Anat Caspi speaks at White House panel on AI in Transportation

June 17, 2024

CREATE associate director Anat Caspi spoke on ‘Mapping, Visualizing, and Building the Future’ as part of a White House panel on AI in Transportation. Focused on developing a national transportation infrastructure observatory and an accompanying application ecosystem, the panel gathered innovators in transportation, aiming to align end users, researchers, entrepreneurs, and federal agencies while addressing privacy, cybersecurity, and policy challenges. Dr. Caspi highlighted the Taskar Center’s pioneering work on AI and transportation, particularly through the Transportation Data Equity Initiative, OpenSidewalks, and AccessMap.

Congrats to CREATE’s Graduating Ph.D. Students 2024!

May 30, 2024

Four of CREATE’s influential and productive doctoral students are graduating this spring. Please join us in congratulating Avery Mack, Emma McDonnell, Venkatesh Potluri, and Ather Sharif and wishing them well.

Avery Mack, a white, femme-presenting person with curly light brown hair shaved close on one side wearing a green blazer and grey top

Avery Mack will receive their Ph.D. from the Paul G. Allen School of Computer Science & Engineering. Advised by CREATE Director Jennifer Mankoff, their research focuses on representation of people with disabilities in digital technologies like avatars and generative AI tools. They recently have investigated how technology can support people with fluctuating access needs, like neurodiverse people and people with chronic or mental health conditions.

Mack has been an invaluable resource at CREATE, co-leading graduate seminars, presenting workshops on accessibility, and contributing to CREATE’s accessibility research. “CREATE has been a great place to meet other accessibility researchers and get in contact with disabled people in our community,” says Mack. “As someone who tries to align my researchers with community needs and desires, this connection to the Seattle disability community is invaluable.”

Mack, whose thesis is titled, Dissertation Title: Understanding, Designing, and Building Adaptable Technology for Fluctuating Accessibility Needs in Group Settings, is on the job market, interested in a research scientist position in industry.

Emma McDonnell, a white woman in her 20s with short red hair, freckles, and a warm smile. In the background: a lush landscape and the Colosseum.

Earning her Ph.D. from Human Centered Design and Engineering, Emma McDonnell is advised by CREATE associate director Leah Findlater. McDonnell’s research blends computer science, design, and disability studies to explore ways that technology can be designed to align with disability politics and social worlds.

Her dissertation research explores how communication technology, specifically captioning, could be redesigned to encourage mixed ability groups to take a collective approach to accessibility. Along with CREATE associate directors Leah Findlater and  Jon Froehlich, McDonnell studied captioning practices on TikTok and offered steps toward a standard for user-generated captioning. With fellow Ph.D. student, McDonnell presented a workshop on accessible presentations for CREATE’s GAAD Day 2024, contextualizing the importance of accessibility within the longer history of disability discrimination and activism.

Looking ahead

McDonnell is interested in postdoctoral opportunities to continue exploring new possibilities for technology design anchored in critical disability perspectives. 

Venkatesh Potluri leans toward the camera smiling with eyes cast downward

Advised by CREATE Director Jennifer Mankoff in the Paul G. Allen School of Computer Science & Engineering, Venkatesh Potluri’s research examines accessibility barriers experienced by blind or visually impaired (BVI) developers participating in professional programming domains such as user interface design, data science, and physical computing. His work contributes real-world systems to improve developer tools and new interaction techniques to address these access barriers. His thesis is titled, A Paradigm Shift in Nonvisual Programming.

While at the UW, Potluri has been selected as an Apple Scholar and a Google Lime Scholar, and contributed to the Accessibility Guide for Data Science and STEM Classes. He presented a paper on a large-scale analysis of the accessibility of Jupyter notebooks, a new tool that enables blind and visually impaired people to create their own data visualizations to explore streaming data.

Asked about his experience with CREATE, Potluri responded, "Since CREATE's founding, I've been thrilled by its mission to take a holistic approach to accessibility with disabled experts and stakeholders—from education to research to translation. I'm grateful to have been part of this beacon of high-quality research informed by a deep understanding of disability. I aspire to carry the torch forward and help make the world accessible!"

Future plans

Potluri will join the University of Michigan as an assistant professor in the Information School in Fall 2024.

 

Headshot of Ather Sharif outside on a sunny balcony with blue sky behind himGraduating with a Ph.D. from the Paul G. Allen School of Computer Science & Engineering, Ather Sharif is co-advised by CREATE faculty Katharina Reinecke and CREATE associate director Jacob O. Wobbrock. Sharif’s research on Human-Computer Interaction (HCI) focuses on making online data visualizations accessible to screen-reader users. He pioneered the first-of-its-kind system, VoxLens, that utilizes voice assistants for screen-reader users to extract information from online data visualizations. He also created UnlockedMaps, an open-data map that assists users with mobility disabilities to make informed decisions regarding their commute.

Sharif has garnered many awards while at the UW, including:

Sharif credits CREATE leaders, who include his advisors as well as Richard Ladner, Jennifer Mankoff, and Anat Caspi, to name a few, “who are not only prominent allies for disabled people but are always willing to advise and guide students to be the best researchers they can be.”

“I cannot begin to express how incredible it is to have CREATE as part of our ecosystem,” says Ather. “It advances the state of accessible technologies for people with disabilities through cutting-edge research. Personally, as someone with a disability, it means the world to me. As a researcher, CREATE has funded almost all of my research work at UW.”

After graduation, Sharif will be traveling on a 2024 UW Bonderman Fellowship. He plans to visit Brazil, Argentina, Peru, Costa Rica, Japan, Vietnam, South Korea, and Thailand to learn about disability rights history and distinct physical infrastructure for wheelchair users and enhance his perspectives, challenge his viewpoints, and identify real-life barriers disabled people face. 

With too many accomplishments amongst them to list here, these almost-minted Ph.D.s collaborated on projects that have contributed to CREATE’s growth and success. In addition to mentoring undergraduate students, publishing and presenting papers, and working in labs and with researchers, here are a few of the ways Sharif, Potluri, McDonnell and Mack have worked together:

  • Avery Mack and Venkatesh Potluri contributed to the Accessibility Guide for Data Science and STEM Classes, available via the CREATE website, A11y in Action resource link. They, with the other lead contributors, received the 2024 UW Digital Accessibility Team Award as part of UW Accessible Technology’s Global Accessibility Awareness Day celebration. 
  • Potluri and Mack also co-led 5 CREATE Accessibility Seminars to discuss relevant reading and share accessibility research.
  • Mack and Ather Sharif collaborated with Lucille Njoo to dispel common myths about students with disabilities in an article in the Winter 2024 Allen School DEIA newsletter.
  • McDonnell and Mack presented an accessible presentations workshop as part of UW’s 2024 Global Accessibility Awareness Day celebration.

CREATE AI+Accessibility Hackfest – Winter ’24

March 6, 2024 – post-event update

The event featured invited speakers Heather Nolis, Ian Stenseng, and Shaun Kane and exciting workshops on building custom GPT and creating accessible Jupyter notebooks. See the full lineup of brainstorming, hacking, and presentation sessions.

The 3-day hackfest attendees included those with no experience in coding or hacking, others with advanced experience in generative AI and building software or tools, and, at the center, attendees with lived experiences of disabilities who contributed their experiences and expertise to invent an accessible AI-enabled future.

Prizes awarded

While appreciation and congratulations go to all participants, these projects were awarded prizes:

First place: LookLoud.ai

Nishit Bhasin and Lakshya Garg

LookLoud.ai is voice-activated assistance technology, powered by GPT-4 Vision, and designed to make e-commerce accessible to everyone. Users can navigate, select, and buy products using simple voice commands. 

Second prize: AI Posture Monitor & Intervention Alerts for Home Health

Max Smoot, Lige Yang, and Richard Li

AI Posture Monitor & Intervention Alerts for Home Health monitors someone’s seated position to identify when they are in an at-risk posture and subsequently alerts a caretaker with recommended corrections.

Third prize: Formflow Ai

Abdul Hussein, Abreham Tegenge, and Aelaph Elias

Formflow.ai reads PDFs, mail, and forms and gives an easy-to-read summarization, with the goal of helping people read and understand documents and forms. 

Fourth place: Clearview Assist

Dhruv Khanna, Ritika Rajpal, Minal Naik, and Menita Agarwal

ClearView Assist is a Chrome extension designed to assist internet users with low vision or blindness by simplifying cluttered web pages based on user tasks and allowing individuals to interact with digital content via voice commands to articulate their browsing objectives.

Fifth place: Student Success Portal

Mia Vong, Cameron Jacob Miller, Keyvyn Rogers, and Jerid Stevenot

Student Success Portal provides AI-powered assistance for challenges in supporting K-12 students with Individualized Education Programs (IEPs).

Sessions, workshops and hack time

  • Introductory session about the potential of AI for accessibility (also on Zoom)
  • Invited speaker Ian Stenseng, Director of Innovation & Accessibility at The Lighthouse for the Blind, Inc. (also on Zoom)
  • Brainstorming project ideas
    • Learn from community members with lived experiences of disabilities to make sure your hack is solving a real accessibility need.
  • Lunch (provided) and conversation, mentoring, team forming, idea hatching
  • Invited speaker Heather Nolis, Principal Machine Learning Engineer of the Digital AI Team and Chair of the Accessibility Community at T-Mobile (ACT) at T-Mobile (also on Zoom)
  • Optional Workshops and hack time
  • Hack time
  • Pizza dinner and opportunities to get feedback from mentors

Saturday

  • Work time
  • Lunch (provided) and opportunity to present for feedback from mentors
  • Presentation of judging rubric
  • Invited speaker, Shaun Kane, Researcher at Google AI and Director of the Superhuman Computing Lab at University of Colorado Boulder (also on Zoom)
  • Hack time

Sunday

  • Optional hack time
  • How to present accessibly & sample pitch presentation (also on Zoom)

Monday

  • Presentations to judges (also on Zoom)
  • Judges deliberation
  • Announcements, prizes, and closing keynote (also on Zoom)

Judges

Speakers

Workshops

Brainstorming ideas

Relevant topics will be driven by community needs to increase access to technology, and to the world through technology. These topics could include, for example:

  • AI’s use for generating plain language summaries of rights
  • Accessibility of AI tools and interfaces
  • Using AI to increase the accessibility of written and visual content
  • Robotic control for access
  • Tools for designing accessible physical objects
  • Using AI to get feedback on the accessibility of things you’re making
  • AI for embodied agent interactions
  • AI applications for health and wellbeing
  • Modalities for human/generative AI interactions such as voice or touch
  • Guidelines or ideas around agents that that may be used for accessibility
  • What disability simulation might look like in the age of AI agents
  • Best practices and pitfalls

Winter 2023 CREATE Research Showcase

December 12, 2023

Students from CSE 493 and additional CREATE researchers shared their work at the December 2023 CREATE Research Showcase. The event was well attended by CREATE students, faculty, and community partners. Projects included, for example: an analysis of the accessibility of transit stations and a tool to aid navigation within transit stations; an app to help colorblind people of color pick makeup; and consider the accessibility of generative AI while also considering ableist implications of limited training data.

CSE 493 student projects

In it’s first offering Autumn quarter 2023, CSE’s undergraduate Accessibility class has been focusing on the importance of centering first-person accounts in disability-focused technology work. Students worked this quarter on assignments ranging from accessibility assessments of county voting systems to disability justice analysis to open-ended final projects.

Alti Discord Bot »

Keejay Kim, Ben Kosa, Lucas Lee, Ashley Mochizuki

Alti is a Discord bot that automatically generates alt text for any image that gets uploaded onto Discord. Once you add Alti to your Discord server, Alti will automatically generate alt text for the image using artificial intelligence (AI).

Enhancing Self-Checkout Accessibility at QFC »

Abosh Upadhyaya, Ananya Ganapathi, Suhani Arora

Makes self-checkout more accessible to visually impaired individuals

Complexion Cupid: Color Matching Foundation Program »

Ruth Aramde, Nancy Jimenez-Garcia, Catalina Martinez, Nora Medina

Allows individuals with color blindness to upload an image of their skin, and provides a makeup foundation match. Additionally, individuals can upload existing swatches and will be provided with filtered photos that better show the matching accuracy.

Twitter Content Warnings »

Stefan D’Souza, Aditya Nair

A chrome extension meant to be used in conjunction with twitter.com in order to help people with PTSD

Lettuce Eat! A Map App for Accessibly Dietary Restrictions »

Arianna Montoya, Anusha Gani, Claris Winston, Joo Kim

Parses menus on restaurants’ websites to provide information on restaurants’ dietary restrictions to support individuals with specific dietary requirements, such as vegan vegetarian, and those with Celiac disease.

Form-igate »

Sam Assefa

A chrome extension that allows users with motor impairments to interact with google forms using voice commands, enhancing accessibility.

Lite Lingo: Plain Text Translator »

Ryan Le, Michelle Vu, Chairnet Muche, Angelo Dauz

A plain text translator to help individuals with learning disabilities

Matplotalt: Alt text for matplotlib figures »

Kai Nailund

[No abstract]

PadMap: Accessible Map for Menstrual Products »

Kirsten Graham, Maitri Dedhia, Sandy Cheng, Aaminah Alam

Our goal is to ensure that anywhere on campus, people can search up the closest free menstrual products to them and get there in an accessible way.

SCRIBE: Crowdsourcing Scientific Alt Text »

Sanjana Chintalapati, Sanjana Sridhar, Zage Strassberg-Phillips

A prototype plugin for arXiv that adds alt text to requested papers via crowdwork.

PalPalette »

Pu Thavikulwat, Masaru Chida, Srushti Adesara, Angela Lee

A web app that helps combat loneliness and isolation for young adults with disabilities

SpeechIT »

Pranati Dani, Manasa Lingireddy, Aryan Mahindra

A presentation speech checker to ensure a user’s verbal speech during presentation is accessible and understandable for everyone.

Enhancing Accessibility in SVG Design: A Fabric.js Solution »

Julia Tawfik, Kenneth Ton, Balbir Singh, Aaron Brown

A Laser Cutter Generator’ interface which displays a form to select shapes and set dimensions for SVG creation.

CREATE student and faculty projects

Designing and Implementing Social Stories in Technology: Enhancing Collaboration for Parents and Children with Neurodiverse Needs

Elizabeth Castillo, Annuska Zolyomi, Ting Zhou

Our research project, conducted through interviews in Panama, focuses on the user-centered design of technology to enhance autism social stories for children with neurodiverse needs. We aim to improve collaboration between parents, therapists, and children by creating a platform for creating, sharing, and tracking the usage of social stories. While our initial research was conducted in Panama, we are eager to collaborate with individuals from Japan and other parts of the world where we have connections, to further advance our work in supporting neurodiversity.

An Autoethnographic Case Study of Generative Artificial Intelligence’s Utility for Accessibility

Kate S Glazko, Momona Yamagami, Aashaka Desai, Kelly Avery Mack, Venkatesh Potluri, Xuhai Xu, Jennifer Mankoff

With the recent rapid rise in Generative Artificial Intelligence (GAI) tools, it is imperative that we understand their impact on people with disabilities, both positive and negative. However, although we know that AI in general poses both risks and opportunities for people with disabilities, little is known specifically about GAI in particular. To address this, we conducted a three-month autoethnography of our use of GAI to meet personal and professional needs as a team of researchers with and without disabilities. Our findings demonstrate a wide variety of potential accessibility-related uses for GAI while also highlighting concerns around verifiability, training data, ableism, and false promises.

Machine Learning for Quantifying Rehabilitation Responses in Children with Cerebral Palsy

Charlotte D. Caskey, Siddhi R. Shrivastav, Alyssa M. Spomer, Kristie F. Bjornson, Desiree Roge, Chet T. Moritz, Katherine M. Steele

Increases in step length and decreases in step width are often a rehabilitation goal for children with cerebral palsy (CP) participating in long-term treadmill training. But it can be challenging to quantify the non-linear, highly variable, and interactive response to treadmill training when parameters such as treadmill speed increase over time. Here we use a machine learning method, Bayesian Additive Regression Trees, to show that there is a direct effect of short-burst interval locomotor treadmill training on increasing step length and modulating step width for four children with CP, even after controlling for cofounding parameters of speed, treadmill incline, and time within session.

Spinal Stimulation Improves Spasticity and Motor Control in Children with Cerebral Palsy

Victoria M. Landrum, Charlotte D. Caskey, Siddhi R. Shrivastav, Kristie F. Bjornson, Desiree Roge, Chet T. Moritz, Katherine M. Steele

Cerebral palsy (CP) is caused by a brain injury around the time of birth that leads to less refined motor control and causes spasticity, a velocity dependent stretch reflex that can make it harder to bend and move joints, and thus impairs walking function. Many surgical interventions that target spasticity often lead to negative impacts on walking function and motor control, but transcutaneous spinal cord stimulation (tSCS), a novel, non-invasive intervention, may amplify the neurological response to traditional rehabilitation methods. Results from a 4-subject pilot study indicate that long-term usage of tSCS with treadmill training led to improvements in spasticity and motor control, indicating better walking function.

Adaptive Switch Kit

Kate Bokowy, Mia Hoffman, Heather A. Feldner, Katherine M. Steele

We are developing a switch kit for clinicians and parents to build customizable switches for children with disabilities. These switches are used to help children play with computer games and adapted toys as an early intervention therapy.

Developing a Sidewalk Improvement Cost Function

Alex Kirchmeier, Cole Anderson, Anat Caspi

In this ongoing project, I am developing a Python script that uses a sidewalk issues dataset to determine the cost of improving Seattle’s sidewalks. My intention is to create a customizable function that will help users predict the costs associated with making sidewalks more accessible.

Exploring the Benefits of a Dynamic Harness System Using Partial Body Weight Support on Gross Motor Development for Infants with Down Syndrome

Reham Abuatiq, PT, MSc1; Mia Hoffman, ME, BSc2; Alyssa Fiss, PT, PhD3; Julia Looper, PT, PhD4; & Heather Feldner, PT, PhD, PCS1,5,6

We explored the benefits of a Dynamic Harness System Using Partial Body Weight Support (PBWS) within an enriched play environment on Gross Motor Development for Infants with Down Syndrome using randomized cross over study design. We found that the effectiveness of the PBWS harness system on gross motor development was clearly evident. The overall intervention positively affected activity levels, however, the direct impact of the harness remains unclear.

StreetComplete for Better Pedestrian Mapping

Sabrina Fang, Kohei Matsushima

StreetComplete is a gamified, structured, and user-friendly mobile application for users to improve OpenStreetMap data by completing pilot quests. OpenStreetMap is an open-source, editable world map created and maintained by a community of volunteers. The goal of this research project is to design pilot quests in StreetComplete to accurately collect information about “accessibility features,” such as sidewalk width and the quality of lighting, to improve accessibility for pedestrian mapping.

Transit Stations Are So Confusing!

Jackie Chen, Milena Johnson, Haochen Miao, and Raina Scherer

We are collecting data on the wayfinding nodes in four different Sound Transit light rail stations, and interpreting them through the GTFS-pathways schema. In the future, we plan on visualizing this information through AccessMaps such that it can be referenced by all users.

Optimizing Seattle Curbside Disability Parking Spots

Wendy Bu, Cole Anderson, Anat Caspi

The project is born out of a commitment to enhance the quality of life for individuals with disabilities in the city of Seattle. The primary objective is to systematically analyze and improve the allocation and management of curbside parking spaces designated for disabled individuals. By improving accessibility for individuals with disabilities, the project contributes to fostering a more equitable and welcoming urban environment.

Developing Accessible Tele-Operation Interfaces for Assistive Robots with Occupational Therapists

Vinitha Ranganeni, Maya Cakmak

The research is motivated by the potential of using tele-operation interfaces with assistive robots, such as the Stretch RE2, to enhance the independence of individuals with motor limitations in completing activities of daily living (ADLs). We explored the impact of customization of tele-operation interfaces and a deployed the Stretch RE2 in a home for several weeks facilitated by an occupational therapist and enabled a user with quadriplegia to perform daily activities more independently. Ultimately, this work aims to empower users and occupational therapists in optimizing assistive robots for individual needs.

HuskyADAPT: Accessible Design and Play Technology

HuskyADAPT Student Organization

HuskyADAPT is a multidisciplinary community at the University of Washington that supports the development of accessible design and play technology. Our community aims to initiate conversations regarding accessibility and ignite change through engineering design. It is our hope that we can help train the next generation of inclusively minded engineers, clinicians, and educators to help make the world a more equitable place.

A11yBoard for Google Slides: Developing and Deploying a Real-World Solution for Accessible Slide Reading and Authoring for Blind Users

Zhuohao (Jerry) Zhang, Gene S-H Kim, Jacob O. Wobbrock

Presentation software is largely inaccessible to blind users due to the limitations of screen readers with 2-D artboards. This study introduces an advanced version of A11yBoard, initially developed by Zhang & Wobbrock (CHI2023), which now integrates with Google Slides and addresses real-world challenges. The enhanced A11yBoard, developed through participatory design including a blind co-author, demonstrates through case studies that blind users can independently read and create slides, leading to design guidelines for accessible digital content creation tools.

“He could go wherever he wanted”: Driving Proficiency, Developmental Change, and Caregiver Perceptions following Powered Mobility Training for Children 1-3 Years with Disabilities

Heather A. Feldner, PT, MPT, PhD; Anna Fragomeni, PT; Mia Hoffman, MS; Kim Ingraham, PhD; Liesbeth Gijbels, PhC; Kiana Keithley, SPT; Patricia K. Kuhl, PhD; Audrey Lynn, SPT; Andrew Meltzoff, PhD; Nicole Zaino, PhD; Katherine M. Steele, PhD

The objective of this study was to investigate how a powered mobility intervention for young children (ages 1-3years) with disabilities impacted: 1) Driving proficiency over time; 2) Global developmental outcomes; 3) Learning tool use (i.e., joystick activation); and 4) Caregiver perceptions about powered mobility devices and their child’s capabilities.

Access to Frequent Transit in Seattle

Darsh Iyer, Sanat Misra, Angie Niu, Dr. Anat Caspi, Cole Anderson

The research project in Seattle focuses on analyzing access to public transit, particularly frequent transit stops, by considering factors like median household income. We scripted in QGIS, analyzed walksheds, and examined demographic data surrounding Seattle’s frequent transit stops to understand the equity of transit access in different neighborhoods. Our goal was to visualize and analyze the data to gain insights into the relationship between transit access, median household income, and other demographic factors in Seattle.

Health Service Accessibility

Seanna Qin, Keona Tang, Anat Caspi, Cole Anderson

Our research aims to discover any correlation between median household income and driving duration from census tracts to the nearest urgent care location in the Bellevue and Seattle region

Conveying Uncertainty in Data Visualizations to Screen-Reader Users Through Non-Visual Means

Ather Sharif, Ruican Zhong, and Yadi Wang

Incorporating uncertainty in data visualizations is critical for users to interpret and reliably draw informed conclusions from the underlying data. However, visualization creators conventionally convey the information regarding uncertainty in data visualizations using visual techniques (e.g., error bars), which disenfranchises screen-reader users, who may be blind or have low vision. In this preliminary exploration, we investigated ways to convey uncertainty in data visualizations to screen-reader users.

UW News: Can AI help boost accessibility? CREATE researchers tested it for themselves

November 2, 2023 | UW News

Generative artificial intelligence tools like ChatGPT, an AI-powered language tool, and Midjourney, an AI-powered image generator, can potentially assist people with various disabilities. They could summarize content, compose messages, or describe images. Yet they also regularly spout inaccuracies and fail at basic reasoningperpetuating ableist biases.

This year, seven CREATE researchers conducted a three-month autoethnographic study — drawing on their own experiences as people with and without disabilities — to test AI tools’ utility for accessibility. Though researchers found cases in which the tools were helpful, they also found significant problems with AI tools in most use cases, whether they were generating images, writing Slack messages, summarizing writing or trying to improve the accessibility of documents.

Four AI-generated images show different interpretations of a doll-sized “crocheted lavender husky wearing ski goggles,” including two pictured outdoors and one against a white background.

The team presented its findings Oct. 22 at the ASSETS 2023 conference in New York.

“When technology changes rapidly, there’s always a risk that disabled people get left behind,” said senior author Jennifer Mankoff, CREATE’s director and a professor in the Paul G. Allen School of Computer Science & Engineering. “I’m a really strong believer in the value of first-person accounts to help us understand things. Because our group had a large number of folks who could experience AI as disabled people and see what worked and what didn’t, we thought we had a unique opportunity to tell a story and learn about this.”

The group presented its research in seven vignettes, often amalgamating experiences into single accounts to preserve anonymity. For instance, in the first account, “Mia,” who has intermittent brain fog, deployed ChatPDF.com, which summarizes PDFs, to help with work. While the tool was occasionally accurate, it often gave “completely incorrect answers.” In one case, the tool was both inaccurate and ableist, changing a paper’s argument to sound like researchers should talk to caregivers instead of to chronically ill people. “Mia” was able to catch this, since the researcher knew the paper well, but Mankoff said such subtle errors are some of the “most insidious” problems with using AI, since they can easily go unnoticed.

Yet in the same vignette, “Mia” used chatbots to create and format references for a paper they were working on while experiencing brain fog. The AI models still made mistakes, but the technology proved useful in this case.

“When technology changes rapidly, there’s always a risk that disabled people get left behind.”

Jennifer Mankoff, CREATE Director, professor in the Allen School

Mankoff, who’s spoken publicly about having Lyme disease, contributed to this account. “Using AI for this task still required work, but it lessened the cognitive load. By switching from a ‘generation’ task to a ‘verification’ task, I was able to avoid some of the accessibility issues I was facing,” Mankoff said.

The results of the other tests researchers selected were equally mixed:

  • One author, who is autistic, found AI helped to write Slack messages at work without spending too much time troubling over the wording. Peers found the messages “robotic,” yet the tool still made the author feel more confident in these interactions.
  • Three authors tried using AI tools to increase the accessibility of content such as tables for a research paper or a slideshow for a class. The AI programs were able to state accessibility rules but couldn’t apply them consistently when creating content.
  • Image-generating AI tools helped an author with aphantasia (an inability to visualize) interpret imagery from books. Yet when they used the AI tool to create an illustration of “people with a variety of disabilities looking happy but not at a party,” the program could conjure only fraught images of people at a party that included ableist incongruities, such as a disembodied hand resting on a disembodied prosthetic leg.

“I was surprised at just how dramatically the results and outcomes varied, depending on the task,” said lead author Kate Glazko, a UW doctoral student in the Allen School. “”n some cases, such as creating a picture of people with disabilities looking happy, even with specific prompting — can you make it this way? — the results didn’t achieve what the authors wanted.”

The researchers note that more work is needed to develop solutions to problems the study revealed. One particularly complex problem involves developing new ways for people with disabilities to validate the products of AI tools, because in many cases when AI is used for accessibility, either the source document or the AI-generated result is inaccessible. This happened in the ableist summary ChatPDF gave “Mia” and when “Jay,” who is legally blind, used an AI tool to generate code for a data visualization. He could not verify the result himself, but a colleague said it “didn’t make any sense at all.”  The frequency of AI-caused errors, Mankoff said, “makes research into accessible validation especially important.”

Mankoff also plans to research ways to document the kinds of ableism and inaccessibility present in AI-generated content, as well as investigate problems in other areas, such as AI-written code.

“Whenever software engineering practices change, there is a risk that apps and websites become less accessible if good defaults are not in place,” Glazko said. “For example, if AI-generated code were accessible by default, this could help developers to learn about and improve the accessibility of their apps and websites.”

Co-authors on this paper are Momona Yamagami, who completed this research as a UW postdoctoral scholar in the Allen School and is now at Rice University; Aashaka DesaiKelly Avery Mack and Venkatesh Potluri, all UW doctoral students in the Allen School; and Xuhai Xu, who completed this work as a UW doctoral student in the Information School and is now at the Massachusetts Institute of Technology. This research was funded by Meta, Center for Research and Education on Accessible Technology and Experiences (CREATE), Google, an NIDILRR ARRT grant and the National Science Foundation.


For more information, contact Glazko at glazko@cs.washington.edu and Mankoff at jmankoff@cs.washington.edu.


This article was adapted from the UW News article by Stefan Milne.