Skip Navigation

How to Dehumanize Accessibility with AI

Published


Context

On November 29th, 2024, I scrolled upon a LinkedIn post about a tool called a11y-ai. It immediately caught my eye (not in a good way) so I decided to go to the tool’s website to read more.

It was enraging, to say the least. I was so angered by what I read, that I created my first TikTok, even after having already posted a loud thread on Bluesky (requires login to view).

I also wrote most of the words below, then disabled life happened and I wasn’t able to finish editing right away. In each section, I respond to quotes from a web page or post on the tool’s website.

Why? AI is being sold as some magical solution to the poor state of web accessibility, and there are times when I can’t just sit by and watch misinformation be spread about AI and accessibility.

I’m an accessibility consultant, multiply disabled, and rely on web accessibility. And it really sucks to watch people reach for AI tools SO QUICKLY over the much more credible and reliable accessibility expertise of disabled people, especially disabled people like myself who do this work for a living.

My Reviews

The homepage

“Accessibility explained by those impacted.

Check any website for barriers and generate the perspective of those who face them.”

Firstly, generative AI is not impacted by inaccessibility. It is not a disabled person. It cannot explain web accessibility from the perspective of a disabled person.

Secondly, it is entirely unnecessary to generate the perspective of people who encounter inaccessible websites. We are here, capable of sharing our perspectives, and we have been doing so for decades.

The plagiarism machine’s been trained on our work, no doubt, and without credit or compensation.

“A11y-AI is an experiment using the storytelling capabilities of generative AI to bring the impact of accessibility violations to life. It aims to inspire web professionals to take action.”

Generative AI does not have the storytelling capabilities of disabled people. Those of us still alive are screaming our stories right now and have been, again, for decades. There is nothing for generative AI to “bring to life.”

As for those of us who are no longer here, many of their stories and wisdom are documented in multiple formats. You just have to care enough to read or listen to them.

The About page

“A11y-AI merges the world of AI and storytelling to make web accessibility more understandable and relatable. It’s an empathy experiment utilising generative AI to transform technical reports into empathy-building experiences.”

To say it again: generative AI cannot tell stories. It is a regurgitation machine. Removing the human factor of inaccessibility stories does not build empathy. It dehumanizes the stories. It dehumanizes US.

Additionally, inaccessibility is not a result of a lack of empathy. It’s a result of ableism. To still position lack of empathy as the main problem in almost 2025, is a failure to consider vital historical context.

While digital accessibility is a younger topic than physical accessibility, we cannot effectively try to improve digital accessibility without referencing previous history.

By looking at several decades of history before the inception of digital accessibility, we can be sure “lack of empathy” is not the core problem now because it wasn’t a core problem then.

It’s also important to recognize that disability activists and accessibility advocates have already tried the approach of empathy-building. It does not work, and this is not news.

“We aim to raise awareness so that the internet can become more accessible and thus more user-friendly. A11y-AI gives an insight into the reality of people’s lives and helps you understand their challenges. Although they are synthetic characters, they represent many people who feel the same way.”

No. Absolutely not. The “synthetic characters” do not represent us. They will never represent us. We — real, living disabled people — represent us.

Just like there isn’t an empathy problem in accessibility, there also isn’t an awareness problem at the core. People would be more aware if it were not for deeply ingrained, systemic ableism.

Creating AI caricatures of disabled people does not help us dismantle systemic ableism. By taking us out of the equation, and positioning fake, digital characters as more credible narrators of our stories than we who actually live those stories, accessibility is literally dehumanized.

Silencing us, the sources of truth, will only reinforce systemic ableism.

And to re-emphasize something I said earlier: AI cannot give insight into the reality of our lives. It does not live our challenges, so it cannot understand our challenges, so it cannot help anyone else understand our challenges.

Can we also note the paradox in saying “reality” in one sentence and then “synthetic” in the next??

A talk titled “AI for empathic assessment of accessibility”

“Existing methods for testing are often very technical and not very attractive, neither for developers nor for users.”

Many existing methods for testing accessibility do have a steep learning curve. I can acknowledge that. However, this is not unique from other parts of digital product development. There are many disciplines that go into building products, and accessibility is one of them.

This means, just like all of the other disciplines, companies must hire experts in the accessibility discipline. Accessibility testing does not need to be attractive. It needs to be effective.

Disabled people are the ultimate experts in accessibility. Many of us also work in the discipline of accessibility, doing exactly what A11y-AI sets out to do.

So not only is this tool being positioned above disabled people in general, but it is also being positioned above the disabled people (such as myself) who are doing accessibility testing and remediation for our livelihoods.

Stealing our stories AND OUR JOBS does not build empathy. It puts us in even more precarious positions, especially when considering facts like: In the United States, disabled people are more likely to be self-employed than nondisabled people (8.4 percent versus 5.9 percent in 2023, according to the US Bureau of Labor Statistics).

I have thoughts on the language like “very technical” in the next major section.

“A11y-AI uses the possibilities of Large Language Models to present the effects of barriers in the web on people with disabilities in a sensitive and personal way.”

It is impossible to be personal if there are not persons involved. Removing the persons is not sensitive, nor personal. LLMs are not persons. Fake, digital characters are not persons. Only people who encounter inaccessibility can credibly present the effects of it, as I’ve written multiple times at this point.

“In a second step, the system uses Axe-Core tool to analyze the website entered and outputs a report.”

It is vital for people to understand that only a small portion of accessibility issues are automatically detectable with tools like Deque Axe. It is also important to understand that these kinds of tools can, and often do, provide misleading results.

Automated testing tools always require a human being to verify the results, and a human being is always needed because manual accessibility testing is always needed.

To reference some data — the Government Digital Service (GDS) published an audit of automated testing tools in 2017. Across 142 test cases and 13 automated tools, the most issues found by a tool was 40%. The lowest was 13%. Take a look at the overview and detailed audit results.

“In a third step, GPT analyzes the content of the website and provides the context. In the presentation, for example, an online music store was presented: Claudia should contextually refer to the website in her response, which is why this step makes sense. This not only makes the system appear more authentic, but also allows it to deal with the barrieres that arise in a humorous way.”

The system cannot appear “authentic” (unless you don’t know anything substantial about what it’s like to live as a disabled person). It will never experience the daily, constant pain and further deterioration of health that is caused by systemic ableism.

It is not affected by:

And this list represents less than a grain of sand within an entire beach of what disabled people are forced to endure EVERY day.

I also take issue with the alleged need for comments to be “appealing” or “humorous.” Nothing appealing nor humorous about inaccessibility. Inaccessibility is PAINFUL in every single sense of the word. When disabled people encounter inaccessibility, we are harmed every. single. time.

Inaccessibility can cause both direct, immediate harm and indirect, delayed harm. Animations, for example, can cause direct and immediate harm like induced physical sickness.

If we try to work around inaccessibility, we risk injuring ourselves in the process. It is always true that we are forced to use extra energy to do the workaround. Since we already have limited energy reserves, this impacts our capacity to do anything else for days, weeks, months, or even years to come.

If we choose to save our energy and not try a workaround, we’re forced to miss out on whatever it is we were trying to do.

No one should expect disabled people to react positively to any of this.

Making one of this tool’s main marketing points that it is “appealing” or “humorous” in its feedback will only harm us further. It will set an expectation that all accessibility feedback must be “appealing” or “humorous” to be taken seriously. It will embolden people to continue ignoring feedback from disabled people that isn’t “nice” enough, which is already a huge problem today.

One of my favorite pieces on this problem is by Alice Wong — writer, editor, disability justice activist, and a 2024 MacArthur Fellow — “Disabled Outrage and #PodSaveJon”. In this January 2024 piece, Alice describes the pattern she’s seen in over 10 years of being on Twitter. Please read it!

A page titled “Anticipating the story behind the spec”

“Our experimental tool “A11y-AI” uses the storytelling capabilities of generative AI to turn technical accessibility reports into tangible, comprehensible user experiences.”

The premise this tool is based upon is an allegedly unmet need for “technical accessibility reports” to be translated into “tangible, comprehensible user experiences.” The “technical accessibility reports” referenced are the ones that come out of automated testing tools.

Deque’s Axe is mentioned on other pages about this tool. Axe was intentionally built for a technical audience — developers — to encourage them to do automated accessibility testing alongside other automated testing that happens in development. Source: axe-core README.md file, accessed December 10th, 2024.

If developers don’t understand the results of Axe, it’s not because the results or the standards they’re built upon are “too technical.” It’s because the developers lack experience in testing accessibility and interpreting the results. It indicates a skill gap and skill gaps are closed with education, coaching, and practice. This is how we address any skill gap in any discipline.

It’s why we hire people with multiple years of experience for a lot of jobs – they already know how to do something you need, they can do it efficiently and accurately, and they can teach others that knowledge.

To suggest that AI is needed to bridge accessibility skill gaps is to suggest that there aren’t human experts already out there capable of doing so. It devalues an entire body of expertise.

A body of expertise in which I have personally endured an incredible amount of pain and suffering, not because people are unaware of the importance of accessibility or lack empathy towards disabled people, but because of ableism and capitalistic greed.

I resent the notion that people need accessibility feedback spoon-fed to them in an “appealing” or “humorous” tone, or that accessibility testing is “too technical.” The message doesn’t need to be softened or immediately “comprehensible” by everyone.

The fact of the matter is — nondisabled people will never be able to comprehend the experiences of disabled people.

And that’s fine. I repeat: IT’S FINE.

Disabled people don’t need nondisabled people to understand the details of our accessibility needs. We need nondisabled people to believe us and take us seriously.

Nondisabled people are confused by disabled bodies and minds. They’ve been unfortunately misled by inaccurate narratives in media and entertainment that are told most often by other nondisabled people.

When disabled people tell our stories in our own words, nondisabled people do not view this as us achieving some level of justice by correcting the record. They choose to continue believing the false narratives given to them their whole lives, and choose to believe disabled people are not being truthful.

Ellen Samuels discusses this “profound anxiety regarding disability imposture” of “the disability con” in Fantasies of Identification: Disability, Gender, Race (page 18) and traces its roots through centuries of history, media, and culture.

Samuels demonstrates this fear’s emergence at two different points in American culture is “in response to new extensions of social benefits to disabled people and others understood as the “worthy” poor” (page 19).

Aside: see Appendix A for more great works that discuss “the disability con.”

“Most people – even developers – find it pretty hard to get excited about abstract, technical specifications. Accessibility standards are no exception. But many things become more tangible when empathy comes into play, when you can see the people behind the specifications and thus better understand the concrete effects that limited accessibility brings with it.”

I can see the intent here (and throughout all the content I’m reviewing here). Everyone who already cares about accessibility wants everyone else to start caring about it.

“How do I convince people to give a shit?” (or a form of it) is one of the most common questions I’m asked. It’s probably the most common but I haven’t counted.

I’ve maintained the same answer since the beginning of my career and will continue to firmly believe and know it is true. We can’t convince everyone to care about accessibility for the right reasons. Some people are deep-to-their-core-un-empathetic.

You can keep trying to gentle-parent them [1] but I will not. My energy, abilities, and years are limited. I do not have a will, a way, or the time to deal with immature adults.

I’ll gladly spend my time with those who want to learn and make the world a more accessible place for disabled people. They deserve it. 🥰

“That’s why we asked ourselves: How can we transform abstract and technical accessibility reports into more personal, tangible and empathy-boosting experiences? How can we use advanced AI technologies to recognize the potential impact of accessibility issues on real people’s lives? And how can we deliver these “aha” experiences to the people who design, develop and operate today’s web?”

I think my overall response to this tool and its marketing can be summarized in 2 sentences.

  1. You don’t need to do that (emphasized).
  2. You (emphasized) don’t need to do that.

Summary

You might be wondering: “So, then, what should be done?” Maybe you think I’m a luddite (derogatory). Oh well! You read this far, might as well keep reading. 🤷🏻‍♀️

Hire the people who do this kind of work for a living. Pay us adequately for our expertise. Listen and believe us when we share our stories with you. Read our stories that are already out there.

AI is not the path to fixing inaccessibility or understanding web accessibility specifications. The path to understanding is hiring disabled people and hiring web accessibility experts. If you find someone who is both, you’ve hit the jackpot.

Appendix A: Suggested Reading

This list is not exhaustive. Many I’ve read, some I haven’t but plan to. If you’d like to suggest more readings, please send me an email!

Appendix B: Footnotes

  1. Many thanks to Heather Buchel for hilariously using the phrase "gentle-parented" in the context of how some people expect to be given accessibility feedback. It gave me a hearty laugh! [back to article]
Back to Top