← Back to context

Comment by noduerme

5 days ago

Just coming off a wild ride where a client was sued by a non-customer and a rapacious legal firm, who claimed that said client's website was not sufficiently accessible.

The day after the lawsuit was filed, a company specializing in accessibility testing mysteriously contacted the client, offering a solution. Client had not even gotten notice of the litigation yet.

The net result of this was several tens of thousands of dollars spent actually removing Aria tags and using standard modern HTML on their aging website, to barely meet some threshold that appeared to be compliant.

The company who did the "work", and I mean, it was barely any work, maybe 100 LoC, stands by it and says the client won't get sued again, as long as they pay for ongoing compliance testing. So it's all a fucking racket.

I pointed out to the client that I didn't think that this half-assed effort was remotely sufficient to actually improve accessibility, but they had an interesting response. Which was this:

In 3 years, all this compliance shit will be out the window, because AI screen readers and agents are going to make the whole point moot.

I can't really disagree with that.

> In 3 years, all this compliance shit will be out the window, because AI screen readers and agents are going to make the whole point moot.

Since the whole compliance racket is totally disconnected from actual accessibility outcomes, why would AI have any impact here?

There’s a standard and a law and money to be made.

  • Because the law doesn't stipulate which methods you use to make something accessible to people with disabilities. It just requires that everyone have equal access. To litigate, someone has to show that they couldn't access something, and that the failure to access it caused them some measurable harm. In a couple years you can show a judge that a free LLM screen reader could have solved their access issues, and my guess is that those cases will then be thrown out, and the predatory law firms will move on to something juicier.

    • It’s not just about reading the screen, it’s about using it. It’s quite the downgrade going through a slow API hoping to be able to navigate compared to the 10x speed (or more) of screen readers. Then you have the problem of LLMs being stateless and sensitive information etc etc.

    • >In a couple years you can show a judge that a free LLM screen reader could have solved their access issues

      It seems like it'd be equally trivial to demonstrate that said reader doesn't work on some combination of hardware and software.

Excellent response. I said it before, forcing the responsibility of making thing accessible on the world makes no sense now (made sense before though). Just use an AI to interact with the app/website; it can provide whatever sort of accessibility you need. It should be built as a chrome extension or even native...

  • Does it count if you have to pay for a service for something to be made accessible? I'm sure a blind person nowadays could hire an assistant to use their computer for them, but that would not be reasonable.

    • So the whole world needs to pay (spend more) to accommodate to your problem? What next? You don't have legs so everyone around you must carry you?

      Look, the goal is not to make things accessible. The goal is, out of generosity of the society (and I fully support it), to allow disabled people to access the services. Before recent times, it was only meaningfully possible thru building accessible friendly websites/apps.

      This has always been a hit and miss. Usually only the super big companies did it right (see Google apple Ms) or people using big company's framework with some carefulness. But for the most, it didn't work out. That's why so much complaint against websites being inaccessible.

      Now we have a better way. Use a local AI to consume or interact with the content. This will empower the disabled far more than they currently are. Apple will probably lead the way here.

      1 reply →

  • > Just use an AI

    You mean like (a) assume this perfect AI that already exists never makes a mistake because AI doesn't do that, (b) accept all the security risks like prompt injection, (c) accept the fact that you don't deserve to enjoy the same privacy over your life as others do?

    • A. Any decent ones are more than good enough to provide content in an accessible way.

      B. Like you accept the risk of running scripts? Or getting zero day thru your browser? Ridiculous. Sandbox it.

      C. Local LLM is enough. Nothing needs to go to a server.

> In 3 years, all this compliance shit will be out the window, because AI screen readers and agents are going to make the whole point moot.

If that's the case, why didn't they spend a few bucks on Claude to do the accessibility changes?