@inproceedings{azhar-etal-2026-representation,
title = "On the Representation Bias of Disabilities in Generative Text-to-Image Models",
author = "Azhar, Asfandyar and
Khan, Zaid and
Mandal, Shaurjya and
Shah, Nidhish and
Youssef, Alaa and
Adenaiye, Oluwasanmi and
Langlotz, Curtis and
Dicianno, Brad",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.416/",
pages = "8770--8785",
ISBN = "979-8-89176-462-0",
abstract = "Emerging text-to-image (T2I) models perpetuate biases, creating stereotypical portrayals of disability. This study investigates whether an agent-based, human-in-loop, AI workflow, that iteratively refines user-provided queries, can produce fairer representations of disability compared to standard models like Stable Diffusion, DALL{\textperiodcentered}E-3, and MidJourney. Sixteen raters (eight with disabilities and eight without) evaluated images generated from models, prompted with textual inputs that targeted professional, medical, or social contexts. In an initial survey, raters were asked if an image depicted a disability, the type of disability if present, and if the setting aligned with the given text prompt to the model. Qualitative comments were collected. A post-survey asked raters to choose which image of a pair (standard vs. agent-based workflow) more accurately represented disability. Results revealed that standard models disproportionately produced images featuring physical disabilities or wheelchairs ($p < 0.001$), consistent with broader dataset biases. In contrast, the agent-based model{'}s images were deemed more representative in approximately two-thirds of comparisons ($p < 0.0001$), especially among raters with disabilities. Qualitative analyses underscored the need for nuanced, contextually rich portrayals.These findings suggest disability bias and that targeted debiasing can help mitigate stereotypes in generative AI, fostering more equitable outputs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="azhar-etal-2026-representation">
<titleInfo>
<title>On the Representation Bias of Disabilities in Generative Text-to-Image Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Asfandyar</namePart>
<namePart type="family">Azhar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zaid</namePart>
<namePart type="family">Khan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaurjya</namePart>
<namePart type="family">Mandal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nidhish</namePart>
<namePart type="family">Shah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alaa</namePart>
<namePart type="family">Youssef</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Oluwasanmi</namePart>
<namePart type="family">Adenaiye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Curtis</namePart>
<namePart type="family">Langlotz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Brad</namePart>
<namePart type="family">Dicianno</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Emerging text-to-image (T2I) models perpetuate biases, creating stereotypical portrayals of disability. This study investigates whether an agent-based, human-in-loop, AI workflow, that iteratively refines user-provided queries, can produce fairer representations of disability compared to standard models like Stable Diffusion, DALL·E-3, and MidJourney. Sixteen raters (eight with disabilities and eight without) evaluated images generated from models, prompted with textual inputs that targeted professional, medical, or social contexts. In an initial survey, raters were asked if an image depicted a disability, the type of disability if present, and if the setting aligned with the given text prompt to the model. Qualitative comments were collected. A post-survey asked raters to choose which image of a pair (standard vs. agent-based workflow) more accurately represented disability. Results revealed that standard models disproportionately produced images featuring physical disabilities or wheelchairs (p < 0.001), consistent with broader dataset biases. In contrast, the agent-based model’s images were deemed more representative in approximately two-thirds of comparisons (p < 0.0001), especially among raters with disabilities. Qualitative analyses underscored the need for nuanced, contextually rich portrayals.These findings suggest disability bias and that targeted debiasing can help mitigate stereotypes in generative AI, fostering more equitable outputs.</abstract>
<identifier type="citekey">azhar-etal-2026-representation</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.416/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>8770</start>
<end>8785</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T On the Representation Bias of Disabilities in Generative Text-to-Image Models
%A Azhar, Asfandyar
%A Khan, Zaid
%A Mandal, Shaurjya
%A Shah, Nidhish
%A Youssef, Alaa
%A Adenaiye, Oluwasanmi
%A Langlotz, Curtis
%A Dicianno, Brad
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F azhar-etal-2026-representation
%X Emerging text-to-image (T2I) models perpetuate biases, creating stereotypical portrayals of disability. This study investigates whether an agent-based, human-in-loop, AI workflow, that iteratively refines user-provided queries, can produce fairer representations of disability compared to standard models like Stable Diffusion, DALL·E-3, and MidJourney. Sixteen raters (eight with disabilities and eight without) evaluated images generated from models, prompted with textual inputs that targeted professional, medical, or social contexts. In an initial survey, raters were asked if an image depicted a disability, the type of disability if present, and if the setting aligned with the given text prompt to the model. Qualitative comments were collected. A post-survey asked raters to choose which image of a pair (standard vs. agent-based workflow) more accurately represented disability. Results revealed that standard models disproportionately produced images featuring physical disabilities or wheelchairs (p < 0.001), consistent with broader dataset biases. In contrast, the agent-based model’s images were deemed more representative in approximately two-thirds of comparisons (p < 0.0001), especially among raters with disabilities. Qualitative analyses underscored the need for nuanced, contextually rich portrayals.These findings suggest disability bias and that targeted debiasing can help mitigate stereotypes in generative AI, fostering more equitable outputs.
%U https://aclanthology.org/2026.findings-emnlp.416/
%P 8770-8785
Markdown (Informal)
[On the Representation Bias of Disabilities in Generative Text-to-Image Models](https://aclanthology.org/2026.findings-emnlp.416/) (Azhar et al., Findings 2026)
ACL
- Asfandyar Azhar, Zaid Khan, Shaurjya Mandal, Nidhish Shah, Alaa Youssef, Oluwasanmi Adenaiye, Curtis Langlotz, and Brad Dicianno. 2026. On the Representation Bias of Disabilities in Generative Text-to-Image Models. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 8770–8785, Budapest, Hungary. Association for Computational Linguistics.