Can AI Identify Ableism?
Despite the widespread adoption of AI to detect and mitigate harmful content, little is known about how effectively these models identify and explain speech targeted at PwD. To this end, I curated and labelled a first-of-its-kind ableist speech dataset, and obtained harm assessments from AI models and PwD to measure their alignment on harm. Our findings revealed significant gaps in AI's ability to identify and explain ableist speech! PwD described AI outputs as “dehumanizing,” highlighting how LLMs echo a clinical, outsider perspective on disability rather than lived experiences. [FAccT '25]
Supervised By:
Dr. Aditya Vashistha
Ableism (and AI) Across Cultures
Ableism in the Global South is intensified by existing socio-economic disparities, infrastructural challenges, and cultural stigmas, yet most research on disability and AI remains anchored in Western contexts. To this end, I adapted an ableist speech dataset to Hindi, an official langauge of India, gathered AI assessments, and recruited disabled participants across India to share their assessments of ableism. Largely, LLMs struggled to identify ableist speech in Hindi and surfaced deep cultural disconnects, with LLMs largely attuned to Western framings of ableism (i.e., “inspiration porn"). [AIES '25]
Supervised By:
Dr. Aditya Vashistha
Disability Bias in AI Hiring Scenarios
A recent Bureau of Labor Statistics report [February 2023] reports a troubling statistic: people with disabilities (PwD) were much less likely to be employed than those with no disability. With the rise of AI-enabled recruiting tools, it is more important than ever to examine how these technologies might exacerbate biases against PwD. Consolidating a corpus of LLM-generated conversations between hiring managers, I apply a Disability Justice (cite) lens to identify harm, focussing on intersectional identities and evaluations of harm.
Supervised By:
Dr. Aditya Vashistha, Dr. Tanu Mitra
Benchmarking Ableism
Work in Progress!
Ableist Microagressions Online
With growing disability advocacy on social media, disabled people are susceptible to discrimination, harassment, and
harm, especially implicit and subtle forms of harm. We studied how people with disabilities experience microaggressions (e.g., patronization, invalidation) on social media platforms,
shedding light into technology-mediated ableism.
[ASSETS '22]
Lead by:
Sharon Heung, Cornell Tech
Supervised by:
Dr. Aditya Vashistha, Cornell University
Dr. Megh Marathe, Michigan State
Video Accessibility for Blind People
We explored ways to improve contextual and 360° video accessibility for blind and low vision viewers with immersion and engagement as key pillars, facets not provided by current-day audio descriptions (AD). We conducted design workshops and synthesized key accessibility guidelines through discussions with professional describers, voice actors, sound designers, and AD users. [ASSETS '23, CHI '24]
Lead by:
Lucy Jiang, University of Washington
Supervised by:
Dr. Abigail Stangl, Georgia Tech
Dr. Shiri Azenkot, Cornell Tech
Accessible Chatbot for nyc.gov
I designed and prototyped an accessible and assistive chatbot for the NYC government website. Powered by state-of-the-art LLMs, and compatible with assistive technologies (i.e., screen readers, voice input), the chatbot ensured equitable access to information for website visitors with disabilities. [PiTech '23]
Supervised by:
Arthur Jacobs, NYC OTI
Tactile Materials by TVIs
Mid-air haptic interfaces enable rich 3D interactions but are inaccessible to users with disabilities. We designed an interactive simulation of a contactless elevator control panel with mid-air touch feedback and accessibility considerations. Despite being fully contactless, the controls are tactile, they emit braille, and closely resemble the mental model of ordinary elevator buttons. [CHI EA '21]
Collaborators:
Tanay Singhal, University of Waterloo
Hands-Free Virtual Reality
People with neuromotor impairments have difficulty operating virtual reality controllers, especially if they use a wheelchair. They may lower fine motor ability, or have spasticity in their muscles. Working with a physiotherapist, we developed the functionality of using head orientation to onboard the Oculus and navigate menus in virtual reality. [Presentation to XR Access '21]
Supervised by:
Dr. Steven Feiner, Columbia University
Accessible Mid-Air Haptics Design
Mid-air haptic interfaces enable rich 3D interactions but are inaccessible to users with disabilities. We designed an interactive simulation of a contactless elevator control panel with mid-air touch feedback and accessibility considerations. Despite being fully contactless, the controls are tactile, they emit braille, and closely resemble the mental model of ordinary elevator buttons. [CHI EA '21]
Collaborators:
Tanay Singhal, University of Waterloo