@aardrian@tink The owner of the last agency I worked with was a "classic Tech Bro" (was trapped at burning man this year) who LOVES AI.
When we first connected, I told him I thought AI mostly just exacerbated the "StackOverflow problem" of people dropping code they don't understand into projects that can ruin them in big but subtle ways.
Fast forward a few months, and he's excited about how "AI can solve A11Y issues for us!"
Nah, dude, your devs just need to learn that a div is not a button.
@aardrian I hate doing it, but the "you're gonna get sued" stick is literally what I had to pull out to get them to give a shit. They were building sites for utility providers that were just completely broken for screen reader and keyboard users.
"What's the minimum we have to do to not get sued?"
@aardrian Having spent the last year and a half funding projects looking to improve outcomes for people with disabilities using AI and related technologies, I am hopeful, yes.
Also agree that using AI to help address software inaccessibility is a hard nut to crack, but worth pursuing.
For whatever it’s worth, I’m still skeptical when it comes to AI generally, especially language models. That said, I think we’re starting to understand how to use them effectively to improve accessibility.
@Aaron I agree there is lots of potential and I am hopeful fake-AI can get past the scammers and environmental damage to genuinely help more people.
I just wish it wasn’t hyped beyond merit at the expense of existing technologies that can already so much good work (or caused them to rebrand as “AI”).
Sadly, I am at the age where it sounds like an old man afraid of new technology when I am really just wary of the tech culture that raised me.
@aardrian Wait wait wait... they trained their AI in accessibility by showing it code they're to incompetent to know is actually inaccessible, causing the AI to regurgitate inaccessible output?
@heydon@aardrian I think a decision-tree (type of thing) would perform better than an LLM. Tick-boxes and get the correct output for your scenario, rather than the fuzzy-matching process provided by an LLM.
@aardrian IMHO, nobody can fix accessibility, but we can/should try… I would apply the same thinking toward AI. Now if you mix money into this affair, that’s a different matter altogether. So, I would concentrate on the money issue, first and foremost.
Add comment