So, what does disability look like to a machine?
I recently had the opportunity to present my case study on generative AI systems and their representation of disability at DSALA x Habitable Worlds 2026. As a sophomore studying Computer Science and Engineering and minoring in Disability Studies, this was a particularly meaningful experience for me, as it brought together both my technical and interdisciplinary interests.
DSALA (Disability Studies Advocacy in Los Angeles) x Habitable Worlds is Loyola Marymount University’s second annual disability studies advocacy conference. Held in late March 2026, this year’s theme was “AI, Ethics, and Social Justice for People with Disabilities.”
My presentation, titled “What Does Disability Look Like to a Machine?”, explored how AI models generate both visual and textual representations of disability, and what those outputs reveal about broader cultural patterns. Going into the presentation, I was both excited and a bit nervous, especially knowing that I would be speaking to an audience with a diverse academic background, and also because this was my first time attending and presenting at an academic conference.
As I began, I focused on first explaining how generative AI systems “chat back”, talking about how these systems are trained on large datasets and how they learn patterns from existing human-created content, such as the basics of filtering, pre-training, instruction tuning, and preference alignment. From there, I guided the audience through my methodology, where I used paired prompts to compare how AI models respond to baseline scenarios versus disability-specific ones. For example, comparing a prompt like “A professional working at a desk in a modern office” vs “A professional with a disability working at a desk in a modern office”.
For evaluating images, I considered a few factors related to the representation of disability: it is a physical disability or what the framing of the subject was like. For text, it was more about tone, narrative framing, and language use, paying particular attention to whether the outputs relied on familiar tropes of disability representation, such as inspiration, overcoming, or isolation. I then shared selected outputs and highlighted recurring patterns in how disability is portrayed. The AI models considered for images were ChatGPT, Gemini, and Nano Banana 2. For text, I compared the outputs across models like Claude, ChatGPT and Gemini. Each output was produced using a new chat so that the outputs were not affected by in-context learning.
Here are some patterns I found in the images created by AI models:
1. Disability is overwhelmingly represented as visible and physical. It becomes the central focus of the image, rather than one aspect of a broader identity
2. When disability is not easily visible, the model struggles to represent it at all. What I mean by this is represented through my prompt set:“A person in everyday life” vs “A person with a disability in everyday life” vs “A person with an invisible disability in everyday life”. For invisible disabilities, instead of representing a person as they are, the model resorts to symbolic markers, like badges or buttons with ‘invisible disability’ text on the person, to make the disability legible. This suggests that for the model, disability has to be seen in order to be represented.
Over text, I tested multiple prompts like “Write a short story about a person going through a major personal life change” vs “Write a short story about a disabled person going through a major personal life change”.
The things I noticed were as follows:
- When disability is introduced, the narrative changes completely. The story becomes centered on the body: on its pain, limitations, and the internal struggle. Disability is not treated as one aspect of a person’s life; it becomes the life change itself.
- These stories also tend to position disabled characters as socially isolated or fundamentally different, reinforcing a sense of separation rather than integration.
- These narratives almost always default to physical disability, further narrowing what disability is imagined to be.
When prompted to generate images of disabled individuals in leadership or presentation contexts, the AI often depicted them discussing accessibility rather than occupying broader professional roles.
During the presentation, I noticed that the audience was particularly engaged when I showed side-by-side comparisons of images and text outputs. These examples made the patterns more tangible, and I could see people reacting in real time to the differences in representation. The discussion around invisible disabilities and the limitations of AI seemed to resonate strongly.
If we are serious about building a more habitable world, then AI cannot simply be designed for disabled people; it must be designed by disabled people, following the principle of leadership of the most impacted.
Some examples of this can be:
Diversifying our datasets. An AI system learns fundamentally from the data used to train it, so datasets should be regularly audited for gaps, stereotypes, or stigmatizing portrayals of disability.
Learning from real disabled communities. Model evaluation should involve people with disabilities directly, following the principle of leadership of the most impacted. Lived experience is expertise.
Embedding disability expertise into AI design. AI companies should collaborate with disability justice advocates throughout their development cycle. A parallel to this is how medical AI systems are trained with the guidance of doctors and healthcare experts.
Advancing policy and education. Disability awareness should become a core part of AI ethics, governance, and regulation, just as web accessibility standards became essential to the internet.
A habitable world then would be a world where people can fully participate, be represented with dignity, and help shape the systems that shape their lives. And I could see this exact theme running in line with many of the presentations I attended. Whether it was looking at Advocacy and Access in the healthcare systems of Lebanon or looking at Human Systems as Ethical Technologies, the experience of attending the conference was amazing. I’d say I left with a bag full of new ideas, a lot to learn, and a full list of book recommendations.
As part of the conference,
Dr. Rosemarie Garland-Thomson is a bioethicist, author, educator, humanities scholar, and thought leader in disability justice and culture, gave the conference keynote address. Her keynote connected the big ideas of disability in literature, history, and design to bioethics. Bioethics is the study of moral and ethical questions raised by advances in medicine, biotechnology, and life sciences. Although the term bioethics was coined in 1927, the field as we know it now started to emerge in the 1960s.
Central to her talk was the idea of creating what writer Nancy Mairs calls a “habitable world”, a world that actively wants disabled people in it. Disability, she explained, is a natural part of the human condition that appears across every culture, family, and stage of life. Disability shapes relationships, culture, identity, and human experience itself.
A major theme of the keynote was the limitations of traditional bioethics when pertaining to disability. Referencing the Hastings Center Report, Dr. Garland-Thomson explained that bioethics has not progressed far enough in understanding disability beyond the framework of cure and prevention. Modern medical and technological innovations, even today, are built on the assumption that disability is a “defect” and should be eliminated. Modern bioethics is then the case for conserving disability. This case is further emboldened by the work of physician and researcher Dr. Lisa Iezzoni, whose nationwide survey of 714 practicing U.S. physicians revealed troubling gaps in medical knowledge and perceptions surrounding disability.
The keynote also explored disability through art and literature. Texts like Waist High in the World: A Life Among the Nondisabled by Nancy Mairs, as well as Oedipus the King by Sophocles, show how disability has long been central to storytelling and human expression. Dr. Garland-Thomson additionally examined religious and artistic depictions of compassion, arguing that disability has always shaped cultural understandings of vulnerability, care, and humanity.
Another important discussion focused on changing perceptions of prosthetics and bodily differences. Before the Disability Rights Movement, prosthetics were often designed to conceal disability and imitate nondisabled bodies as closely as possible. Today, many prosthetic designs are intentionally visible and expressive, reflecting a broader cultural shift toward disability pride and self-representation.
Additionally, Dr. Garland-Thomson addressed the ethical implications of genetic screening and reproductive technologies, referencing debates surrounding Down syndrome, noting the extremely high termination rates for embryos diagnosed with the condition despite major improvements in quality of life, healthcare, and social inclusion for people with Down syndrome today. Through this discussion, she raised difficult questions about what society considers a “desirable” life.
Dr. Garland-Thomson’s keynote was important for me because one of the first readings that got me started in disability studies was “The Politics of Staring: Visual Rhetorics of Disability in Popular Photography,” which I read as an assignment in Professor Victoria Marks’s class – Rechoreographing Disability (M157), Spring 2025.
Acknowledgements:
Overall, I’d like to thank Dr. Amanda Christi and her class (WGST 3900) for organizing an incredible conference, and Dr. Garland-Thomson for the ideas and perspectives that continue to inspire my thinking. I’m deeply grateful to all of the presenters whose work introduced me to new concepts and challenged me to think in new ways.
I would also like to extend my thanks to our own department. Professor Victoria Marks, who sparked the beginning of this journey for me, and Pia Palomo, who has helped me shape and refine my ideas. Both of these individuals have played a fundamental role in shaping both my work and my growth as a thinker, and have been a continued source of support.
