Pith. sign in

REVIEW 3 major objections 3 minor 2 cited by

Vision-Language Models Generate More Homogeneous Stories for Phenotypically Black Individuals

T0 review · 3 major / 3 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Vision-language models generate more homogeneous stories about Black individuals with higher perceived racial phenotypicality.

desk verdict The question is good, but the statistics are broken: the reported p-values rest on non-independent, pooled pairwise similarities and cannot be taken at face value. read the letter →

arxiv 2412.09668 v2 pith:I2ENTLBA submitted 2024-12-12 cs.CV

Messi H.J. Lee , Soyeon Jeon This is my paper · ORCID
classification cs.CV
keywords racialphenotypicalityhomogeneitybiasvision-languagemodelsstereotypingsentenceembeddingscosinesimilarityintersectionalityGAN-generatedfaces
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 vision-language models (VLMs) treat members of the same racial group differently based on how typically Black their facial features look. The authors claim that when a VLM writes a short story about a computer-generated Black face, the stories become significantly more similar to each other for faces rated higher in phenotypicality. The effect appears in all three models tested (GPT-4o mini, GPT-4 Turbo, and Llama-3.2), with Black women driving the effect in two of the three models. This matters because it extends homogeneity bias from language models to multimodal AI and suggests that even within a single racial category, AI reproduces the human tendency to stereotype people with more phenotypically Black features more strongly.

What carries the argument

The machinery is a measurement pipeline that combines three components: GAN Face Database pairs that hold identity constant while manipulating facial features associated with perceived Blackness; a free-generation prompt that asks the VLM to write a 50-word story about the individual in the image; and a homogeneity metric built from Sentence-BERT embeddings (all-mpnet-base-v2) that computes pairwise cosine similarity between story embeddings. Linear mixed-effects models with Pair ID as random intercepts test whether phenotypicality, gender, and their interaction predict this similarity, with the load-bearing assumption that higher cosine similarity means less individualized representation.

What would settle it

Re-run the story-generation and similarity pipeline but add a covariate that measures the pairwise visual similarity of the face images themselves, for instance cosine similarity of image embeddings. If the phenotypicality effect on story homogeneity disappears or drops sharply once visual similarity is controlled, the central claim of stereotyping-driven homogeneity is not supported.

Watch

Extended reading notes

Core claim

The central claim is that VLMs generate more homogeneous stories about Black individuals with higher perceived racial phenotypicality than about those with lower phenotypicality, where homogeneity is measured as pairwise cosine similarity of sentence embeddings of the generated stories. Using 10 pairs of GAN-generated faces matched for identity but manipulated in phenotypicality, the authors elicited 50-word stories from three VLMs and fit mixed-effects models. They found significant positive effects of phenotypicality in all models (bs = 0.044, 0.15, and 0.080), significant gender effects with Black women represented more homogeneously than Black men, and a positive phenotypicality-by-gender interaction in GPT-4o mini and Llama-3.2, meaning the phenotypicality effect is largely carried by Black women. The paper interprets this as evidence that VLMs mirror the human racial phenotypicality bias documented in social psychology.

Load-bearing premise

The high-phenotypicality face images may be more visually similar to one another than the low-phenotypicality ones are, so the more similar stories could simply reflect more similar input images rather than social stereotyping.

Editorial extensions

If this is right

  • VLMs inherit not only between-group but also within-group stereotypical homogeneity, so audits should measure variation across phenotypicality rather than only across racial categories.
  • Because the effect is concentrated in Black women for two of three models, intersectional analysis is necessary to identify who is most affected by AI stereotyping.
  • Embedding-based cosine similarity of generated texts can serve as a scalable quantitative audit tool for homogeneity bias in generative models.
  • The pattern mirrors documented human racial phenotypicality bias, suggesting that debiasing methods must address feature-based stereotyping rather than simply making models colorblind.
  • The mixed evidence across models, with GPT-4 Turbo showing no interaction, indicates that homogeneity bias varies by model architecture and training data.

Reading between the lines

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

  • A plausible alternative mechanism is that high-phenotypicality images share more low-level visual features with each other than low-phenotypicality images do; a direct test would add image-similarity covariates to the mixed models to see if the phenotypicality effect survives.
  • A testable extension is to apply the same pipeline to real face photos matched by human-rated phenotypicality; if the effect disappears with real photos, it may be an artifact of GAN-generated feature distributions rather than a general property of VLMs.
  • Another extension is to check whether the homogeneity effect appears in other tasks such as image captioning or question answering, which would indicate whether it is a general representational bias or specific to narrative generation.
  • The interaction with gender suggests that the phenotypicality manipulation may also alter gender-typical appearance; future work should verify that the manipulation affects male and female faces equally on dimensions other than race.
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 / 3 minor

Summary. The paper investigates whether vision-language models (VLMs) generate more homogeneous stories about Black individuals with higher versus lower phenotypicality, using images from the GAN Face Database and measuring pairwise cosine similarity of Sentence-BERT embeddings of generated stories. Three VLMs are tested (GPT-4o mini, GPT-4 Turbo, Llama-3.2). The authors report a significant main effect of phenotypicality in all models, a robust gender effect (women more homogeneous), and an interaction in two models that they interpret as the phenotypicality effect being driven primarily by Black women. They frame these results as evidence that VLMs mirror human phenotypicality bias.

Significance. The research question is timely and important: extending AI bias auditing from between-group to within-group comparisons is a genuine gap, and the use of controlled GAN-generated stimuli is a methodological strength that could enable cleaner causal claims than real-world images. If the findings survive appropriate statistical treatment, they would make a useful contribution to the study of intersectional bias in multimodal models. However, the paper's current statistical analysis has a serious unit-of-analysis inconsistency that undermines the reported p-values and effect sizes, and there are internal contradictions about which models show the interaction. These issues need to be resolved before the empirical claims can be assessed.

major comments (3)
  1. [Sec. 2.3–2.4, Tables S2–S4] The reported observation count is inconsistent with the described design. The text states that cosine similarity is calculated between all combinations of stories generated for each image (Sec. 2.3) and that 50 stories per image yield 2,500 measurements per Pair ID (Sec. 2.2). With 40 images (10 sets × 2 genders × 2 phenotypicality levels), a within-image pairwise design would produce 40 × C(50,2) = 49,000 observations. Tables S2–S4 instead report 499,000 observations for every model, which equals 4 × C(500,2) — i.e., all 500 stories from the 10 images in a condition pooled together. This implies that most pairwise observations compare stories generated from different images, so the stated random intercept "Pair ID" (defined as the image set in Sec. 2.4) is undefined for the majority of observations. Moreover, in the pooled design each story embedding is reused in 499 pairwise comparisons, so the 499,000 observations are not independent; the effective sample size is at most 2,000 stories (or 40 images). The reported standard errors (e.g., 0.0026 for a coefficient of 0.044) are therefore almost certainly grossly inflated in precision, and the p < .001 values cannot be taken at face value. The authors must clarify the actual unit of analysis, re-fit the models with appropriate random effects (e.g., random intercepts for image and story identity, or analyze image/condition-level summaries), and re-report all significance tests.
  2. [Sec. 3.3 vs. Sec. 4.2, Tables S4 and S6] There is an internal contradiction about which models exhibit the gender × phenotypicality interaction. Sec. 3.3 states that the interaction is significantly positive for GPT-4o mini and Llama-3.2, and not significant for GPT-4 Turbo. Sec. 4.2, however, says 'In two of three VLMs—GPT-4 Turbo and Llama-3.2—the effect of phenotypicality ... was significantly greater for women than for men.' These statements are incompatible. In addition, the simple-slopes results in Table S6 show that for both GPT-4o mini and Llama-3.2, the effect of phenotypicality within women has a 95% confidence interval that includes zero (e.g., GPT-4o mini women: 0.040, [-0.014, 0.094]; Llama-3.2 women: 0.053, [-0.012, 0.12]), which does not support the claim that the main effect is 'primarily driven by Black women.' The authors should reconcile the narrative with the reported statistics and either temper the interpretation or provide appropriate supporting analyses.
  3. [Sec. 2.1, Sec. 5] The interpretation of the phenotypicality effect as a social-stereotyping phenomenon is potentially confounded by low-level visual similarity. The GAN manipulation holds other facial characteristics constant within a pair (lower vs. higher phenotypicality from the same set), but it does not ensure that the ten higher-phenotypicality images resemble each other more than the ten lower-phenotypicality images do (e.g., systematically darker skin or coarser hair). If the VLM generates more similar stories for images that share low-level visual features, the observed homogeneity difference could be a visual-similarity artifact rather than a reflection of social phenotypicality bias. The paper provides no control or covariate for pairwise image similarity. At minimum, the authors should discuss this alternative explanation explicitly and, ideally, include an image-similarity covariate or a per-image random slope to demonstrate that the effect is not explained by low-level visual resemblance.
minor comments (3)
  1. [Sec. 2.2] The power-analysis numbers do not match the described design: C(50,2) = 1,225, but the text says 1,245 measurements per pair of stimuli, and then says 2,500 per Pair ID. The arithmetic should be corrected and clarified.
  2. [Sec. 3.3, Table S4] For GPT-4 Turbo, the text reports the interaction as 'b = -0.0048, p = .038' but also calls it not significant. The SE (0.0055) gives z ≈ -0.87, p ≈ .38, consistent with the likelihood-ratio test but not with p = .038. This is likely a typographical error but should be fixed.
  3. [Figures 3 and Tables S2–S4] The captions refer to 'all four VLMs' and Tables S2–S4 say 'across all four VLMs,' but only three models are used in the study. This should be corrected to 'three.'

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the paper measures a new empirical phenomenon with a borrowed metric; the only self-citation is the metric's origin, which is not load-bearing.

full rationale

The paper's central claim is an empirical measurement, not a derivation. Homogeneity is defined as pairwise cosine similarity between sentence embeddings of stories generated by VLMs, and the effect of phenotypicality is estimated as a fixed effect in mixed models on newly collected outputs. Nothing is fitted to the target result, and no prediction is an algebraic consequence of the measure. The image pairs and stories are new data; the GANFD stimuli and VLM outputs are external to the cited prior work. The main self-citations (Lee et al., 2024, for the cosine-similarity measure and power-analysis effect size; Lee et al., 2025, for prior VLM results) supply methodology and context, not the conclusion. Even though the first author overlaps with those papers, the cited measure is a standard embedding-similarity operation and the present study's findings depend on fresh model generations, so the citation is not load-bearing. The reviewer's concern about non-independence of the 499,000 pairwise observations is a statistical validity issue, not circularity. Score 2 reflects the minor same-group self-citation in the method, not any reduction of the result to its inputs.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical entities, particles, or mechanisms. Its assumptions are about the validity of the measurement construct, the stimulus selection, and the statistical model. The main unexamined assumption is the visual-similarity confound, which could threaten the stereotyping interpretation.

free parameters (2)
  • Effect size for power analysis = 0.30
    Chosen from Lee et al. (2024) for power analysis (Section 2.2), not fitted to the current data. It does not affect the reported results, only the planned number of stories.
  • Number of stories per image = 50
    Selected to reach the power target (Section 2.2). A design choice, not a fitted parameter.
assumptions (4)
  • domain assumption Sentence-BERT cosine similarity between story embeddings is a valid operationalization of homogeneity.
    The paper uses pairwise cosine similarity of all-mpnet-base-v2 embeddings as the sole measure of homogeneity (Section 2.3). The Limitations section acknowledges this is a black-box proxy and does not identify which textual features drive similarity.
  • domain assumption The GAN Face Database human perceived-Blackness ratings are a valid measure of racial phenotypicality.
    Image selection and the lower/higher phenotypicality labels depend entirely on human 'Blackness' ratings in GANFD (Section 2.1). No independent validation of these ratings is provided.
  • domain assumption Pairwise cosine similarity observations are sufficiently independent for mixed-effects models despite shared stories.
    Each story appears in many pairwise comparisons, creating non-independence. The model uses only Pair ID random intercepts, not story-level random effects (Section 2.4), which may underestimate standard errors.
  • domain assumption The high-phenotypicality GAN faces are not more visually similar to each other than the low-phenotypicality faces in a way that drives story similarity.
    The design claims to hold other visual characteristics constant (Section 2.1), but the GAN manipulation changes features along a phenotypicality axis, so all high-phenotypicality images may share low-level features that could themselves increase story similarity. This is not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Vision-Language Models Generate More Homogeneous Stories for Phenotypically Black Individuals." pith.science (2026). https://pith.science/paper/I2ENTLBA

@misc{pith2026241209668,
  author       = {Pith},
  title        = {Pith review of: Vision-Language Models Generate More Homogeneous Stories for Phenotypically Black Individuals},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I2ENTLBA}},
  note         = {Machine review of arXiv:2412.09668}
}
read the original abstract

Vision-Language Models (VLMs) extend Large Language Models' capabilities by integrating image processing, but concerns persist about their potential to reproduce and amplify human biases. While research has documented how these models perpetuate stereotypes across demographic groups, most work has focused on between-group biases rather than within-group differences. This study investigates homogeneity bias-the tendency to portray groups as more uniform than they are-within Black Americans, examining how perceived racial phenotypicality influences VLMs' outputs. Using computer-generated images that systematically vary in phenotypicality, we prompted VLMs to generate stories about these individuals and measured text similarity to assess content homogeneity. Our findings reveal three key patterns: First, VLMs generate significantly more homogeneous stories about Black individuals with higher phenotypicality compared to those with lower phenotypicality. Second, stories about Black women consistently display greater homogeneity than those about Black men across all models tested. Third, in two of three VLMs, this homogeneity bias is primarily driven by a pronounced interaction where phenotypicality strongly influences content variation for Black women but has minimal impact for Black men. These results demonstrate how intersectionality shapes AI-generated representations and highlight the persistence of stereotyping that mirror documented biases in human perception, where increased racial phenotypicality leads to greater stereotyping and less individualized representation.

Figures

Figures reproduced from arXiv: 2412.09668 by the authors.

Figure 1
Figure 1. Summary of the experimental setup. We collect 50-word stories about Black individuals differing in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Two face stimulus pairs representing (A) Black men and (B) Black women. In each pair, the left image depicts [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Standardized cosine similarity values of Black individuals with lower versus higher phenotypicality ratings [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Standardized cosine similarity values of Black men and women. Higher standardized cosine similarity means [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Standardized cosine similarity values of Black men and women with lower and higher phenotypicality. Higher [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Token-Level Entropy Reveals Demographic Disparities in Large Language Models

    cs.CL 2025-01 unverdicted novelty 7.0 of 10

    The abstract's claim of demographic entropy disparities is directly contradicted by the paper's own results section, which finds token sampling uncertainty does not explain homogeneity bias.

  2. Autonomous VR-Based Risk Detection for Situational Awareness in Dangerous Settings

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A robot using GPT-4o labeled hazards in a simulated disaster room, and VR users preferred and rated these annotations highly, though the study lacks a controlled baseline comparison.

Reference graph

Works this paper leans on

38 extracted references · 18 canonical work pages · cited by 2 Pith papers

  1. [3]

    In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21)

    On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21). Association for Computing Machinery, New York, NY , USA, 610–623. https://doi.org/10.1145/3442188.3445922 Federico Bianchi, Pratyusha Kalluri, Esin Durmus, Faisal Ladhak, Myra Cheng, De...

    arXiv 2021

  2. [5]

    In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.)

    CoMPosT: Characterizing and Evaluating Caricature in LLM Simulations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational Linguistics, Singapore, 10853–10875. https://doi.org/10.18653/v1/2023.emnlp-main.669 Sunipa Dev, Masoud Monajatipoor...

    doi:10.18653/v1/2023.emnlp-main.669 2023

  3. [6]

    Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologies. https://doi.org/10.48550/arXiv.2108.12084 arXiv:2108.12084 Sunipa Dev, Emily Sheng, Jieyu Zhao, Aubrie Amstutz, Jiao Sun, Yu Hou, Mattie Sanseverino, Jiin Kim, Akihiro Nishi, Nanyun Peng, and Kai-Wei Chang

    arXiv doi:10.48550/arxiv.2108.12084

  4. [10]

    https: //doi.org/10.48550/arXiv.2407.21783 Peter Green, Catriona MacLeod, and Phillip Alday

    The Llama 3 Herd of Models. https: //doi.org/10.48550/arXiv.2407.21783 Peter Green, Catriona MacLeod, and Phillip Alday

    arXiv doi:10.48550/arxiv.2407.21783

  5. [11]

    Nature Medicine 30, 2 (Feb

    Deep Learning-Aided Decision Support for Diagnosis of Skin Disease across Skin Tones. Nature Medicine 30, 2 (Feb. 2024), 573–583. https://doi.org/10.1038/s41591-023-02728-3 9 A PREPRINT - MARCH 21, 2025 Nao Hagiwara, Deborah A. Kashy, and Joseph Cesario

    work page doi:10.1038/s41591-023-02728-3 2024

  6. [19]

    In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24)

    Large Language Models Portray Socially Subordinate Groups as More Homogeneous, Consistent with a Bias Observed in Humans. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24). Association for Computing Machinery, New York, NY , USA, 1321–1340. https://doi.org/10.1145/3630106.3658975 Messi H. J. Lee, Soyeon Je...

    arXiv 2024

  7. [21]

    Political Analysis 31, 3 (July 2023), 366–379

    Cross-Lingual Classification of Political Texts Using Multilingual Sentence Embeddings. Political Analysis 31, 3 (July 2023), 366–379. https://doi.org/10.1017/pan.2022.29 Gechun Lin

    work page doi:10.1017/pan.2022.29 2023

  8. [22]

    American Journal of Political Science n/a, n/a (March 2025), 1–17

    Using Cross-Encoders to Measure the Similarity of Short Texts in Political Science. American Journal of Political Science n/a, n/a (March 2025), 1–17. https://doi.org/10.1111/ajps.12956 Patricia W. Linville, Gregory W. Fischer, and Peter Salovey

    work page doi:10.1111/ajps.12956 2025

Show all 38 references
  1. [24]

    In Proceedings of the Third Workshop on Narrative Understanding, Nader Akoury, Faeze Brahman, Snigdha Chaturvedi, Elizabeth Clark, Mohit Iyyer, and Lara J

    Gender and Representation Bias in GPT-3 Generated Stories. In Proceedings of the Third Workshop on Narrative Understanding, Nader Akoury, Faeze Brahman, Snigdha Chaturvedi, Elizabeth Clark, Mohit Iyyer, and Lara J. Martin (Eds.). Association for Computational Linguistics, Virt...

    doi

  2. [26]

    In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society

    Social Biases through the Text-to-Image Generation Lens. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. ACM, Montr\’{e}al QC Canada, 786–808. https: //doi.org/10.1145/3600211.3604711 José C. Pinheiro and Douglas M. Bates

    2023

  3. [27]

    In Mixed- Effects Models in S and S-PLUS

    Linear Mixed-Effects Models: Basic Concepts and Examples. In Mixed- Effects Models in S and S-PLUS. Springer, New York, NY , 3–56.https://doi.org/10.1007/0-387-22747-4_ 1 10 A PREPRINT - MARCH 21, 2025 George A. Quattrone and Edward E. Jones

    2025 doi

  4. [29]

    https://doi.org/10.48550/arXiv.1908.10084 arXiv:1908.10084 [cs] Mansour Sami, Ashkan Sami, and Pete Barclay

    Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks. https://doi.org/10.48550/arXiv.1908.10084 arXiv:1908.10084 [cs] Mansour Sami, Ashkan Sami, and Pete Barclay

    1908 doi arXiv

  5. [30]

    In 2023 IEEE International Conference on Software Maintenance and Evolution (ICSME)

    A Case Study of Fairness in Generated Images of Large Language Models for Software Engineering Tasks. In 2023 IEEE International Conference on Software Maintenance and Evolution (ICSME). 391–396. https://doi.org/10.1109/ICSME58846.2023.00051 Henrik Singmann, Ben Bolker, Jake W...

    2023

  6. [31]

    Journal of Experimental Social Psychology 48, 4 (2012), 867–878

    The Role of Skin Color and Facial Physiognomy in Racial Categoriza- tion: Moderation by Implicit Racial Attitudes. Journal of Experimental Social Psychology 48, 4 (2012), 867–878. https://doi.org/10.1016/j.jesp.2012.02.019 Elena V . Stepanova and Michael J Strube

    2012 doi

  7. [33]

    Journal of Computer-Mediated Communication 29, 1 (Nov

    Smiling Women Pitching down: Auditing Representational and Presentational Gender Biases in Image-Generative AI. Journal of Computer-Mediated Communication 29, 1 (Nov. 2023), zmad045. https://doi.org/10.1093/jcmc/zmad045 T. Joel Wade, Melanie Judkins Romano, and Leslie Blue

    2023 doi

  8. [34]

    Journal of Applied Social Psychology 34, 12 (2004), 2550–2558

    The Effect of African American Skin Color on Hiring Preferences. Journal of Applied Social Psychology 34, 12 (2004), 2550–2558. https://doi.org/10.1111/j. 1559-1816.2004.tb01991.x Le Xue, Manli Shu, Anas Awadalla, Jun Wang, An Yan, Senthil Purushwalkam, Honglu Zhou, Viraj Prab...

    2004

  9. [35]

    https://doi.org/10.48550/arXiv.2408.08872 arXiv:2408.08872 Dora Zhao, Angelina Wang, and Olga Russakovsky

    xGen-MM (BLIP-3): A Family of Open Large Multimodal Models. https://doi.org/10.48550/arXiv.2408.08872 arXiv:2408.08872 Dora Zhao, Angelina Wang, and Olga Russakovsky

    doi

  10. [36]

    In 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Understanding and Evaluating Racial Biases in Image Captioning. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV). IEEE, Montreal, QC, Canada, 14810–14820. https://doi.org/10.1109/ICCV48922.2021.01456 Kankan Zhou, Eason Lai, and Jing Jiang

    2021

  11. [37]

    VLStereoSet: A Study of Stereotypical Bias in Pre-trained Vision- Language Models. In Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Com- putational Linguistics and the 12th International Joint Conference on Natural Language Processing (Vo...

    2025

  12. [38]

    The woman in the image is a beautiful black woman with curly hair and dark brown eyes. She has a serious expression and is looking at the camera

    and Claude 3.7 Sonnet (Anthropic, 2025), were excluded from our analysis because they refused to generate stories based on facial images. BLIP-3 produced visual descriptions instead (e.g., "The woman in the image is a beautiful black woman with curly hair and dark brown eyes. ...

    2025

  13. [1980]

    Journal of Personality and Social Psychology38, 1 (1980), 141–152

    The Perception of Variability within In-Groups and out-Groups: Implications for the Law of Small Numbers. Journal of Personality and Social Psychology38, 1 (1980), 141–152. https://doi.org/10.1037/0022-3514.38.1.141 Nils Reimers and Iryna Gurevych

    1980 doi

  14. [1989]

    Journal of Personality and Social Psychology 57, 2 (1989), 165–188

    Perceived Distributions of the Characteristics of In-Group and out-Group Members: Empirical Evidence and a Computer Simulation. Journal of Personality and Social Psychology 57, 2 (1989), 165–188. https://doi.org/10.1037/0022-3514.57.2.165 Li Lucy and David Bamman

    1989 doi

  15. [1991]

    Skin Tone and Stratification in the Black Community. Amer. J. Sociology97, 3 (1991), 760–778. arXiv:2781783 Newton M. Kinyanjui, Timothy Odonga, Celia Cintas, Noel C. F. Codella, Rameswar Panda, Prasanna Sattigeri, and Kush R. Varshney

    1991

  16. [1995]

    Annals of Human Biology 22, 1 (1995), 69–73

    Skin Colour, Measures of Socioeconomic Status, and Blood Pressure among Blacks in Erie County, NY. Annals of Human Biology 22, 1 (1995), 69–73. https://doi.org/10.1080/03014469500003712 Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad ...

    1995 doi

  17. [2000]

    Journal of Behavioral Medicine 23, 4 (Aug

    Is Skin Color a Marker for Racial Discrimination? Explaining the Skin Color- Hypertension Relationship. Journal of Behavioral Medicine 23, 4 (Aug. 2000), 329–338. https://doi.org/10. 1023/a:1005580300128 Messi H.J. Lee, Jacob M. Montgomery, and Calvin K. Lai

  18. [2002]

    https://doi.org/10.2307/3090169 arXiv:3090169 Kimberly Barsamian Kahn and Paul G

    Skin Color and the Perception of Attractiveness among African Americans: Does Gender Make a Difference? Social Psychology Quarterly 65, 1 (2002), 77–91. https://doi.org/10.2307/3090169 arXiv:3090169 Kimberly Barsamian Kahn and Paul G. Davies

    2002 doi

  19. [2004]

    Personality and Social Psychology Review: An Official Journal of the Society for Personality and Social Psychology, Inc 8, 4 (2004), 383–401

    Perspectives on Racial Phenotypicality Bias. Personality and Social Psychology Review: An Official Journal of the Society for Personality and Social Psychology, Inc 8, 4 (2004), 383–401. https: //doi.org/10.1207/s15327957pspr0804_4 Art D. Marsden, Alexandria Jaurique, Mackenzi...

    2004 doi

  20. [2009]

    Journal of Applied Social Psychology 39, 1 (2009), 134–168

    The Hidden Prejudice in Selection: A Research Investigation on Skin Color Bias. Journal of Applied Social Psychology 39, 1 (2009), 134–168. https://doi.org/10.1111/j. 1559-1816.2008.00433.x Mark E. Hill

    2009

  21. [2011]

    Group Processes & Intergroup Relations 14, 4 (2011), 569–580

    Differentially Dangerous? Phenotypic Racial Stereotypicality Increases Implicit Bias among Ingroup and Outgroup Members. Group Processes & Intergroup Relations 14, 4 (2011), 569–580. https://doi.org/10.1177/1368430210374609 Verna M. Keith and Cedric Herring

    2011 doi

  22. [2012]

    Journal of Experimental Social Psychology 48, 4 (July 2012), 892–898

    The Independent Effects of Skin Tone and Facial Features on Whites’ Affective Reactions to Blacks. Journal of Experimental Social Psychology 48, 4 (July 2012), 892–898. https://doi.org/10.1016/j.jesp.2012.02.001 Matthew S. Harrison and Kecia M. Thomas

    2012 doi

  23. [2014]

    Fitting Linear Mixed-Effects Models Using Lme4. https://doi.org/10.48550/arXiv.1406.5823 arXiv:1406.5823 Douglas Bates, Martin Maechler, Ben Bolker [aut, cre, Steven Walker, Rune Haubo Bojesen Christensen, Henrik Singmann, Bin Dai, Fabian Scheipl, Gabor Grothendieck, Peter Gre...

    doi arXiv

  24. [2018]

    The Journal of General Psychology 145, 1 (Jan

    Attractiveness as a Function of Skin Tone and Facial Features: Evidence from Categorization Studies. The Journal of General Psychology 145, 1 (Jan. 2018), 1–20. https: //doi.org/10.1080/00221309.2017.1394811 Luhang Sun, Mian Wei, Yibing Sun, Yoo Ji Suh, Liwei Shen, and Sijia Yang

    2018 arXiv

  25. [2019]

    https://doi.org/10.48550/arXiv.1910.13268 arXiv:1910.13268 [cs] E

    Estimating Skin Tone and Effects on Classification Performance in Dermatology Datasets. https://doi.org/10.48550/arXiv.1910.13268 arXiv:1910.13268 [cs] E. A. Klonoff and H. Landrine

    1910 doi arXiv

  26. [2021]

    https://doi.org/10.48550/arXiv.2101.05783 arXiv:2101.05783 [cs] Anthropic

    Persistent Anti-Muslim Bias in Large Language Models. https://doi.org/10.48550/arXiv.2101.05783 arXiv:2101.05783 [cs] Anthropic

    doi arXiv

  27. [2022]

    In Findings of the Association for Computational Linguistics: AACL-IJCNLP 2022, Yulan He, Heng Ji, Sujian Li, Yang Liu, and Chua-Hui Chang (Eds.)

    On Measures of Biases and Harms in NLP. In Findings of the Association for Computational Linguistics: AACL-IJCNLP 2022, Yulan He, Heng Ji, Sujian Li, Yang Liu, and Chua-Hui Chang (Eds.). Association for Computational Linguistics, Online only, 246–267. https://doi.org/10.18653/...

    2022 doi

  28. [2023]

    https://doi.org/10.48550/arXiv.2211.03759 arXiv:2211.03759 Joy Buolamwini and Timnit Gebru

    Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale. https://doi.org/10.48550/arXiv.2211.03759 arXiv:2211.03759 Joy Buolamwini and Timnit Gebru

    doi arXiv

  29. [2024]

    https://doi.org/ 10.48550/arXiv.2410.06415 arXiv:2410.06415 L

    Biased AI Can Influence Political Decision-Making. https://doi.org/ 10.48550/arXiv.2410.06415 arXiv:2410.06415 L. Gleiberman, E. Harburg, M. R. Frone, M. Russell, and M. L. Cooper

    doi

  30. [2025]

    https://doi.org/10.48550/arXiv.2503.05093 arXiv:2503.05093 [cs] Russell V

    Visual Cues of Gender and Race Are Associated with Stereotyping in Vision-Language Models. https://doi.org/10.48550/arXiv.2503.05093 arXiv:2503.05093 [cs] Russell V . Lenth, Ben Bolker, Paul Buerkner, Iago Giné-Vázquez, Maxime Herve, Maarten Jung, Jonathon Love, Fernando Migue...

    doi arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.