registered classification by 'meaning.' Without this, the reported dissociation between meaning and presentation/data level could be an artifact of item selection rather than a general property (see weakest assumption on representativeness).
  3. [Results] Results: No effect sizes, confidence intervals, or explicit tests for the stability of relative orderings across human/AI conditions are referenced, making it difficult to evaluate the strength of the 'largely stable' ordering claim.
minor comments (1)
  1. [Abstract] Abstract: The parenthetical examples of alterations could be expanded with one sentence on how many fall into each level (presentation vs. data) to clarify the balance of the stimulus set.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below, indicating where we will revise the paper to improve clarity and rigor.

read point-by-point responses

  1. Referee: [Methods] Methods section: The description provides no information on survey design details, statistical tests, response rates, or controls for order effects and social desirability bias. The central claims about meaning-driven acceptability independent of level and the consistent AI penalty rest on these unexamined choices.

    Authors: We agree that the Methods section requires expansion to include these details. The revised manuscript will add subsections describing the survey instrument design (including question ordering and randomization), the specific statistical tests performed (e.g., mixed-effects models), response rate and completion metrics, and any steps taken to mitigate order effects or social desirability bias (such as anonymous administration and neutral phrasing). revision: yes

  2. Referee: [Methods / Results] The 15 alterations lack a described sampling frame, coverage validation, or pre-registered classification by 'meaning.' Without this, the reported dissociation between meaning and presentation/data level could be an artifact of item selection rather than a general property (see weakest assumption on representativeness).

    Authors: We acknowledge the need for greater transparency on item selection. In revision, we will add a dedicated subsection in Methods explaining the sampling frame for the 15 alterations, the process used to classify them by transformation meaning (including the rationale for balancing presentation- and data-level changes), and any coverage validation performed during pilot testing. While the classification was developed iteratively by the research team rather than pre-registered, we will document this procedure explicitly so readers can evaluate potential selection effects. revision: yes

  3. Referee: [Results] Results: No effect sizes, confidence intervals, or explicit tests for the stability of relative orderings across human/AI conditions are referenced, making it difficult to evaluate the strength of the 'largely stable' ordering claim.

    Authors: We will revise the Results section to report effect sizes (e.g., Cohen's d or partial eta-squared), 95% confidence intervals for key comparisons, and explicit statistical tests (such as rank correlation or permutation tests) assessing the stability of relative orderings between human and AI conditions. These additions will provide quantitative support for the stability claim. revision: yes

Circularity Check

0 steps flagged

No circularity: direct empirical survey with no derivations or fitted predictions

full rationale

This paper reports results from an expert survey (N=95) on acceptability of 15 visualization alterations. There are no equations, parameters, derivations, or first-principles claims that could reduce to inputs by construction. All findings derive from direct participant responses rather than any model fitting or self-referential logic. No load-bearing self-citations of theorems or ansatzes appear in the described methodology or results.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Empirical survey study with no mathematical derivations or fitted parameters; the central claim rests on the domain assumption that expert judgments are valid proxies for acceptable practice and that the chosen alterations cover the relevant design space.

axioms (1)
  • domain assumption Expert self-reported judgments on visualization alterations accurately reflect real-world concerns about data perception and integrity.
    The study uses these judgments as the sole basis for distinguishing permissible from sensitive alterations.

pith-pipeline@v0.9.1-grok · 5744 in / 1247 out tokens · 26497 ms · 2026-07-02T17:10:33.036863+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 32 canonical work pages

  1. [1]

    P. W. Anani, F. D. Mintah, A. A. A. Danso, and E. W. Churcher. Filtered reality: Exploring the motives and socio-demographic fac- tors of smartphone beauty filter usage among university students in Ghana.Cogent Arts & Humanities, 11(1):2392381, 2024. doi: 10. 1080/23311983.2024.2392381 4

    work page arXiv 2024

  2. [2]

    J. Baca, D. Carruth, M. Stephens, and C. Lewis. Evidence for effect of aesthetic on interpretation of visualizations by engineers and non- engineers. InTareq Ahram and Christianne Falc ˜ao (eds) Usability and User Experience. AHFE (2022) International Conference, vol. 39,

    2022

  3. [3]

    doi: 10.54941/ahfe1001718 1

    work page doi:10.54941/ahfe1001718

  4. [4]

    R. C. Basole and T. Major. Generative AI for visualization: Oppor- tunities and challenges.IEEE Computer Graphics and Applications, 44(2):55–64, 2024. doi: 10.1109/MCG.2024.3362168 2

    work page doi:10.1109/mcg.2024.3362168 2024

  5. [5]

    Bateman, R

    S. Bateman, R. L. Mandryk, C. Gutwin, A. Genest, D. McDine, and C. Brooks. Useful junk? The effects of visual embellishment on com- prehension and memorability of charts. InProc. of CHI, pp. 2573–

  6. [6]

    doi: 10.1145/1753326.1753716 1, 2

    ACM, 2010. doi: 10.1145/1753326.1753716 1, 2

    work page doi:10.1145/1753326.1753716 2010

  7. [7]

    M. A. Borkin, A. A. V o, Z. Bylinskii, P. Isola, S. Sunkavalli, A. Oliva, and H. Pfister. What makes a visualization memorable?IEEE Trans- actions on Visualization and Computer Graphics, 19(12):2306–2315, Dec 2013. doi: 10.1109/TVCG.2013.234 1, 2

    work page doi:10.1109/tvcg.2013.234 2013

  8. [8]

    Borland and R

    D. Borland and R. M. Taylor II. Rainbow color map (still) considered harmful.IEEE Computer Graphics and Applications, 27(2):14–17,

  9. [9]

    doi: 10.1109/MCG.2007.323435 2

    work page doi:10.1109/mcg.2007.323435 2007

  10. [10]

    A. Carusi. Scientific visualisations and aesthetic grounds for trust. Ethics and Information Technology, 10(4):243–254, 2008. doi: 10. 1007/s10676-008-9159-5 1

    2008

  11. [11]

    Cawthon and A

    N. Cawthon and A. V . Moere. The effect of aesthetic on the usability of data visualization. In2007 11th International Conference Informa- tion Visualization (IV ’07), pp. 637–648, 2007. doi: 10.1109/IV.2007 .147 1, 2

    work page doi:10.1109/iv.2007 2007

  12. [12]

    W. S. Cleveland and R. McGill. Graphical perception: Theory, ex- perimentation, and application to the development of graphical meth- ods.Journal of the American Statistical Association, 79(387):531– 554, 1984. doi: 10.1080/01621459.1984.10478080 2

    work page doi:10.1080/01621459.1984.10478080 1984

  13. [13]

    Correll and M

    M. Correll and M. Gleicher. Error bars considered harmful: Exploring alternate encodings for mean and error.IEEE Transactions on Visu- alization and Computer Graphics, 20(12):2142–2151, 2014. doi: 10. 1109/TVCG.2014.2346298 2

    work page arXiv 2014

  14. [14]

    H. Farid. Digital doctoring: Can we trust photographs? In B. Harring- ton, ed.,Deception: From Ancient Empires to Internet Dating, chap. 5. Stanford University Press, 2009. doi: 10.1515/9781503626607-007 4

    work page doi:10.1515/9781503626607-007 2009

  15. [15]

    Freiling, S

    I. Freiling, S. K. Yeo, and H. Xue. When AI and humans produce par- tial truths: Examining acceptability of perceived error and perceived associated harms.Health Communication, 0(0):1–12, 2026. PMID: 41482663. doi: 10.1080/10410236.2025.2608202 2

    work page doi:10.1080/10410236.2025.2608202 2026

  16. [16]

    Gulati, M

    A. Gulati, M. Mart ´ınez-Garcia, D. Fern´andez, M. A. Lozano, B. Lepri, and N. Oliver. What is beautiful is still good: The attractiveness halo effect in the era of beauty filters.Royal Society Open Science, 11(11),

  17. [17]

    doi: 10.1098/rsos.240882 4

    work page doi:10.1098/rsos.240882

  18. [18]

    Huang, W

    L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qin, and T. Liu. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions.ACM Trans. Inf. Syst., 43(2), Jan. 2025. doi: 10.1145/ 3703155 1

    2025

  19. [19]

    E. A. Jensen, K. Borkiewicz, J. P. Naiman, S. Levy, and J. Carpen- ter. Picture perfect science communication: How public audiences respond to informational labels in cinematic-style 3D data visualiza- tion.PLOS ONE, 19(10):e0307733, 2024. doi: 10.1371/journal.pone .0307733 4

    work page doi:10.1371/journal.pone 2024

  20. [20]

    Joshi and P

    A. Joshi and P. Rheingans. Illustration-inspired techniques for visu- alizing time-varying data. InIEEE Visualization, pp. 679–686, 2005. doi: 10.1109/VISUAL.2005.1532857 4

    work page doi:10.1109/visual.2005.1532857 2005

  21. [21]

    P. M. Kumarage and M. Saarela. Explainable generative AI: A two- stage review of existing techniques and future research directions.AI, 7(1), 2026. doi: 10.3390/ai7010031 2

    work page doi:10.3390/ai7010031 2026

  22. [22]

    M. C. Laupichler, N. Knoth, J. Schleiss, and T. Raupach. Algorithm aversion revisited: The role of AI literacy and attitudes towards AI in shaping perceptions of AI-generated texts.British Journal of Educa- tional Technology, 2025. doi: 10.1111/bjet.70035 2

    work page doi:10.1111/bjet.70035 2025

  23. [23]

    Lin and M

    C. Lin and M. A. Thornton. Visualization aesthetics influence trust in science, news, and social media, 2021. doi: 10.31234/osf.io/dnr9s 1

    work page doi:10.31234/osf.io/dnr9s 2021

  24. [24]

    S. Liu, H. Miao, and P.-T. Bremer. ParaView-MCP: An autonomous visualization agent with direct tool use. InIEEE Visualization and Visual Analytics (VIS), pp. 61–65, 2025. doi: 10.1109/VIS60296.2025 .00018 1

    work page doi:10.1109/vis60296.2025 2025

  25. [25]

    M. A. Lones. Pitfalls and risks of generative AI in machine learning. Patterns, p. 101534, 2026. doi: 10.1016/j.patter.2026.101534 2

    work page doi:10.1016/j.patter.2026.101534 2026

  26. [26]

    Mucundorfeanu, D

    M. Mucundorfeanu, D. C. Balaban, and M. Mauer. Exploring the ef- fectiveness of digital manipulation disclosures for instagram posts on source credibility and authenticity of social media influencers.Inter- national Journal of Advertising, 44(1):131–163, 2025. doi: 10.1080/ 02650487.2024.2381973 4

    work page arXiv 2025

  27. [27]

    Nasim Saravi, J

    A. Nasim Saravi, J. Horacsek, U. Alim, and J. D. Silva. Transfer- ring transfer functions (TTF): A guided approach to transfer func- tion optimization in volume visualization.Computers & Graphics, 124:104067, 2024. doi: 10.1016/j.cag.2024.104067 1

    work page doi:10.1016/j.cag.2024.104067 2024

  28. [28]

    A. V . Pandey, A. Manivannan, O. Nov, M. Satterthwaite, and E. Bertini. The persuasive power of data visualization.IEEE Trans- actions on Visualization and Computer Graphics, 20(12):2211–2220,

  29. [29]

    doi: 10.1109/TVCG.2014.2346419 2

    work page doi:10.1109/tvcg.2014.2346419 2014

  30. [30]

    A. V . Pandey, K. Rall, M. L. Satterthwaite, O. Nov, and E. Bertini. How deceptive are deceptive visualizations? An empirical analysis of common distortion techniques. InProc. of CHI, p. 1469–1478. ACM, New York, NY , USA, 2015. doi: 10.1145/2702123.2702608 2

    work page doi:10.1145/2702123.2702608 2015

  31. [31]

    We are visual thinkers, not verbal thinkers!

    H. Park, J. Eirich, A. Luckow, and M. Sedlmair. “We are visual thinkers, not verbal thinkers!”: A thematic analysis of how profes- sional designers use generative AI image generation tools. InProceed- ings of the 13th Nordic Conference on Human-Computer Interaction, NordiCHI ’24. Association for Computing Machinery, New York, NY , USA, 2024. doi: 10.1145/...

    work page doi:10.1145/3679318.3685370 2024

  32. [32]

    Mucundorfeanu, D

    A. Pfeuffer, H. R. Hatfield, N. Evans, and J. Kim. Illegally beau- tiful? The role of trust and persuasion knowledge in online image manipulation disclosure effects.International Journal of Advertising, 44(4):696–717, 2025. doi: 10.1080/02650487.2024.2403311 4

    work page doi:10.1080/02650487.2024.2403311 2025

  33. [33]

    Rautek, S

    P. Rautek, S. Bruckner, E. Gr ¨oller, and I. Viola. Illustrative visual- ization: New technology or useless tautology?SIGGRAPH Comput. Graph., 42(3), Aug. 2008. doi: 10.1145/1408626.1408633 4

    work page doi:10.1145/1408626.1408633 2008

  34. [34]

    J. Rho, M. A. Rau, and X. Zhu. Comparing the deceptive impact of misleading data visualizations: Implications for adaptive data literacy support in computer-based learning environments.Computers & Ed- ucation, 242:105508, 2026. doi: 10.1016/j.compedu.2025.105508 2

    work page doi:10.1016/j.compedu.2025.105508 2026

  35. [35]

    Filipov, A

    V . Schetinger, S. Di Bartolomeo, M. El-Assady, A. McNutt, M. Miller, J. P. A. Passos, and J. L. Adams. Doom or deliciousness: Challenges and opportunities for visualization in the age of generative models. Computer Graphics Forum, 42(3):423–435, 2023. doi: 10.1111/cgf. 14841 2

    work page doi:10.1111/cgf 2023

  36. [36]

    G. G. Starr. Aesthetic experience models human learning.Frontiers in Human Neuroscience, V olume 17 - 2023, 2023. doi: 10.3389/fnhum. 2023.1146083 1

    work page doi:10.3389/fnhum 2023

  37. [37]

    Tang and C

    K. Tang and C. Wang. StyleRF-V olVis: Style transfer of neural radi- ance fields for expressive volume visualization.IEEE Transactions on Visualization & Computer Graphics, 31(1):613–623, Jan. 2025. doi: 10.1109/TVCG.2024.3456342 1, 2

    work page doi:10.1109/tvcg.2024.3456342 2025

  38. [38]

    T. L. Weissgerber, N. M. Milic, S. J. Winham, and V . D. Garovic. Beyond bar and line graphs: Time for a new data presentation paradigm.PLOS Biology, 13(4):1–10, 04 2015. doi: 10.1371/journal. pbio.1002128 2

    work page doi:10.1371/journal 2015

  39. [39]

    P. Wlekły. Aesthetics and usability of statistics data visualisation through charts: An eyetracking study as a tool for chart analysis. European Research Studies Journal, 27(B):848–868, 2024. doi: 10. 35808/ersj/3679 1

    2024

  40. [40]

    Y . Ye, J. Hao, Y . Hou, Z. Wang, S. Xiao, Y . Luo, and W. Zeng. Gener- ative AI for visualization: State of the art and future directions.Visual Informatics, 8(2):43–66, 2024. doi: 10.1016/j.visinf.2024.04.003 2

    work page doi:10.1016/j.visinf.2024.04.003 2024