Skip to main content

University of Michigan School of Information

Menu

Media Center

UMSI welcomes media inquiries

Disability centered conversations are often overlooked in the field of artificial intelligence

UMSI Research. Misfitting with AI: How Blind People Verify and Contest AI Errors. Rahaf Alharbi,. PhD Candidate. Sarita Shoenebeck, Professor. Robin Brewer, Assistant Professor.

Wednesday, 10/30/2024

By Noor Hindi

Blind people use artificial intelligence-enabled visual assistance technologies (AI VAT) to gain visual access in their everyday lives. Though discussions around AI often focus on its errors and implicit biases, little research has been done on how blind people navigate errors in AI. 

New work by UMSI PhD candidate Rahaf Alharbi, professor Sarita Schoenebeck and assistant professor Robin Brewer looks into the errors that blind people experience when working with AI

“In this study, we were motivated to understand how blind people verify AI errors,” Alharbi says. “There's a focus on AI for visual information access for blind people and a huge part of our field is dedicated to understanding and building technologies for visual access. But one of the missing dimensions was around how blind people deal with errors that can lead to harmful consequences.” 

These errors, Alharbi says, include processing errors and cultural biases that can negatively impact the way a blind person receives critical information and the accuracy of the information they receive. For example, AI VAT technologies, when navigating complex documents, can mix up dates and times. 

Additionally, cross-cultural bias within the AI system can impact where and when a blind person can use this technology. 

“I think within accessibility, there's a tendency to think of blind people as a homogeneous community,” she says. “In reality, their experiences with access are often informed by other identity factors like culture. We found that a lot of these documents don't support languages that are more prevalent, like Arabic or Khmer.” Additionally, “our study shows that blind people are already doing much of this work around error detection,” Brewer noted. 

One participant of the study highlighted how AI VAT, when used in grocery stores, does not recognize products that are less common in the United States. Outside of identifying and verifying errors, the paper discusses design implications that include blind people in the development and maintenance of AI technologies.  

Alharbi, a fifth-year PhD candidate, has long published work that focuses on disability-centric responsible AI practices. Alharbi’s advisers are Schoenebeck and Brewer. Her expected graduation date is spring 2025. 

“My engineering background pushed me toward taking a disability justice and disability studies approach to technology,” she says. “I want to critique the prevalent savior assumptions around disabled people and advocate against this grain ”

Misfitting With AI: How Blind People Verify and Contest AI Errors” is available on arXiv and is authored by Rahaf Alharbi, Pa Lor, Jaylin Herskovitz, Sarita Schoenebeck and Robin Brewer. 

RELATED

Learn more about Rahaf AlharbiSarita Schoenebeck and Robin Brewer by visiting their UMSI faculty profiles. 

Keep up with research from UMSI experts by subscribing to our free research roundup newsletter!