The AI-Accessibility outlook for 2025

Despite most of the "What to look out for in Accessibility in 2025" articles focusing on AI (and all sounding oddly similar to each other...), we identified the European Accessibility Act (EAA) as potentially the biggest digital accessibility game changer for 2025. That is not to say AI isn't already making a big impact and set to break more ground in the year ahead.

Last October, I spoke with Joe DiNero from Helen Keller Services for the Blind and UsableNet about how AI is making an impact on accessibility (even before the explosion of Large Language Models and ChatGPT), and the progress and potential of AI for accessibility currently is breathtaking. 

You can read the summary of our discussion about AI and it's potential or watch it in full. The AI-Accessibility Frontier: Progress, Potential, and Pitfalls

In our discussion, we touched on some key points worth considering and keeping in mind when thinking about AI and accessibility.

The current state of AI in digital accessibility is very exciting for people with disability at the moment with a number of impressive products and services being created or enhanced using AI, and made available to users to start using or to trial and give feedback for the final products.

At the same time, one of the biggest limitations of AI currently, is it's accuracy and accountability - and this is as true if not more so when it relates to accessibility. 

Accuracy is extremely important in accessibility. When used as an enhancement and add on, AI can be very helpful, but the technology often falls down for functions where accuracy is imperative. For instance, a tool that can scan and announce what is in front of the camera in a general sense for someone who is blind or has low vision can be a nice addition. 

For example, describing a landscape, city skyline, beach, etc. Even if some of the information isn't 100% accurate, it can still provide some interesting and useful observations. 70% accuracy might be quite good, and the user can take the information into account with that in mind as they continue to navigate the world with their existing, trusted, tools.

If the technology was designed for real time navigation, however (perhaps to replace a guide dog or other assistive technology) that 70% accuracy that was good in the previous use case, could be devastatingly harmful in this case. 

The person may be lead into dangerous situations, walking into objects or onto roads with oncoming traffic. There can be serious implications for this kind of assistive technology, even with a significantly high degree of accuracy. Even 95% and more may still not be high enough for it to be useful.

For digital accessibility, the same types of situations can apply; a user asking AI for an overview of a page layout may gain useful insights before they then navigate the page with their existing technology. On the other hand, there may be significant implications if they relied on AI to interpret an important contract or a financial obligation if it is not accurate.

It is essential to continue to provide accessible digital platforms, products and services to ensure all people have the ability to access them in a robust and accurate way, and are not forced to rely on a 3rd party interpretation of content, whether that is through another person (whom may not always or ever be available), or through an AI agent; particularly for the time being as accuracy is such an issue.

As for how AI can help with the implementation of accessibility in these platforms, products and services, so far, AI has struggled to provide consistently accurate information to designers, developers and content authors. 

While it may be able to give good general tips and high level advice, digital accessibility has a very broad set of potential issues and the resolution to the same issue can vary depending on the context of the content or feature, so it is very difficult for AI to be able to provide accurate recommendations and this shows in the results so far.

Ultimately, we cannot rely on AI to replace assistive technology, nor to build accessibility into the digital world for us, without ensuring we know and understand how to do both ourselves. The current state of AI makes it a powerful tool when used as a tool we know how to use, and a dangerous one when we use it without knowing if it has done the job it was supposed to do.

If used right, AI is going to be an incredible enhancement to the lives of people with disability all over the world. Let's help to make that happen by playing our part and laying a solid accessible digital foundation to build on for AI to enhance and take to a whole new level.

A reminder you can watch our webinar in full The AI-Accessibility Frontier: Progress, Potential, and Pitfalls, and contact us or look out for our upcoming public training sessions if you're interested in learning more.