Critical Examination of AI-Based Text-to-Image Tools for Disability Representation

Topic Description

Generative artificial intelligence (GAI) become more and
more accessible to mainstream users in the recent years. Tools that use text-toimage
models uses in-texts prompts and generate an image corresponding to the
prompt. Considering the data that they are trained on, there are major concerns
regarding how they may recreate the stereotypes that we encounter in real life,
including how disabled people are perceived and represented.
The main goal of this thesis to 1) critically examine work of (resulting images)
selected generative artificial intelligence based text-to-image tools through the
lenses of different image analysis methods (Iconographic Analysis, Iconological
Analysis, Contextual Analysis, Psychological Analysis, Feminist/Gender Analysis,
Marxist/Socio-Economic Analysis, Structuralist and Semiotic Analysis and Postcolonial
Analysis). Through the questions of these analysis methods, the mechanism
behind the prompt engineering will be critically examined. 2) Generate
list of questions for people to critically reflect upon generated images for representing
disabled people, from Disability Justice perspective 3) Come up with
recommendations for prompting that would lead to more diverse and justice oriented
representations of disabled people.
Contribution: This thesis is expected to contribute to the understanding
of disability representation in generative artificial intelligence based
text-to-image tools. It is part of the Real-World Lab Accessibility, also see:
http://www.accessibility.kit.edu.

Task Summary

1. Desk Research: Research on selected text-to-image models on disability representation, art/image analysis methods and Disability Justice perspective.
2. Critical analysis: Apply structural analysis using image analysis methodologies to critically examine results of selected text-toimage models for representing disabled people
3. Questions and Recommendations: Come up with questions and recommendations for prompt engineering methodologies to gain justice oriented perspective when representing disabled people

Requirements

• Critical thinking
• Prompt engineering
• HCI

Contact

zeynep.yildiz∂kit.edu