AI and Accessibility abdicating engagement UNESCO

 AI and Accessibility: abdicating engagement? | UNESCO

 

Bennett, A (2025) AI and Accessibility: abdicating engagement? UNESCO. Online at: https://www.unesco.org/en/articles/ai-and-accessibility-abdicating-engagement (Accessed 03 September 2026)

 

Further, “digital accessibility” refers to the principles and processes of making digital content that disabled users can fully engage with. That is, ensuring that both the digital output can be accessed by those using assistive technologies (such as screen reader software) and the content itself, (such as image description, transcripts and captions) are functional and readily available to all users. 

 

Accessibility by design is always easier than a retrofit, whether addressing physical or digital accessibility. 

 

Adopting Universal design principles is a far more inclusive approach than addressing contextual requirements as an afterthought. For example, the Web Accessibility Initiative (WAI) advocates for new technologies and digital content to be designed to be accessible from inception (sometimes products designed this way are termed “born accessible”).

 

In contrast, the ways in which some generative AI tools are marketed presents them as means to render content more accessible without having to adjust practice or mindset. This perpetuates the perception of digital accessibility as an added task, rather than an integral part of creating digital content. 

 AI tools promising automatic accessibility fixes with little to no need for human intervention are tempting, offering a quick and easy solution to often overworked staff. However, real accessibility requires engagement.

Moreover, this corresponds with attitudes prevalent in education. Whilst the accessibility of digital teaching materials is generally recognised as important, and even a legal requirement for teaching materials in some countries, it frequently remains an afterthought.

Consequently, creation of accessible digital materials is typically excluded from teacher training programmes, leaving staff without the requisite design skills. In this environment, AI tools promising automatic accessibility fixes with little to no need for human intervention are tempting, offering a quick and easy solution to often overworked staff. 

 

However, real accessibility requires engagement. An example of this is auto-generated alternative text, or “alt text” 

 

The mere presence of an alternative text might make your content appear accessible, but unless it is meaningful, it is merely an illusion of inclusivity. 

 

Even when AI can correctly identify the subject of an image, it cannot identify the message you wanted to convey – a disadvantage, for example, to people with blindness or those who have limited vision.

 

Unhelpful, incomplete or inaccurate image description does not provide equivalent access and could even create confusion or misinformation for anyone relying on this to fully engage with the content. The mere presence of an alternative text might make your content appear accessible, but unless it is meaningful, it is merely an illusion of inclusivity. 

 

Accessibility checker tools also aid the development of web resources, running automated checks to identify accessibility issues. But like AI-generated image descriptions, these tools should only function as a starting point: they can help in the creation of a tool that is technically accessible, but they cannot replace the work done with disabled users to make a tool practically useful. For example, automated accessibility checkers can help you determine if a webpage can be accessed with a screen reader, but they cannot tell you how it will be experienced by human users. Similarly, a checker can identify if alternative text for an image is present, but not if that text is appropriate.

 

Automated accessibility checkers could help clear the first hurdles of making resources digitally accessible, but they cannot replace user research and testing as a means of understanding user experience.

 

A 2024 article looking at the rise of synthetic users – AI-generated respondents for user research – concluded that, whilst helpful for some tasks such as preparing for user testing, synthetic users should not replace testing by human users, as they cannot accurately predict human experience.

 

The authors go so far as to warn the reader to “Avoid adopting this tool if your stakeholders will see it as a replacement for user research.” While these tools have great potential to support accessible development, then, they can't replace final rounds of user testing by the people who are meant to use them.

 

Reliance on automation and AI tools eliminates nuance, giving a poorer experience for disabled users. To continue down this path could lead to an automated era of disability erasure. The design of policies, services and products for disabled people has historically been done by people without disabilities, leading to the adoption of the motto “Nothing About Us Without Us” by disability movements. However, dependence on AI tools without human input and disabled persons’ lived experiences is antithetical to that principle of disabled inclusion. If AI tools lead to automation replacing all manual testing and synthetic users replacing user research, we now potentially risk the design process excluding not only disabled people but also any human testing in favour of AI.

 

Reliance on automation and AI tools eliminates nuance, giving a poorer experience for disabled users. To continue down this path could lead to an automated era of disability erasure. The design of policies, services and products for disabled people has historically been done by people without disabilities, leading to the adoption of the motto “Nothing About Us Without Us” by disability movements. However, dependence on AI tools without human input and disabled persons’ lived experiences is antithetical to that principle of disabled inclusion. If AI tools lead to automation replacing all manual testing and synthetic users replacing user research, we now potentially risk the design process excluding not only disabled people but also any human testing in favour of AI.

 

Uncritical reliance on these tools risks perpetuating the notion of accessibility as an afterthought, promising to automate accessible features without the creator providing nuance or attention to the experience of disabled users. 

 

It’s easy to see why AI tools that promise to resolve digital accessibility issues with little to no input are attractive – they claim to present an easy solution, one all the more attractive to staff potentially lacking digital skills and confidence, or with limited time to prepare teaching materials. However, whilst in future with automated and AI tools may help us create accessible content in some form, a meaningful accessible experience requires engagement, both in education and society as a whole