Comment by wizzwizz4 | Hacker News Reader

Comment by wizzwizz4 | Hacker News Reader

Parent Full thread

The only viable approach I can think of is to completely rewrite everything from scratch. It's a huge undertaking, but I honestly think it's less work than getting the existing software infrastructure to work accessibly. Even heroic efforts like AccessKit just aren't heroic enough.

We've got a few decent speech synths, but information about how things should be read out isn't passed through to them. That's handled by a screen reader program… except screen readers can't represent half the semantics they should, so people regularly bypass them, which leads to (a) UI inconsistency; and (b) the systems being useless if you need something other than a screen reader. AI scraper bots are the straw that broke the camel's back, so virtually no (current) website is accessible via a basic web browser any longer. UI customisability was low in the Windows 95 days, but we've managed to go backwards from there.

We might as well go the whole way, and design something that's actually usable, then put together case-by-case compatibility layers. Here's how we translate Home Office Design System HTML, here's how we translate Stacks Design System HTML, here's how we translate MediaWiki HTML, here's how we translate Wordpress Gutenberg HTML, here's how we translate Moodle HTML… here's how we represent the OpenDocument content model for reading and writing, here's how we represent the SVG content model for reading and writing, here's how we represent a login flow…

"Ghetto systems" as the saying goes aren't a solution either. Those have been tried decades ago. Remember "IBM Homepage reader"? No? Probably before your time. Its a nice idea, in isolation. However, the translation layers you talk about are never going to be sufficient. You're just moving the problem. Now, if you want to use a system which doesn't have a translation layer for your ghetto system, you're out of luck again.

One of the translation layers can be to-spec WAI-ARIA: then anyone who can be bothered to implement their websites correctly will. (There are currently no correct implementations of WAI-ARIA: only vague approximations of partial implementations.) I don't think there's a way to salvage untagged PDF forms, except taking them case-by-case.

You're right to point out issues with "ghetto systems" – but we don't have a single computer system that actually works, and everything that does exist has fundamental design flaws that make accessibility prohibitively difficult. (Wayland somehow managed to be a downgrade from X11, which is quite a feat.)

I think a basic ghetto system with a full development environment and easily-accessible documentation would rapidly become not a ghetto system.

As a completely ignorant infra engineer, is AI not also a potential solution that avoids rewriting entirely? Like, it feels like translating between visuals and audio signals would be one of the few things it’s really great at.

Perhaps a bit wasteful, but feels more likely than getting the entire software industry (including companies with inverse financial incentives) to get on board.

No. There are AI systems currently in-use, and I expect they could be improved significantly with a purpose-built model (e.g. one that describes what's actually there, rather than describing how it's supposed to make you feel), but there's no way to bring the friction down without compromising accuracy (even further). Describing an image involves editorial decisions that just aren't present when doing language-to-language translation, there's no way to avoid that, and transformer models (etc) are bad at editorial decisions more complicated than "rank the salience from most to least".

A current-gen AI system has all the problems of a ghetto system (see https://news.ycombinator.com/item?id=44297481), with the added bonus of being inherently unpredictable.

https://www.bemyeyes.com/ serves a similar role: higher-latency, different privacy concerns, way better accuracy (most of the time). You can see why this isn't the same genre of thing as "make computer systems that people can actually use without assistance".

The two problems I see with the AI-screen-reader approach is wastefulness and confabulations. It is one thing to have to deal with your screen reader not reading something. However, it is quite a different thing having to hope that your screen reader doesn't outright lie to you. And they do. I recently had BeMyEyes (uses OpenAI vision model) describe an iOS screenshot to me. It imagined an error which definitely was not on-screen. I cliamed I have a storage full warning, for a 256GB iPhone. Thing is, storage is not full, and the iPhone has 128GB. I really dont want such technology to be used for screen reading, its bad at best, and dangerous at worst.

M-X professor-hubert-j-farnsworth-mode

Good news, everybody! Rewriting everything from scratch has recently become a lot, lot easier, and much more fun!

Leela AI (not to be confused with Turanga Leela) learns to speak in LLOOOOMM / Cursor:

https://youtu.be/Sn057QrCUm8?t=5367

https://leela.ai/how-leelas-ai-is-different/

PS: Does anyone know how to make the Mac "say" command use a new fangled "Personal Voice" that you can record of your own?

Reply on news.ycombinator.com