Putting AI to Work in Accessibility Technology

SECTION 1

My journey with assistive technology: from ear trumpets to AI

Born profoundly deaf, I've always worn hearing aids. Over the years, I've witnessed a complete transformation of this technology. What started as old-fashioned ear trumpets turned into body-worn devices, then behind-the-ear and in-ear devices, before entering the digital era with rechargeable batteries, cochlear implants, and wireless devices with AI features. They've gone from bulky, cumbersome objects to tiny, smart gadgets. How I feel about wearing these devices has also changed: I used to feel self-conscious about them; today I feel far more confident.

At work and at home, I live and breathe captions, and I've closely followed the evolution of that technology. From basic, delayed text to today's near real-time, AI-assisted accuracy, the advances have been remarkable. The recent ability to translate captions into more than 140 languages adds global reach on top. Captions let me fully take part in meetings and engage confidently with colleagues and clients on calls.

But where do we go from here? The possibilities AI opens up for the future are genuinely exciting. And as an expert at Adservio, I'm also aware that technology never springs into being fully formed. The intersection between AI and accessibility will no doubt bring considerable advances, but only if it's handled thoughtfully and responsibly. That's what this article is about.

SECTION 2

AI already at work in assistive technology

Beyond my own experience, AI-based technology is already making a concrete difference in millions of people's lives, across four families of needs.

### Hearing and captioning enhanced by AI

The latest hearing aids embed noise-suppression and speech-separation models that isolate a single voice in a noisy environment, a task long out of reach for classic signal processing. Near real-time captioning, now translated on the fly into more than 140 languages, has become mainstream in video-conferencing tools as well as physical environments.

### Vision and remote visual assistance

Apps like Seeing AI use computer-vision models to recognize objects, text, or faces and describe them to the user in real time through a smartphone camera. Be My Eyes, for its part, connects people who need sighted assistance with volunteers or, increasingly, with a multimodal AI assistant able to describe a scene even before a human steps in. Consumer smart glasses are now generalizing this kind of embedded visual assistance.

### Mobility: smart prosthetics and wheelchairs

Artificial limbs now "learn" a user's individual movement patterns and adapt to provide smoother, more natural control, improving mobility and functional independence. AI-powered wheelchairs, for their part, autonomously navigate around obstacles and respond to voice commands, offering greater independence in daily life.

### Cognition: AI assistants for everyday life

AI also helps with the daily cognitive load, with assistants capable of managing tasks and schedules, anticipating contextual reminders, or breaking down complex instructions into simpler steps, valuable support for people with memory disorders or executive-function challenges.

@cite:comment-l-ia-rend-accessibles-les-services-gouvernementaux

SECTION 3

AI as an accelerator for accessible design and development

Beyond being a feature embedded in products, AI has the potential to significantly accelerate how accessible technologies are designed and built in the first place.

### Automated accessibility testing

AI-assisted accessibility scanners can automatically audit applications and websites against WCAG 2.2 AA, now the practical baseline beyond WCAG 2.1,as well as the early building blocks of WCAG 3.0, still being stabilized. These tools streamline the testing process but remain unable, on their own, to cover every criterion: screen-reader compatibility, cognitive usability, and complex interactive components still require manual evaluation by trained professionals.

### User-needs analysis and personalization

AI can analyze user interactions and feedback to identify unmet needs, recommend relevant accessibility features, and personalize an interface for widely diverse profiles, rather than aiming for an average user who doesn't actually exist.

### Generating standards-compliant code

On the development side, AI works at an impressive pace: give an AI coding assistant a set of accessibility guidelines, and it can automatically generate components that meet those standards from the outset, rather than fixing them afterward. All this speed deserves praise, but AI has its limits: it remains just another tool in the developer's toolbox, and its impact depends entirely on how it's used.

SECTION 4

The limits of AI in accessibility

It's a myth to believe you can rely on AI to cover all your accessibility needs. It simply cannot guarantee full compliance on its own.

### Algorithmic bias inherited from training data

Despite their potential to reduce certain human biases, algorithms remain prone to inheriting biases present in their training data, which can make them discriminatory toward certain user profiles, particularly those underrepresented in existing datasets.

### No guarantee of full compliance

Available estimates on AI-assisted audit tools converge on a similar finding: they typically detect between 30 and 40 percent of an interface's real accessibility issues. The rest, keyboard flows, screen-reader navigation consistency, the cognitive load of a complex form, requires human judgment that automation doesn't replace.

@cite:les-biais-de-genre-en-2025

SECTION 5

Building responsible governance for AI in accessibility

Being aware of these limits is the first step toward mitigating them. The next step is putting rigorous processes in place to avoid the most common pitfalls.

### Inclusive data collection and bias-aware design

This starts with ensuring genuinely inclusive data collection and designing algorithms with the explicit goal of avoiding bias, rather than hoping it will disappear on its own as data volumes grow.

### Transparency and explainability

Building transparency and explainability into AI systems ensures accountability and trust: a user or an auditor should be able to understand why a system recommended one adaptation over another.

@cite:testing-des-systemes-ia-construire-la-confiance-dans-le-non

### Co-design with people with disabilities

Collaborating with people with disabilities from the design, development, and testing phases onward remains the single most decisive condition for creating accessibility solutions that are genuinely effective. This human-centered approach must remain the foundation for everything else: AI should support and empower human designers and developers, never replace them entirely. AI isn't infallible, and people naturally tend to accept its most confident-sounding answers without questioning them.

SECTION 6

Spreading a culture of inclusive accessibility

With principles like these in place, teams can maximize the benefits AI provides and bring real advances in accessibility technology to market faster. Raising awareness of inclusion is a central part of my role at Adservio: by putting the value of inclusive development on the agenda, internally and with our clients, we help enable the creation of technologies that truly make a difference in people's lives, just as they have in mine.

Disclaimer: the statements and opinions expressed in this article are those of the author(s) and do not necessarily reflect the positions of Adservio.

FAQ

Frequently asked questions