Ideas
E-Learning

AI and Accessibility: Why Human Engagement Still Matters

road with a sign telling access only

Photo by Patrick Robert Doyle on Unsplash

Listen the Article

Key takeaways

  • Artificial intelligence tools such as subtitles, text-to-speech and image descriptions can remove real barriers for students with disabilities.
  • A 2025 review of 47 studies found significant progress, but also biases, high costs, privacy risks and inequalities in access.
  • Automatic corrections can create the illusion of accessibility when no one checks the result.
  • Explainable AI helps teachers and students understand – and therefore trust – what a tool is doing.
  • In the context of vocational training, AI works best as a support for staff, not as a substitute for their judgement.

Artificial intelligence is changing the way people with disabilities learn, work and travel. Screen readers now describe photos, apps translate speech in real time, and chatbots rephrase complex texts into simple language. For vocational education and training (VET), AI and accessibility have become a key issue. However, the latest research sends a clear message: technology alone does not make learning inclusive. In this article, we examine what AI is already capable of, where its limitations lie, and what VET staff can do about it.

What do we mean by AI and accessibility?

Accessibility means that people with disabilities can use a service, tool or piece of content fully and independently. Digital accessibility applies the same concept to websites, documents, videos and apps. Artificial intelligence and accessibility come together when machine learning helps to remove a barrier. For example, it can convert speech to text, text to speech, or an image into a spoken description.
The need is enormous. According to the World Health Organisation, over 1.3 billion people live with some form of disability, accounting for around 16 per cent of the world’s population. Every vocational education and training organisation, therefore, works with learners who could benefit from better tools.

From niche assistive tools to everyday features

Artificial intelligence has been supporting accessibility for longer than most people realise. As early as 1976, the “Kurzweil Reading Machine” combined text recognition with speech synthesis, enabling blind users to listen to printed pages. Subsequently, speech recognition software enabled people with physical disabilities to write using their voice.
Many of these tools have since become commonplace. YouTube, for example, launched automatic subtitles in 2009 for deaf and hard-of-hearing viewers. Today, millions of people use subtitles simply to watch videos without sound. Predictive text has followed a similar path: it began as an aid for people with dyslexia or limited mobility, and is now found on every smartphone keyboard.
Microsoft sums up this approach with the principle “design for one, extend to many”. In other words, addressing an accessibility need often improves the experience for everyone.
Generative AI adds a new layer. In 2023, the Be My Eyes app introduced a “virtual volunteer” powered by GPT-4 that describes photos and answers questions about them. Be My Eyes is also featured in the IDEAS tools repository, as it helps students read signs and instructions whilst studying abroad.

What does the research say about AI and accessibility?

A recent exploratory review conducted by Abdullah Alsaleh, published in Acta Psychologica, analysed 47 peer-reviewed studies from 2018 to 2025. The review covers the topics of mobility, communication, cognitive support, education and independent living. Overall, the review confirms clear progress, particularly in terms of personalisation and real-time adaptation.
In the field of education, the review highlights adaptive e-learning platforms that modify content for students with dyslexia, as well as real-time transcription for deaf students. Similarly, text-to-speech tools and writing assistants help students with ADHD or physical disabilities to participate in lessons.
At the same time, the author lists several persistent problems:

  • Algorithmic bias, as training data rarely reflects the diversity of experiences related to disability.
  • High costs and unequal access, especially in resource-poor contexts.
  • Data privacy concerns, as many tools collect sensitive personal information.
  • Limited user involvement, as people with disabilities are all too often test subjects rather than co-designers.

The conclusion is simple; developers should create AI not just for people with disabilities, but together with them.

The illusion of access: when automation replaces engagement

An article published in UNESCO’s IdeasLAB, written by Alice Bennett of the University of York, issues an even more forceful warning. Some artificial intelligence tools promise to automatically resolve accessibility issues, with little or no human effort. For overworked teachers, this prospect seems tempting. However, the article argues that true accessibility requires active engagement.
Automatically generated alternative text is a good example of this. An AI can recognise what an image depicts, but it cannot know why that particular image was chosen. Consequently, the description might be correct yet still of no use to a blind reader. Automatic checking systems also have a similar shortcoming: a page that uses only the word “image” as alternative text can pass the test, even if it is of no practical use.
The article also highlights that teacher training rarely covers how to create accessible materials. As a result, staff often lack the necessary skills and resort to quick fixes. This is even more important now, as the EU accessibility standards coming into force in 2025 apply to both commercial services and public bodies. It is worth bearing Bennett’s conclusion in mind: “If digital teaching materials are not accessible, they are not complete.”

Why explainable AI matters for trust

Many AI models operate like a black box. Not even their developers are always able to explain how they arrive at a result. As IBM explains, explainable AI (XAI) is a set of methods that makes these results understandable to humans.
Why should this matter to a vocational training provider? Trust depends on understanding. When a tool simplifies a text or suggests a support measure, staff need to know what that choice is based on. Otherwise, they are unable to spot errors or biases. IBM also emphasises the human aspect of explainability: teams need training to understand how and why AI makes decisions. Similarly, Alsaleh’s review calls for explainable methods, so that educators and healthcare professionals can follow and challenge the AI’s decisions.
Transparency transforms AI from a mysterious helper into a tool that people can verify, adjust and challenge.

What this means for inclusive VET mobility

The IDEAS project operates precisely in this area. Across Europe, support for disadvantaged students in terms of mobility remains low: CEDEFOP’s Mobility Scoreboard gives this aspect an average score of 2 out of 5. Artificial intelligence tools can reduce some of the barriers that force these students to stay at home, such as language difficulties, reading problems or the stress of being in an unfamiliar place.
For this reason, the IDEAS partners are creating a repository of digital and AI-based tools. Each entry describes the barriers that a tool helps to overcome, its strengths and weaknesses, and its suitability for mobility projects. The collection includes Google Translate for everyday communication, Cboard for students who communicate using symbols, and text-to-speech tools for complex documents. Furthermore, the interactive Evaluation Matrix enables staff to assess any new tool against the IDEAS inclusion criteria.

Five practical tips for VET teachers and trainers

  1. Use AI for the first draft. Generate alternative texts, captions or simplified texts, then review and edit them.
  2. Test with real users. Automated checking tools are useful, but only students can tell you whether a resource works for them.
  3. Ask how the tool works. Choose tools that explain their results and allow you to correct them.
  4. Protect personal data. Never share sensitive information about students with a public AI tool.
  5. Build accessibility from the start. Design accessible materials right from the planning stage, not as an afterthought.

Frequently asked questions

Can AI automatically make teaching materials accessible?
Not entirely. AI can produce useful drafts, such as captions or image descriptions. However, a person must check that the result is accurate and meaningful for the learner.

What is explainable AI?
Explainable AI (XAI) is a set of methods that helps people understand how an AI system has arrived at a result. It fosters trust and accountability and makes it easier to identify any biases.

Which AI tools help vocational training students during a period of mobility abroad?
Among the most useful are translation apps, text-to-speech, speech-to-text, visual assistance apps and communication cards. The IDEAS tool repository describes each of these with practical examples of how they can be used.

Technology helps, people include

Artificial intelligence will continue to redefine the concept of accessibility, and this is good news for many students. However, these tools are only as effective as the people using them. That is why IDEAS invests first and foremost in staff skills. Explore the IDEAS tools library, try out the Assessment Matrix, and let us know which tools are making a difference in your organisation.

Sources