Pith. sign in

REVIEW 3 major objections 5 minor 103 references

Kaleidoscope Gallery: Exploring Ethics and Generative AI Through Art

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read This paper claims that DALL-E 3, asked to picture five ethical theories, produces a visual grammar of morality that experts read as Western, hierarchical, and biased by gender and geography.

desk verdict A genuinely useful art-based method for probing how T2I models visualize ethics, undercut by a headline generalization the closed evaluation loop cannot support. read the letter →

arxiv 2505.14758 v1 pith:QUZTSUOA submitted 2025-05-20 cs.CY cs.AI

classification cs.CYcs.AI
keywords CriticalDesignVisualEthicsEthicalTheoriesGenAIText-to-ImageGeneration(T2I)ResearchThroughbiasDALL-E3
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

Text-to-image models such as DALL-E 3 are not neutral illustrators of abstract ideas: this paper sets out to show that when the model is asked to picture five families of ethical theory, its images carry a consistent Western, hierarchical, and gender-and-geography-biased visual grammar. To demonstrate this, the paper interviews ten ethics experts, turns their definitions of ethical theories into prompts, generates twenty images with DALL-E 3, and has the same experts interpret the resulting Kaleidoscope Gallery. The interpretations distill into eight themes—moral compass, representations of morality, bias of representation, difference and diversity, social construct, decision-making procedures, signs and symbolism, and image composition—which give a vocabulary for naming what the model encodes. If the paper is right, this matters because AI-generated images are increasingly used to stand in for complex concepts in classrooms, media, and public communication, where the bias pattern would travel with them.

What carries the argument

The mechanism that carries the argument is a closed human-in-the-loop pipeline. Ten ethics experts first supply definitions and practices of ethical theories; inductive coding reduces those accounts to five families (Virtue, Duty-based, Consequentialism, Contractualism, Pluralism) and to paired 'definition' and 'practice' prompts; DALL-E 3 renders each prompt twice, producing the twenty-image Kaleidoscope Gallery; the same experts then interpret the images grouped by family; and iterative thematic analysis turns their commentary into the eight-theme codebook. The physical kaleidoscope shown to participants at the start of the formative interviews is the design probe that primes the 'ever-changing' framing linking ethics and generative models. The eight-theme codebook is the key diagnostic object: it converts subjective expert commentary into a reusable vocabulary for detecting the model's biases.

What would settle it

Have a fresh panel of evaluators, blind to the prompts and to the family grouping, tag the same twenty Kaleidoscope Gallery images for the eight themes; the claim predicts the Western, hierarchical, and gender-and-geography pattern will reproduce, while the closed-loop alternative predicts it will weaken or shift.

Watch

Extended reading notes

Core claim

In the paper's telling, the discovery is that DALL-E 3 can translate abstract ethical theories into images that expert viewers recognize as coherent visualizations of those theories, but the visual language of those images is systematically lopsided. The ten expert participants described images in terms of an internal moral compass, cosmic and religious imagery, scales and legal symbols, structured and hierarchical governance, and learned symbolic associations; they also flagged male-coded moral exemplars in virtue ethics, the Americas foregrounded in duty-based images, sexualized depictions of women in contractualism, and US-centered or stereotyped portrayals of global cultures. From these readings the paper builds three categories (morality, society, learned associations), eight themes, and seventeen sub-themes, and it concludes that the model is hierarchical in social construct, western in worldview, and biased in gender and geography—even while succeeding at conceptualizing complex ethical concepts.

Load-bearing premise

The claim assumes that the ten experts' evaluations measure DALL-E 3's own representations rather than re-reading the definitions those same experts supplied for the prompts—a concern the paper itself acknowledges through its Western institutional framing, family-grouped presentation, and small twenty-image sample.

Editorial extensions

If this is right

  • AI-generated illustrations of abstract ideas should be treated as culturally positioned artifacts, not neutral diagrams, especially when used in classrooms, newsrooms, or public policy materials.
  • The eight-theme codebook gives researchers a transferable vocabulary for auditing text-to-image models, so bias checks need not start from scratch with every new model.
  • Separating 'definition' from 'practice' prompts proved generative; applying the same two-track prompting to other abstract concepts such as fairness, privacy, or justice could expose where a model's conceptual grammar shifts between theory and application.
  • Because the paper deliberately chose DALL-E 3 for its prompt-following ability, a direct corollary is that other text-to-image models should be expected to produce different and possibly less biased visual grammars, and the paper explicitly calls for such comparisons.
  • The findings strengthen the case for participatory and community-centered design of generative systems, since the bias pattern is visible to domain experts but not to a purely technical evaluation.

Reading between the lines

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

  • A testable extension: turn the eight-theme codebook into a quantitative probe—generate a fixed prompt set, count occurrences of male-coded exemplars, Western geography, scales, and blue-orange palettes, and compare across models or prompt variants.
  • An implication the paper leaves implicit: running the same pipeline with non-Western ethical traditions (the paper notes its own Western skew) would likely require new themes, not a simple reversal of the existing ones.
  • A caution supported by the paper's own limitations: because experts saw images grouped by family and knew the study's purpose, a blind replication would separate model-level bias from expectation-driven reading.
  • If the hierarchy and geography findings generalize, they may point to a broader compositional bias in text-to-image models—social pyramids and US-centered maps could appear just as readily for abstract concepts such as order, progress, or community.
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

3 major / 5 minor

Summary. The paper presents a qualitative, art-based study ('Kaleidoscope Gallery') in which ten ethics experts were first interviewed about ethical theories, their responses were distilled into five ethical families (Virtue, Duty-based, Consequentialism, Contractualism, Pluralism) and two prompts per family (definition and practice), and DALL-E 3 was used to generate twenty images. The same ten experts then evaluated the images, which were grouped by ethical family, during follow-up interviews. Thematic analysis of those evaluations yielded three categories, eight themes, and seventeen sub-themes, and the paper argues that the images reveal how T2I models visually and conceptually represent ethical theories, while also exhibiting western, hierarchical, gendered, and geographic biases. The paper positions the work within Visual Ethics, critical AI art, and research-through-design, and it closes with cautions, limitations, and future work.

Significance. If the central claims are accepted, the paper offers a novel methodological contribution: using generative imagery and expert interpretation as a critical lens on T2I models' representation of complex philosophical concepts. The study is transparent about its qualitative coding process (consensus then split coding, three-step iterative analysis), provides participant quotes and image references for each theme, and acknowledges several limitations in §5.4. The eight themes in Table 2 could serve as a useful vocabulary for discussing how AI-generated imagery encodes moral, social, and symbolic content. However, the significance is currently constrained by the study's closed evaluation loop and by the gap between the evidence gathered (ten experts, twenty images, one model) and the generalizing language used in the abstract and §5.3. The work's value as a design exploration and critical provocation is clear; its value as an empirical measurement of T2I model properties is not yet established.

major comments (3)
  1. [§3.2, §3.3, §3.5, §5.3] The load-bearing claim in §5.3 that T2I models are 'hierarchical in social construct, western in worldview, and biased in gender and geography' is not supported by the study design, because the evaluation loop is closed: the same ten experts supplied the ethical-theory definitions from which the prompts were built (§3.2, §3.3), and the same experts then judged images they knew were grouped by ethical family (§3.5). The evaluation therefore functions partly as a member-check of the prompt source rather than an independent measurement of model properties. To support the bias attribution, the paper should either reframe the claim as describing the human-AI-human loop (experts, prompts, images, experts) or add an independent validation step, such as blind presentation of unlabelled images, a fresh evaluator panel, or baseline prompts unrelated to the expert interviews.
  2. [§3.4, §5.3, §5.4] The plural 'T2I models' in §5.3 and in the abstract overgeneralizes from a single model and a small sample. Images were generated only with DALL-E 3 (§3.4), with two prompts per family and two images per prompt, yielding twenty images total. Section 5.4 acknowledges that this 'small sample might limit potential arguments about inherent bias within the models,' but the wording of the central finding in §5.3 does not carry that caveat. The authors should either restrict the claim to DALL-E 3 or present comparative evidence from additional T2I models, as planned in §5.4.
  3. [§1, §3.5, §5.3] The distinction between RQ1 (visual representation) and RQ2 (conceptual representation) is not operationalized. Both research questions are answered with the same evidence: expert interpretations of images that were grouped by family and prompted by expert-derived definitions. The abstract and §5.3 use 'conceptualize complex concepts,' but the paper does not define what would count as conceptual representation as opposed to visual representation, especially given the paper's own citation of West et al. (§2.1.2) noting that generative capability may not entail understanding. The claim that T2I models 'conceptualize' ethical theories therefore needs either a stated operational definition or a more cautious phrasing, such as 'prompt-conditioned visual synthesis that experts read as conceptually meaningful.'
minor comments (5)
  1. [Figure 2] The summary of study procedure in Figure 2 labels the final analysis step 'Section 1,' but the methodology and analysis are described in Section 3; the label appears to be a leftover placeholder and should be corrected.
  2. [§4.6.2] The text reads 'a classic symbol respresenting justice'; 'respresenting' is a typo for 'representing.'
  3. [§5.1] The phrase 'GenAI models and ethical theories' are seemingly static yet dynamic' mixes singular and plural agreement and reads awkwardly; it should be rephrased.
  4. [§4.3.1] The abbreviation 'NSFW' is introduced without expansion; please define it at first use (e.g., 'not safe for work (NSFW)').
  5. [References] Reference [19] cites a Wikipedia page version for Fricker's concept of epistemic injustice; the authors should cite the primary source (M. Fricker, 'Epistemic Injustice: Power and the Ethics of Knowing,' Oxford University Press, 2007) instead of or in addition to the encyclopedia entry.

Circularity Check

1 steps flagged · score 6.0 of 10

Central bias claim rests on a closed expert-prompt-expert loop: same experts supplied definitions used to build prompts and then judged family-labeled images, making the 'T2I model bias' attribution partly a member-check of the prompt source.

  1. fitted input called prediction [§3.2–§3.5 with claim in §5.3]
    "First, we found that T2I models were able to conceptualize complex concepts, as highlighted in our themes, however, as determined by our experts, showed to be hierarchical in social construct, western in worldview, and biased in gender and geography (RQ1). ... The generated images were grouped according to their corresponding ethical family and presented to the experts in a 30-minute follow-up interview ... we identified the five families of ethical theories spun from both selected and participant-introduced ethical theories."

    The prompt content is fitted to the ten experts' own definitions (§3.2: 'participants were asked to detail their understanding of seven prominent ethical theories'; §3.3: five families 'spun from both selected and participant-introduced ethical theories'), and the same experts then evaluate the family-labeled images (§3.5). The §5.3 inference about 'T2I models' therefore measures whether experts recognize their own prior definitions in the model's rendering of prompts built from those definitions, not model behavior independently. Gender, geographic, and hierarchical attributions are read onto images whose provenance and grouping the evaluators knew, so those attributions are partly constructed by the study loop.

full rationale

The paper is not circular via self-citation: background citations are external, the kaleidoscope metaphor is borrowed, and no uniqueness theorem is invoked. The circularity is the empirical loop. In §3.2 the same ten experts supplied the theory definitions; §3.3 coded those into the five prompt families; §3.4 DALL-E 3 generated images; §3.5 the same experts, seeing family-labeled exhibits, evaluated them. The §5.3 conclusion about model-level bias is therefore partly a re-description of the evaluators' own inputs: the prompts are a re-encoding of expert concepts, so the evaluation functions as a member-check of the prompt source. Some independent content remains (composition, color, symbolism, and the model's actual renderings), so the paper is not entirely reducible to its inputs; the score is 6 rather than 8-10. The paper's explicit limitations in §5.4 partially mitigate the over-generalization but do not remove the circularity from the §5.3 wording.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The central claim rests on a small set of researcher choices (five-family reduction, definition/practice prompt split, four images per family, DALL-E 3 as sole model) and on domain assumptions about expert elicitation, non-blind evaluation, and dual-researcher coding. The Kaleidoscope Gallery and the five-family taxonomy are constructed artifacts whose evidential role depends entirely on the study design; no physics-style invented entities are introduced. Each ledger entry is load-bearing for the statement that T2I models represent ethics with the observed biases.

free parameters (5)
  • Five-family taxonomy = Virtue, Duty-based, Consequentialism, Contractualism, Pluralism
    The inductive coding reduced seven seed theories plus participant-introduced theories to five families; this researcher choice determines all prompts and gallery content.
  • Definition/practice prompt split = 2 prompts per family
    Each family was rendered as one 'definition' and one 'practice' prompt written by the researchers from interview codes; different phrasing would change the images and likely the themes.
  • Images per family = 4 (2 per prompt)
    Twenty images total, chosen 'not to overcrowd' experts; a deliberate design choice that limits coverage, acknowledged in §5.4.
  • Model choice = DALL-E 3
    Selected for prompt-following capabilities; results are model-specific and not compared with other T2I models, acknowledged in §5.4.
  • Expert panel = n=10, snowball-sampled, Western institution
    The pool determines the 'foundation of ethical theories'; the paper acknowledges the Western-philosophy skew, e.g. lacking Ubuntu, in §5.4.
assumptions (4)
  • domain assumption Ten expert interviews provide a valid foundation for the space of ethical theories
    §3.2. The seed list of seven theories and the snowball-sampled panel bound the theory space; no external benchmark establishes completeness or representativeness, and the paper acknowledges the Western skew in §5.4.
  • domain assumption DALL-E 3 outputs can be read as 'the model's representation' of the prompt
    §3.4. The paper answers RQ1/RQ2 from one proprietary model's outputs, with no test-time randomization analysis, no comparison prompts, and no other model, so the attribution to 'the model' is assumed.
  • domain assumption Expert evaluation is not unduly influenced by known family labels
    §3.5. Images were grouped by family with questions tailored to that family; the paper notes the possible influence in §5.4 but still treats the judgments as measurements of the model's representation.
  • domain assumption Two-researcher coding without an inter-rater reliability statistic is reliable
    §3.3 and §3.5 describe consensus and split coding but report no agreement metric and no full codebook, so coding reliability is asserted rather than demonstrated.
invented entities (2)
  • Kaleidoscope Gallery
    purpose: Curated set of twenty DALL-E 3 images used as the stimulus for expert evaluation and as the study's central artifact
    A study artifact, not an independently verifiable entity; its evidential role depends entirely on the closed expert loop.
  • Five ethical families taxonomy
    purpose: Organizing construct for prompt generation, image grouping, and theme analysis
    A researcher-constructed coding output built from standard ethics categories; the specific reduction and its boundaries are not externally validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Kaleidoscope Gallery: Exploring Ethics and Generative AI Through Art." pith.science (2026). https://pith.science/paper/QUZTSUOA

@misc{pith2026250514758,
  author       = {Pith},
  title        = {Pith review of: Kaleidoscope Gallery: Exploring Ethics and Generative AI Through Art},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QUZTSUOA}},
  note         = {Machine review of arXiv:2505.14758}
}
read the original abstract

Ethical theories and Generative AI (GenAI) models are dynamic concepts subject to continuous evolution. This paper investigates the visualization of ethics through a subset of GenAI models. We expand on the emerging field of Visual Ethics, using art as a form of critical inquiry and the metaphor of a kaleidoscope to invoke moral imagination. Through formative interviews with 10 ethics experts, we first establish a foundation of ethical theories. Our analysis reveals five families of ethical theories, which we then transform into images using the text-to-image (T2I) GenAI model. The resulting imagery, curated as Kaleidoscope Gallery and evaluated by the same experts, revealed eight themes that highlight how morality, society, and learned associations are central to ethical theories. We discuss implications for critically examining T2I models and present cautions and considerations. This work contributes to examining ethical theories as foundational knowledge that interrogates GenAI models as socio-technical systems.

Figures

Figures reproduced from arXiv: 2505.14758 by the authors.

Figure 2
Figure 2. Summary of Study Procedure Previous work by Tendulkar et al. [81] addresses this representa￾tion for generating conceptual text via a semantic reinforcement approach known as TReAT—Thematic Reinforcement for Artistic Typography. Given an input word and a theme, TReAT replaces the input word with a clipart of each letter upon the context or theme of the word. In a similar approach, recent work by Iluz et al. [33] fur… view at source ↗
Figure 3
Figure 3. Design probe used to onboard participants to the [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Virtue Ethics (a) Definition 1 (b) Definition 2 (c) Practice 1 (d) Practice 2 [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (2 more)
Figure 7
Figure 7. Figure 7: Contractual Ethics [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Pluralistic Ethics [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

103 extracted references · 41 canonical work pages

  1. [1]

    Evgeni Aizenberg and Jeroen van den Hoven. 2020. Designing for human rights in AI. https://doi.org/10.1177/2053951720949566

  2. [2]

    Baraniuk

    Sina Alemohammad, Josue Casco-Rodriguez, Lorenzo Luzi, Ahmed Imtiaz Hu- mayun, Hossein Babaei, Daniel LeJeune, Ali Siahkoohi, and Richard G. Baraniuk

  3. [3]

    Larry Alexander and Michael Moore. 2024. Deontological Ethics. In The Stanford Encyclopedia of Philosophy (winter 2024 ed.), Edward N. Zalta and Uri Nodelman (Eds.). Metaphysics Research Lab, Stanford University. https://plato.stanford. edu/archives/win2024/entries/ethics-deontological/

  4. [4]

    Prithviraj Ammanabrolu, Liwei Jiang, Maarten Sap, Hannaneh Hajishirzi, and Yejin Choi. 2022. Aligning to Social Norms and Values in Interactive Narratives. https://doi.org/10.48550/arXiv.2205.01975 arXiv:2205.01975

  5. [5]

    Elizabeth Ashford and Tim Mulgan. 2018. Contractualism. In The Stanford Encyclopedia of Philosophy (summer 2018 ed.), Edward N. Zalta (Ed.). Metaphysics Research Lab, Stanford University. https://plato.stanford.edu/archives/sum2018/ entries/contractualism/

  6. [6]

    Bennett, Cecilia Chi-Ham, Geoffrey Barrows, Steven Sexton, and David Zilberman

    Alan B. Bennett, Cecilia Chi-Ham, Geoffrey Barrows, Steven Sexton, and David Zilberman. 2013. Agricultural Biotechnology: Economics, Environment, Ethics, and the Future. Annual Review of Environment and Resources 38, 1 (2013), 249–279. https://doi.org/10.1146/annurev-environ-050912-124612 _eprint: https://doi.org/10.1146/annurev-environ-050912-124612

  7. [7]

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, Wesam Manassra, Prafulla Dhari- wal, Casey Chu, Yunxin Jiao, and Aditya Ramesh. 2023. Improving Image Gener- ation with Better Captions. (2023)

  8. [8]

    Camilla Bignotti and Carolina Camassa. 2024. Legal Minds, Algorithmic De- cisions: How LLMs Apply Constitutional Principles in Complex Scenarios. https://doi.org/10.48550/arXiv.2407.19760 arXiv:2407.19760

Show all 103 references
  1. [9]

    Charlotte Bird, Eddie Ungless, and Atoosa Kasirzadeh. 2023. Typology of Risks of Generative Text-to-Image Models. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society . ACM, Montr\’{e}al QC Canada, 396–410. https://doi.org/10.1145/3600211.3604722

  2. [10]

    Maarten Buyl, Alexander Rogiers, Sander Noels, Iris Dominguez-Catena, Edith Heiter, Raphael Romero, Iman Johary, Alexandru-Cristian Mara, Jefrey Lijffijt, and Tijl De Bie. 2024. Large Language Models Reflect the Ideology of their Creators. http://arxiv.org/abs/2410.18417 arXiv...

  3. [11]

    Michaud, Jacob Pfau, Dmitrii Krasheninnikov, Xin Chen, Lauro Langosco, Peter Hase, Erdem Bıyık, Anca Dragan, David Krueger, Dorsa Sadigh, and Dylan Hadfield-Menell

    Stephen Casper, Xander Davies, Claudia Shi, Thomas Krendl Gilbert, Jérémy Scheurer, Javier Rando, Rachel Freedman, Tomasz Korbak, David Lindner, Pe- dro Freire, Tony Wang, Samuel Marks, Charbel-Raphaël Segerie, Micah Carroll, Andi Peng, Phillip Christoffersen, Mehul Damani, St...

  4. [12]

    Eva Cetinic and James She. 2021. Understanding and Creating Art with AI: Review and Outlook. https://doi.org/10.48550/arXiv.2102.09109 arXiv:2102.09109

  5. [13]

    Kientz, Jason Yip, and Caroline Pitt

    Aayushi Dangol, Michele Newman, Robert Wolfe, Jin Ha Lee, Julie A. Kientz, Jason Yip, and Caroline Pitt. 2024. Mediating Culture: Cultivating Socio-cultural Understanding of AI in Children through Participatory Design. In Designing Interactive Systems Conference . ACM, IT Univ...

  6. [14]

    Antonio Daniele and Yi-Zhe Song. 2019. AI + Art = Human. In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society . ACM, Honolulu HI USA, 155–161. https://doi.org/10.1145/3306618.3314233

  7. [15]

    Ernest. 2018. Pictograms and Ideograms - Origin and History. https://citrinitas. com/a-history-of-visual-communication-pictograms-and-ideograms/

  8. [16]

    Hugging Face. 2024. What is Text-to-Image? https://huggingface.co/tasks/text- to-image

  9. [17]

    Maria Flis and Karol Piotrowski. 2024. The Conceptual Metaphor as an Ethical Kaleidoscope in Field Research. Qualitative Sociology Review 20, 1 (Jan. 2024), 30–41. https://doi.org/10.18778/1733-8077.20.1.03

  10. [18]

    Interaction Design Foundation. 2016. What is Color Theory? https://www. interaction-design.org/literature/topics/color-theory

  11. [19]

    Miranda Fricker. 2024. Epistemic injustice. https://en.wikipedia.org/w/index. php?title=Epistemic_injustice&oldid=1251981459 Page Version ID: 1251981459

  12. [20]

    Batya Friedman and David G. Hendry. 2019. Value Sensitive Design: Shaping Technology with Moral Imagination. The MIT Press. https://mitpress.mit.edu/ 9780262039536/value-sensitive-design/

  13. [21]

    Iason Gabriel. 2020. Artificial Intelligence, Values and Alignment. Minds and Machines 30, 3 (Sept. 2020), 411–437. https://doi.org/10.1007/s11023-020-09539-2 arXiv:2001.09768 [cs]

  14. [22]

    Ainur Gainetdinov. 2023. Diffusion Models vs GANs vs VAEs: Comparison of Deep Generative Models. https://pub.towardsai.net/diffusion-models-vs-gans- vs-vaes-comparison-of-deep-generative-models-67ab93e0d9ae

  15. [23]

    Alexandre Gefen, Léa Saint-Raymond, and Tommaso Venturini. 2021. AI for Digital Humanities and Computational Social Sciences.Lecture Notes in Computer Science (2021). https://doi.org/10.1007/978-3-030-69128-8_12

  16. [24]

    Sourojit Ghosh. 2024. Interpretations, Representations, and Stereotypes of Caste within Text-to-Image Generators. https://arxiv.org/html/2408.01590v1

  17. [25]

    Trystan S. Goetze. 2024. Ai Art is Theft: Labour, Extraction, and Exploitation, or, on the Dangers of Stochastic Pollocks. Proceedings of the 2024 Acm Conference on Fairness, Accountability, and Transparency (2024), 186–196

  18. [26]

    Wilson Wen Bin Goh and Chun Chau Sze. 2019. AI Paradigms for Teach- ing Biotechnology. Trends in Biotechnology (2019). https://doi.org/10.1016/ J.TIBTECH.2018.09.009

  19. [27]

    Leo A. Goodman. 1961. Snowball Sampling. The Annals of Mathematical Statistics 32, 1 (1961), 148 – 170. https://doi.org/10.1214/aoms/1177705148

  20. [28]

    Jan-Christoph Heilinger. 2022. The Ethics of AI Ethics. A Constructive Critique. Philosophy & Technology 35, 3 (July 2022), 61. https://doi.org/10.1007/s13347- 022-00557-9

  21. [30]

    Dan Hendrycks, Collin Burns, Steven Basart, Andrew Critch, Jerry Li, Dawn Song, and Jacob Steinhardt. 2023. Aligning AI With Shared Human Values. http://arxiv.org/abs/2008.02275 arXiv:2008.02275 [cs]

  22. [31]

    John Hospers. 1954. The Concept of Artistic Expression. Proceedings of the Aristotelian Society 55 (1954), 313–344. https://www.jstor.org/stable/4544551 Publisher: [Aristotelian Society, Wiley]

  23. [32]

    Rosalind Hursthouse and Glen Pettigrove. 2023. Virtue Ethics. In The Stanford Encyclopedia of Philosophy (fall 2023 ed.), Edward N. Zalta and Uri Nodelman (Eds.). Metaphysics Research Lab, Stanford University. https://plato.stanford. edu/archives/fall2023/entries/ethics-virtue/

  24. [33]

    Shir Iluz, Yael Vinker, Amir Hertz, Daniel Berio, Daniel Cohen-Or, and Ariel Shamir. 2023. Word-As-Image for Semantic Typography. http://arxiv.org/abs/ 2303.01818 arXiv:2303.01818 [cs]

  25. [34]

    Lam, Minh Chau Mai, Jeff Hancock, and Michael S

    Chenyan Jia, Michelle S. Lam, Minh Chau Mai, Jeff Hancock, and Michael S. Bernstein. 2023. Embedding Democratic Values into Social Media AIs via Societal Objective Functions. arXiv:2307.13912 [cs.HC]

  26. [35]

    Liwei Jiang, Jena D. Hwang, Chandra Bhagavatula, Ronan Le Bras, Jenny Liang, Jesse Dodge, Keisuke Sakaguchi, Maxwell Forbes, Jon Borchardt, Saadia Gabriel, Yulia Tsvetkov, Oren Etzioni, Maarten Sap, Regina Rini, and Yejin Choi. 2022. Can Machines Learn Morality? The Delphi Exp...

  27. [36]

    Mark Johnson. 1994. Moral Imagination: Implications of Cognitive Science for Ethics. University of Chicago Press, Chicago, IL. https://press.uchicago.edu/ucp/ books/book/chicago/M/bo3684141.html

  28. [37]

    Wm Matthew Kennedy and Daniel Vargas Campos. 2024. Vernacularizing Tax- onomies of Harm is Essential for Operationalizing Holistic AI Safety. https: //doi.org/10.48550/arXiv.2410.16562 arXiv:2410.16562

  29. [38]

    Zachary Kenton, Tom Everitt, Laura Weidinger, Iason Gabriel, Vladimir Mikulik, and Geoffrey Irving. 2021. Alignment of Language Agents. http://arxiv.org/abs/ 2103.14659 arXiv:2103.14659 [cs]

  30. [39]

    Kurt Koffka. 2014. Perception: An Introduction To The Gestalt Theory: A Classic Article in the History of Psychology . www.all-about-psychology.com

  31. [40]

    Olya Kudina and Ibo van de Poel. 2024. A sociotechnical system perspective on AI. Minds and Machines 34, 3 (June 2024), 21. https://doi.org/10.1007/s11023- 024-09680-2

  32. [41]

    George Lakoff and Mark Johnson. 2003. Metaphors We Live By . University of Chicago Press, Chicago, IL. https://press.uchicago.edu/ucp/books/book/chicago/ M/bo3637992.html

  33. [42]

    Hannah Landecker. 1999. Between Beneficence and Chattel: The Human Bi- ological in Law and Science. Science in Context 12, 1 (April 1999), 203–225. https://doi.org/10.1017/S0269889700003367 Publisher: Cambridge University Press

  34. [43]

    Dabae Lee, Sheunghyun Yeo, Dabae Lee, and Sheunghyun Yeo. 2022. Developing an AI-based chatbot for practicing responsive teaching in mathematics. (2022). https://doi.org/10.1016/J.COMPEDU.2022.104646

  35. [44]

    Harrison Lee, Samrat Phatale, Hassan Mansoor, Kellie Lu, Thomas Mesnard, Colton Bishop, Victor Carbune, and Abhinav Rastogi. 2023. RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback. http://arxiv. org/abs/2309.00267 arXiv:2309.00267 [cs]

  36. [45]

    Long Lian, Boyi Li, Adam Yala, and Trevor Darrell. 2023. LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models. http://arxiv.org/abs/2305.13655 arXiv:2305.13655 [cs]

  37. [46]

    Juniper Lovato, Julia Zimmerman, Isabelle Smith, Peter Dodds, and Jennifer Karson. 2024. Foregrounding Artist Opinions: A Survey Study on Transparency, Ownership, and Fairness in AI Generative Art. https://doi.org/10.48550/arXiv. 2401.15497 arXiv:2401.15497

  38. [47]

    Elinor Mason. 2023. Value Pluralism. In The Stanford Encyclopedia of Philosophy (summer 2023 ed.), Edward N. Zalta and Uri Nodelman (Eds.). Metaphysics Research Lab, Stanford University. https://plato.stanford.edu/archives/sum2023/ entries/value-pluralism/

  39. [48]

    Marian Mazzone and Ahmed Elgammal. 2019. Art, Creativity, and the Potential of Artificial Intelligence. Arts 8 (Feb. 2019). https://doi.org/10.3390/arts8010026

  40. [49]

    Nan Leslie McDonald and Douglas Fisher. 2006. Teaching Literacy Through the Arts. Guilford Press. Google-Books-ID: lrQukcPC_zYC

  41. [50]

    Meta. 2024. Movie Gen: A Cast of Media Foundation Models. AI at Meta. https://ai.meta.com/static-resource/movie-gen-research-paper

  42. [51]

    W. J. T. Mitchell. 1995. Picture Theory: Essays on Verbal and Visual Representation . University of Chicago Press, Chicago, IL. https://press.uchicago.edu/ucp/books/ book/chicago/P/bo3683962.html

  43. [52]

    Luis Morales-Navarro, Yasmin B Kafai, Francisco Castro, William Payne, Kayla DesPortes, Daniella DiPaola, Randi Williams, Safinah Ali, Cynthia Breazeal, Clifford Lee, Elisabeth Soep, YR Media, Duri Long, Brian Magerko, Jaemarie Solyst, Amy Ogan, Cansu Tatar, Shiyan Jiang, Jie ...

  44. [53]

    Seana Moran. 2014. Introduction: The Crossroads of Creativity and Ethics. InThe Ethics of Creativity, Seana Moran, David Cropley, and James C. Kaufman (Eds.). Palgrave Macmillan UK, London, 1–22. https://doi.org/10.1057/9781137333544_1

  45. [54]

    Albert Musschenga. 2005. Empirical Ethics, Context-Sensitivity, and Contex- tualism. The Journal of Medicine and Philosophy 30, 5 (Oct. 2005), 467–490. https://doi.org/10.1080/03605310500253030

  46. [55]

    Davy Tsz Kit Ng, Jac Ka Lok Leung, Samuel Kai Wah Chu, and Maggie Shen Qiao. 2021. Conceptualizing AI literacy: An exploratory review. Computers and Education: Artificial Intelligence 2 (2021), 100041. https://doi.org/10.1016/j.caeai. 2021.100041

  47. [56]

    Dietmar Offenhuber. 2023. Autographic Design: The Matter of Data in a Self- Inscribing World. MIT Press. Google-Books-ID: bpO0EAAAQBAJ

  48. [57]

    OpenAI. 2024. How your data is used to improve model perfor- mance. https://help.openai.com/en/articles/5722486-how-your-data-is-used-to- improve-model-performance

  49. [58]

    Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and...

  50. [59]

    Dino Pedreschi, Luca Pappalardo, Emanuele Ferragina, Ricardo Baeza-Yates, Albert-László Barabási, Frank Dignum, Virginia Dignum, Tina Eliassi-Rad, Fosca Giannotti, János Kertész, Alistair Knott, Yannis Ioannidis, Paul Lukowicz, Andrea Passarella, Alex Sandy Pentland, John Shaw...

  51. [60]

    William Peebles and Saining Xie. 2023. Scalable Diffusion Models with Trans- formers. arXiv:2212.09748 [cs.CV] https://arxiv.org/abs/2212.09748

  52. [61]

    Bennett, and Emily Denton

    Rida Qadri, Renee Shelby, Cynthia L. Bennett, and Emily Denton. 2023. AI’s Regimes of Representation: A Community-centered Study of Text-to-Image Models in South Asia. In 2023 ACM Conference on Fairness, Accountability, and Transparency. ACM, Chicago IL USA, 506–517. https://d...

  53. [62]

    Prabhakar Raghavan. 2024. Gemini image generation got it wrong. We’ll do better. https://blog.google/products/gemini/gemini-image-generation-issue/

  54. [64]

    Andrew R

    K. Andrew R. Richards and Michael A. Hemphill. [n. d.]. A Practical Guide to Collaborative Qualitative Data Analysis. 37, 2 ([n. d.]), 225–231. https://doi.org/ 10.1123/jtpe.2017-0084 Publisher: Human Kinetics Section: Journal of Teaching in Physical Education

  55. [65]

    William A Gaviria Rojas, Sudnya Diamos, Keertan Ranjan Kini, David Kanter, Vijay Janapa Reddi, and Cody Coleman. 2022. The Dollar Street Dataset: Im- ages Representing the Geographic and Socioeconomic Diversity of the World. Thirty-sixth Conference on Neural Information Proces...

  56. [66]

    Marinela Rusu. 2017. 2. Empathy and Communication through Art. Review of Artistic Education 14 (March 2017). https://doi.org/10.1515/rae-2017-0018

  57. [67]

    Johnny Saldana. 2021. The Coding Manual for Qualitative Researchers. (2021), 1–440. https://www.torrossa.com/en/resources/an/5018667 Publisher: SAGE Publications Ltd

  58. [68]

    Elizabeth S Scott. 2009. Surrogacy and the Politics of Commodification. LA W AND CONTEMPORARY PROBLEMS 72 (2009)

  59. [69]

    Selbst, Danah Boyd, Sorelle A

    Andrew D. Selbst, Danah Boyd, Sorelle A. Friedler, Suresh Venkatasubramanian, and Janet Vertesi. 2019. Fairness and Abstraction in Sociotechnical Systems. In Proceedings of the Conference on Fairness, Accountability, and Transparency . ACM, Atlanta GA USA, 59–68. https://doi.o...

  60. [70]

    William Seymour, Max Van Kleek, Reuben Binns, and Dave Murray-Rust. 2022. Re- spect as a Lens for the Design of AI Systems. InProceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society . ACM, Oxford United Kingdom, 641–652. https://doi.org/10.1145/3514094.3534186

  61. [71]

    Andrew K. Shenton. [n. d.]. Strategies for ensuring trustworthiness in qualitative research projects. 22, 2 ([n. d.]), 63–75. https://doi.org/10.3233/EFI-2004-22201 Publisher: IOS Press

  62. [72]

    Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Nicolas Papernot, Ross Anderson, and Yarin Gal. 2024. AI models collapse when trained on recursively generated data. Nature 631, 8022 (July 2024), 755–759. https://doi.org/10.1038/s41586-024- 07566-y

  63. [73]

    Walter Sinnott-Armstrong. 2023. Consequentialism. In The Stanford Encyclopedia of Philosophy (winter 2023 ed.), Edward N. Zalta and Uri Nodelman (Eds.). Meta- physics Research Lab, Stanford University. https://plato.stanford.edu/archives/ win2023/entries/consequentialism/

  64. [74]

    Katrina Sluis and Erica Molesworth. 2023. Critical AI in the Art Museum. https: //criticalai.art/

  65. [75]

    Taylor Sorensen, Liwei Jiang, Jena Hwang, Sydney Levine, Valentina Pyatkin, Peter West, Nouha Dziri, Ximing Lu, Kavel Rao, Chandra Bhagavatula, Maarten Sap, John Tasioulas, and Yejin Choi. 2023. Value Kaleidoscope: Engaging AI with Pluralistic Human Values, Rights, and Duties....

  66. [76]

    Ramya Srinivasan. 2024. To See or Not to See: Understanding the Tensions of Algorithmic Curation for Visual Arts. In The 2024 ACM Conference on Fairness, Accountability, and Transparency. ACM, Rio de Janeiro Brazil, 444–455. https: //doi.org/10.1145/3630106.3658917

  67. [77]

    Ramya Srinivasan and Devi Parikh. 2022. Building Bridges: Generative Artworks to Explore AI Ethics. http://arxiv.org/abs/2106.13901 2022 CVPR Workshop on Ethical Considerations in Creative Applications of Computer Vision

  68. [78]

    Ramya Srinivasan and Kanji Uchino. 2021. The Role of Arts in Shaping AI Ethics. 2021 AAAI Workshop on Reframing Diversity in AI: Representation, Power, and Inclusion

  69. [79]

    Ramya Srinivasan and Kanji Uchino. U.S. Patent 20230186535A1, Jun. 2023. Image Generation based on Ethical Viewpoints

  70. [80]

    Sarah Sterman, Molly Jane Nicholas, Janaki Vivrekar, Jessica R Mindel, and Eric Paulos. 2023. Kaleidoscope: A Reflective Documentation Tool for a User Interface Design Course. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. ACM, Hamburg Germany...

  71. [81]

    Selvaraju, and Devi Parikh

    Purva Tendulkar, Kalpesh Krishna, Ramprasaath R. Selvaraju, and Devi Parikh

  72. [82]

    Thiroux and Keith W

    Jacques P. Thiroux and Keith W. Krasemann. 2014. Ethics: Theory and Practice (11th edition ed.). Pearson, Boston

  73. [83]

    Mark Timmons. 2012. Moral Theory: An Introduction . Rowman & Littlefield Publishers. Google-Books-ID: qWGp1iK9IlAC

  74. [84]

    Edidiong Enyeneokpon Ukoh, Jude Nicholas, Edidiong Enyeneokpon Ukoh, and Jude Nicholas. 2022. AI Adoption for Teaching and Learning of Physics. (2022). https://doi.org/10.20533/IJI.1742.4712.2022.0222

  75. [85]

    Varshney

    Kush R. Varshney. 2024. Decolonial AI Alignment: Openness, Viśe\d{s}a- Dharma, and Including Excluded Knowledges. http://arxiv.org/abs/2309.05030 arXiv:2309.05030

  76. [86]

    Gergana Vladova, Jennifer Haase, and Sascha Friesike. 2024. Why, with whom, and how to conduct interdisciplinary research? A review from a researcher’s perspective. Science and Public Policy 52, 2 (Nov. 2024), 165–180. https: //doi.org/10.1093/scipol/scae070 _eprint: https://a...

  77. [87]

    Johanna Walker, Gefion Thuermer, Julian Vicens, and Elena Simperl. 2023. AI Art and Misinformation: Approaches and Strategies for Media Literacy and Fact Checking. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. ACM, Montr\’{e}al QC Canada, 26–37. ht...

  78. [88]

    Wendell Wallach and Shannon Vallor. 2020. Moral Machines: From Value Align- ment to Embodied Virtue. In Ethics of Artificial Intelligence, S. Matthew Liao (Ed.). Oxford University Press, 0. https://doi.org/10.1093/oso/9780190905033.003.0014

  79. [89]

    Jiajun Wang, Morteza Ghahremani, Yitong Li, Björn Ommer, and Christian Wachinger. 2024. Stable-Pose: Leveraging Transformers for Pose-Guided Text-to- Image Generation. arXiv:2406.02485 [cs.CV] https://arxiv.org/abs/2406.02485

  80. [90]

    Xinru Wang and Ming Yin. 2023. Watch Out for Updates: Understanding the Effects of Model Explanation Updates in AI-Assisted Decision Making. In Pro- ceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI ’23) . Association for Computing Machinery, New Y...

  81. [91]

    What It Can Create, It May Not Understand

    Peter West, Ximing Lu, Nouha Dziri, Faeze Brahman, Linjie Li, Jena D. Hwang, Li- wei Jiang, Jillian Fisher, Abhilasha Ravichander, Khyathi Chandu, Benjamin New- man, Pang Wei Koh, Allyson Ettinger, and Yejin Choi. 2023. The Generative AI Paradox: "What It Can Create, It May No...

  82. [92]

    Robert Wolfe and Tanushree Mitra. 2024. The Implications of Open Generative Models in Human-Centered Data Science Work: A Case Study with Fact-Checking Organizations. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, Vol. 7. 1595–1607. https://ojs.aaai.org/...

  83. [93]

    Carissa Wong. 2024. AI-generated images and video are here: how could they shape research? Nature (March 2024). https://doi.org/10.1038/d41586-024-00659- 8 Bandiera_abtest: a Cg_type: News Explainer Publisher: Nature Publishing Group Subject_term: Scientific community, Softwar...

  84. [94]

    Stephen Tze-Inn Wu, Daniel Demetriou, and Rudwan Ali Husain. 2023. Honor Ethics: The Challenge of Globalizing Value Alignment in AI. In 2023 ACM Con- ference on Fairness, Accountability, and Transparency . ACM, Chicago IL USA, 593–602. https://doi.org/10.1145/3593013.3594026

  85. [95]

    Yuanshun Yao, Xiaojun Xu, and Yang Liu. 2023. Large Language Model Unlearn- ing. https://doi.org/10.48550/arXiv.2310.10683 arXiv:2310.10683 [cs]

  86. [96]

    Chenshuang Zhang, Chaoning Zhang, Mengchun Zhang, and In So Kweon

  87. [97]

    Junsong Zhang, Yu Wang, Weiyi Xiao, and Zhenshan Luo. 2017. Synthesizing Ornamental Typefaces. Computer Graphics Forum 36, 1 (2017), 64–75. https://doi.org/10.1111/cgf.12785 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/cgf.12785. A FORMATIVE STUDY INTERVIEW SCRIPT

  88. [100]

    http: //arxiv.org/abs/2303.07909 arXiv:2303.07909 [cs]

    Text-to-image Diffusion Models in Generative AI: A Survey. http: //arxiv.org/abs/2303.07909 arXiv:2303.07909 [cs]

  89. [102]

    Thank you for joining me in conversation today! My name is [name], and I am [occupation] in [Institution]

    Introduction Hello [Insert name]! Pleasure to meet you. Thank you for joining me in conversation today! My name is [name], and I am [occupation] in [Institution]. I look forward to our conversation today! Before we begin, I will detail the study we’re to embark on. This resear...

  90. [103]

    [Detail answer] Awesome! Thank you for that

    Interviewee I’d like to begin by allowing you to introduce yourself and the academic discipline you come from. [Detail answer] Awesome! Thank you for that. It’s a pleasure to get to know you!

  91. [104]

    Ethical Theories Let’s begin with Ethical Theories. • To your understanding, what are ethical theories as a syn- thesis, and you can dive back into what ethics are in the first place? This is what you conceptualize as theories prior to categories of what may lie underneath. • ...

  92. [105]

    There are very im- portant cases in which what we identify as ethics may have, or not have been divided as such

    Examples of Ethical Theories Thank you so much for sharing your insight. There are very im- portant cases in which what we identify as ethics may have, or not have been divided as such. The distinctions make us aware. Now, to may proceed. We’re looking to define these definiti...

  93. [2019]

    In Proceedings of the 2019 International Conference on Computational Creativity

    Trick or TReAT: Thematic Reinforcement for Artistic Typography. In Proceedings of the 2019 International Conference on Computational Creativity . PRE-PRINT C&C ’25, June 23–25, 2025, Virtual, United Kingdom Issak et al. ICCC

  94. [2023]

    https://doi.org/10.48550/ arXiv.2307.01850 arXiv:2307.01850

    Self-Consuming Generative Models Go MAD. https://doi.org/10.48550/ arXiv.2307.01850 arXiv:2307.01850

  95. [2025]

    Artificial Intelligence 339 (2025), 104244

    Human-AI coevolution. Artificial Intelligence 339 (2025), 104244. https: //doi.org/10.1016/j.artint.2024.104244

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.