How to Navigate Potential Limitations of AI in Accessibility
In our previous AI article, we highlighted some of the ways AI is enhancing accessibility, both in the classroom and in day-to-day life. Today, we’re looking at some of the potential limitations of AI and how to navigate these potential limits.
Several studies have shown patterns of AI models developing biases against disabled people. Some AI models are trained on data that does not properly represent people with disabilities. Because of these concerns, there are concerns that disabled people may face unintentional discrimination when using programs that utilize AI. To prevent issues like these, the US Access Board has recommended that people with disabilities “consider opting into the minimum amount of information and location sharing needed to access desired services”.
Disability professional Connor Scott-Gardner wrote that in one specific instance where he was using an AI tool to describe the color scheme and content of an image promoting a guide to navigating college for blind students, the software changed the written content to exclude the word “blind”. The updated description read, “My guide for students provides all the information you will need.” This changed the messaging of the image, not denoting that it was specifically for blind students.
When creating written materials with AI, checking for features like proper alternative text, header formatting, and captioning are important to ensure materials being created are accessible and free of errors. To learn more about these functionalities, visit SDSU’s alt text and headers pages.
Others are concerned that overreliance on these tools may backfire and decrease accessibility rather than increase it. Some tools are simply unable to grasp the complexity of the space or properly perform tasks to make materials accessible. In a 2023 study, University of Washington researchers found mixed results when disabled people used AI tools for accessibility assistance. One particular case involved a chatbot that understood that the formatting of tables on a webpage needed to be changed to be accessible, but could not properly perform the task.
Some experts have reiterated the importance of including and engaging people with disabilities in the process of creating materials to ensure they are accessible and properly formatted.
Because of issues like these, the extent to which AI can be relied on for accessibility is limited. Remaining mindful of accessibility best practices and critically evaluating AI outputs and responses can help ensure the accessibility of digital materials. SDSU’s Accessibility Essentials toolkit, Instructional Materials Guide, and Tools and Evaluations guide for accessibility provide tools and examples of ways to ensure accessibility for digital materials.
The impact of AI on people with disabilities is substantial. AI has the potential to address barriers and promote inclusivity, but also poses risks of excluding individuals with disabilities from advancements. It is crucial to remember that AI is not a one-size-fits-all solution, and remain critical of AI outputs to ensure accessibility for all.
