How will AI affect accessibility in tech?

The impacts of AI can be a thorny issue to navigate at the best of times, especially when it comes to generative AI. But when it comes to the effect of AI on the disability community, that picture gets even more murky.

The landscape is a crowded one. AI has long been a part of the digital accessibility story, tools like Dragon Naturally Speaking and Seeing AI – voice-to-text software and a Microsoft product that aims to help those who are blind or have low vision, respectively – are longtime players in the space. At the same time, AI has been heralded as both a visionary joy and a scourge, as it’s marketed across the digital accessibility ecosystem.

Rebecca Rosenberg is the founder and CEO of ReBokeh Technologies, an assistive technology company. She says, in her lived experience both personally and professionally, the potential benefits for disabled people start with how AI is bringing more attention to accessibility in the tech space and beyond.

“I think there are a lot of people who are considering, as their understanding of AI develops, how it could be beneficial to people with disabilities. It’s become something that people want to take a look at, as opposed to kind of an afterthought that it has been in so many other technology and even just infrastructure areas.”

Rosenberg is touching on a grim reality for a lot of accessibility efforts: despite regulatory frameworks like the Americans with Disabilities Act (ADA), the UK’s Equality Act, or even the Web Content Accessibility Guidelines (WCAG), a lot of the conversations about accessibility boil down to the cost of making those changes. Regardless of industry, accessibility is often left until last, if it’s considered at all.

Mike Rispoli, co-founder and CTO of boutique agency Cause of a Kind, says he’s seen that cost-cutting firsthand.

“I’ve found a lot of the reason that the companies tend to tend to skimp on accessibility is that time and money piece. They’re putting so much into… the standard design, if you will, that accessibility is sort of thought of as this afterthought, and it’s always kind of a mad scramble at the end, like, ‘Oh, we didn’t check the screen readers’.”

He’s hopeful that the rise of AI, and its potential to reduce operating costs, can help companies bump accessibility back up the priority list. Rispoli’s experience, that accessibility is always part of the contract but rarely core to the implementation of a design, touches on one of the key ways in which accessibility is talked about in the tech sector, including IT, and compliance.

AI for web accessibility

One of the most obvious ways that ADA compliance has roared onto the scene is via web accessibility overlays, many of which use AI.

Given the litigious nature of the American ecosystem, overlays have become a popular option for companies looking to assuage accessibility concerns. These tools allow the user to do things like change the color contrast of a web page boost text on the screen, or make it easier for a screen reader to detect.

Evangelists of these products tend to market them as one-stop shops, while critics see them as little more than gadgets that take away user agency. Because AI is an imperfect technology, it can also cause issues of its own.

In 2021, Vice reported on an AI accessibility tool that was alleged to actually make pages harder to navigate for users with accessibility needs.

Alexa Heinrich, a disabled social media manager for St Petersburg College in Florida and founder of Accessible Social, says that she’s concerned that AI’s impact on the accessibility sphere could be similar to overlays and their general approach.

“I’m afraid that AI is going to become the next web overlay situation that we have now, where we see these companies installing these codes or these widgets and calling it. ‘Okay, we’re done. This project is complete. Our website is now accessible,’ and that’s not the case.”

This isn’t to say that Heinrich doesn’t use AI tools. She uses generative AI to help her craft better plain language alternatives to what she has written and has collaborated with social media management tool Sprout Social on the company’s AI assistant. She especially sees potential for generative AI tools to help people create better image descriptions and, by extension, alt text.

OpenAI has also demonstrated its multimodal model GPT-4 being used to give improve accessibility for blind and visually impaired users in real time, in partnership with the accessibility app Be My Eyes.

“A lot of people are still kind of finding their feet when it comes to, ‘How do I write good alt text?’ And they get really nervous about it. And now that you have programs like ChatGPT that can describe images, I’ve heard a lot of people using it for that functionality. Which I think is great as long as they’re being intentional about reviewing what AI actually writes. Because, as most of us are aware, AI is capable of mistakes and biases.”

Those biases include algorithmic ableism, where the AI’s biased training data reinforces harmful stereotypes and assumptions about disabled people. Neil Sahota, CEO of ACSI Labs and a United Nations AI advisor, says that those assumptions strike at the core of the tech sector and have to be acknowledged. He would know, having been part of the team that built Watson.

“Unfortunately, technologists, we tend to build to an outcome and we don’t think about the diversity of people’s backgrounds and what they can or cannot do. So, we take a lot of things for granted.”

Sahota gives the example of the rush to default voice assistance, partially due to cost, and how that disadvantages those who are non-speaking. He says that orienting AI towards accessibility means having a deep understanding of how it operates, for better or for worse.

“It’s not software, so don’t think of it as a program. You’ve to think of it as a high-energy intern, right? It’s ready to do whatever you want it to do, but you’ve got to teach it. And so if you want this to be accessible, you want this to be helpful for all cuts of the population, then that’s we what have to teach it.”

So, where do we go from here? Rosenberg says that AI has the potential to increase independence for people with disabilities as long as it is implemented thoughtfully.

“I think AI scales a lot of things that we otherwise would have to rely on other humans for, either by paying them or just by kind of relinquishing some of our control of a situation to ask for help.”

Still, she is mindful that good intentions need to be combined with a skeptical eye in order to further digital accessibility with the use of AI.

“People are so leery, I think, of AI, that we’re thinking about everything.

“We’re really, actually being quite hyper-vigilant about what this means for society, and that has managed to involve people with disabilities in a way that I haven’t seen [in] other technology development.”