I have noticed a number of people are advocating the use of AI/LLMs (Large Language Models) for accessibility testing and/or code remediation.
My onging question is why?
My personal experience with ChatGPT, for example is that it sucks at these tasks. Or it may well be that I suck at using AI to elicit useful/trustworthy results or safe accessible code. What tricks, prompt engineering skills am I missing?
an aside
One accessibility corp. that provided an AI powered “fix my code” site. Have taken it down, probably because it sucked. But are still peddling AI code remediation as a solution
An other aside
One of the uses of AI that I do find worthwhile is the generation of initial drafts of alt text for my article images, for example I asked chatGPT to describe the image in this article. The output:
A surreal image of people wearing knitted full-body masks, with text overlays reading "For accessibility testing" and "For accessibility code remediation." Two figures on the right are marked with large red Xs, suggesting disapproval. The word "AI" appears in the upper left, indicating the role of artificial intelligence in accessibility.
Which is not a bad start, but it’s easy to not notice the inaccuracies, which I did not in the first instance “Two figures on the right are marked with large red Xs, suggesting disapproval.” This part of the alt is incorrect as the X’s pertain to the text “For accessibility testing X” and “For accessibility code remediation. X” they are not placed where they are to obscure anything.
After realising the mistake I updated the text to make it more accurate. note I also clarified the subject of the slide “indicating the role of artificial intelligence in accessibility testing/remediation”
A surreal image of people wearing knitted full-body masks, with text overlays reading "For accessibility testing" and "For accessibility code remediation." each is marked with a large red X, suggesting disapproval. The word "AI" appears in the upper left, indicating the role of artificial intelligence in accessibility testing/remediation.
Further reading
- No, ‘AI’ Will Not Fix Accessibility by Adrain
- 57/30
- “AI” won’t solve accessibility by Eric
3 replies on “AI for accessibility conformance testing? Why?”
I found AI useful for some aspects of accessibility. For example you can use Notbook LM for comparison of different articles. I found this article very helpful https://netz-barrierefrei.de/en/ai-research.html.
Agreed. An accessibility expert I often work with is gung-ho on AI for accessibility. I am not sold on it. It has too many inaccuracies in my experience and is far too confident in making up things that are just plain wrong.
[…] AI for accessibility conformance testing? Why? […]