Generative “AI” and EN 301 549 (incl. Free Webinar!)

Hello from your web accessibility pal, Eric Eggert.

It starts to be spring here in Germany, going outside is much more enjoyable than in the winter, a winter, by the way, that felt incredibly long and sad.

I’m looking forward to being a bit more active during spring and see where that takes me.

Of course, as always 1000 thanks to the people who have taken out a membership, I really appreciate it, and it helps to justify putting time into standards discussions.

In the last few months, I am trying to redefine what being an effective accessibility person means for me. There is so much focus on the product of “doing an audit” although that is rarely the right place for clients to start. I wrote about the infuriating inefficiency of accessibility audits (Abre numa nova janela) 18 months ago.

👋 Eric

“AI” corner

My esteemed colleague Casey Kreer (Abre numa nova janela) conducted research into different “AI” models (Abre numa nova janela) and how they are able or not to generate accessible interfaces. The results are not good.

First, all models were severely error prone when not mentioning accessibility in the prompt. This reflects real-world guidance where accessibility often is not considered, too.

Mentioning accessibility halved the issues with some tools and slightly reduced the issues in others. Giving detailed accessibility descriptions, like how to implement forms properly with labels, for and IDs, did result in widely different results, sometimes better, sometimes worse than the lightly prompted results.

Note that Casey’s research is giving the “AI” models the benefit of the doubt on purpose. Of three attempts, only the two with the least errors (tested automatically through axe) counted.

What really surprised me was the comparison with the WebAIM million report (Abre numa nova janela). Even the best models produced worse results, and for most models they are not even close to matching what we have done manually. I wrote before (Abre numa nova janela) about how the absolute numbers in the report are not as important, but that the trend line of improvements really matters more. With coding “AI”, it looks like the newer models are creating more complex solutions which are more error prone. If “AI” is more widely used, I expect to see a backslide in the WebAIM numbers, too.

(Note: WebAIM and axe run tests differently, but I think due to giving the benefit of the doubt, that it is a very useful comparison.)

Read the announcement of the research (in German). (Abre numa nova janela)

Please support Casey if you can, her research is invaluable and she’s just a great person. Research like this would not be possible without Steady members that support it. Disclaimer: I support Casey’s work, too.

One blog post

• “WCAG is difficult to read, don’t read it” is a self-fulfilling prophecy (Abre numa nova janela) – I did not blog much over the last months, but jumping off from Meeting WCAG is an achievement (Abre numa nova janela) from last newsletter, I needed to follow up with the notion that reading and understanding the actual Web Content Accessibility Guidelines (WCAG) is important. There is a sea of “Easy WCAG” resources out there which can be too simplified or contain errors. While I understand that the text of the standard can be difficult and intimidating, the work to understand it pays off a lot.

EN 301 549 update & webinar

The EN 301 549 is the European Norm that contains WCAG and adds additional requirements and which determines if you meet the requirements set forth in the European Accessibility Act (EAA).

Currently, the norm is in the process of getting updated to what is called version 4.1.0. (The previous versions were, I kid you not, 1.1.2 in 2015, 2.1.2 in 2018, 3.2.1 in 2021.) The plan is to include WCAG 2.2 and change a few of the unclear rules. For example there has long been a debate about the EN requiring dark mode in websites, which seems to get addressed. (You cannot block someone from your site when they are using dark mode now. Of course it is not explained what a “block” is.)

I had a look into the new draft last week, and while it was planned that it would be published in July, this seems less certain now. It’s a substantial update, but some changes feel just unnecessary, including the complication with minimum font sizes that I outlined in my comment on the standard (Abre numa nova janela).

I think it’s a very opaque process, which is not helped by the fact that work is done in Word documents and hence there is no clear commit path.

• Current version of the EN (3.2.1) – PDF (Abre numa nova janela)

• Draft of the next version (4.1.0) – PDF (Abre numa nova janela)

• Changes – Word diff document (Abre numa nova janela)

• Issue tracker for comments (Abre numa nova janela)

Free webinar about the EN on, March 12: Lunchtime! What EN 301 549 brings to the table

Axess Lab asked me to present a lunchtime webinar about the EN 301 549 on March 12. In it, I’ll focus on the additional bits of guidance beyond WCAG that the EN brings.

The webinar happens at 12:30, which is 7:30am on the US East Coast, 20:30 in Tokyo, and 22:30 in Sydney.

Sign up for free by following the instructions on the Axess Lab website. (Abre numa nova janela)

Parting words

Thanks again to everyone who has subscribed to this newsletter, and a special thanks to all members. You know who you are, and you rock! Without you, my engagement in standards and elsewhere would not be possible.

👋 Eric

Socials

My primary social media/Fediverse/Mastodon handle is @yatil@yatil.social (Abre numa nova janela), and you can also find me (increasingly reluctantly) on LinkedIn (Abre numa nova janela).