AI and digital accessibility
In today’s technology-driven world, Artificial Intelligence (AI) constantly improves and expands into new industries. In fact, people commonly use AI as a tool in their everyday lives for things like checking the grammar of an email, unlocking phones through facial recognition software, or using a navigation app to find the fastest route to your destination. Because of its widespread use, it’s no surprise that AI has expanded to the digital accessibility industry.
But can AI determine entirely on its own, what users with disabilities need and want? We’re skeptical. Here’s why.
How AI can help accessibility
AI can be a useful tool for enhancing digital accessibility on websites, mobile apps, and software. For example, live virtual events can use automated captioning powered by AI as a budget-friendly alternative to hiring a captioner. People who are blind or have low vision can use apps like Microsoft’s Seeing AI to hear a description of what’s around them. When used alongside manual accessibility testing with users living with disabilities, automated AI accessibility auditing even has the potential to be a helpful addition to maintaining an accessible website.
Beyond The “Alt” Tag AI Image Description
When images are published online, they have several pieces of data attached to them: image or file name, size and dimensions, captions, copyright information and, of course, the images themselves. Additionally, there is a piece of data called “alt text,” which is short for alternative text. It’s where the image publisher can add a description of the image. The intent of alt text is to help screen readers describe images to people who are blind. Non-human search engines such as Google use alt text as a way to understand the content of a page.
AI technology will be able not only to read text on images, but also read and interpret the content of images.
Consider what this means for the accessibility of images across the 1.13 billion websites (Forbes 2023) that exist across the globe. One study examined roughly 30 e-commerce sites and found that 55% of them failed to meet accessibility requirements with informational images.*
*Informational images are core to the story; if you remove them, the story or content loses meaning. Decorative images are not considered important to the content; they could be stock images, article headers, or decorative ways to break up large blocks of text.
AI and human testing go hand-in-hand
Like a lot of AI applications, AI will enhance — not replace — human testing, editing, and quality assurance. While many manual procedures are vulnerable to human error, AI technology can be a fail-safe when it comes to routine, assembly-line-type work that can lead to fatigue. Likewise, AI technology is far from perfect. It can help prevent human error, but it will be hard to replace the detail and accuracy of the human eye.
AI and Leaving Out the User
While AI is helpful in some ways, it isn’t a replacement for human feedback. For many of us living with disabilities, AI is a helpful convenience. But it doesn’t mean that we want it to make our choices for us. And in its present form, AI falls far short of being accurate and meeting our needs.
A few examples:
- AI accessibility testing and remediation can be clunky and inaccurate, as well as biased by its limited data (see example below).
- It only tests for a fraction of what’s required for accessibility by the Web Content Accessibility Guidelines (WCAG).
Relying solely on automated AI tools for digital accessibility will give you a website that is only partially accessible. It will also reduce the usability of your project by leaving out the very people it intends to include.