I wonder if AI screen readers will ever use the visual appearance of something to determine what it means. So even if you draw on a `<canvas>` it can still read the text. That would be sick.
Here's an explanation of the accessibility features I've added to my canvas library[1]. These features may not be the best approach, or even the most appropriate approach ... but at least they're there, and they can evolve into something better as/when people offer feedback.
Prefab[1] and especially Bubble Cursor [2] look like be super useful additions for the flat-UI era.
You can contact me directly using my handle at my domain, which is 21337.tech
We need to treat the screen like a PNG which an AI can parse to determine what the text is, what text is relevant to read to the user, where and what the buttons are, how to emphasize reading the text based on the visuals, etc. It's the only thing that will scale and while difficult it seems that creating something like this is within reach.
This actually gives me a little hope that AI-interpreted UI could trigger a trend back to visually obvious and unambiguous controls.
All operating systems and platforms have pretty fluent and in-depth APIs for exposing an “accessibility tree” for assistive technologies to use. It’s your responsibility as a developer to build interfaces for actual users, and there’s very little reason not to.
That's actually a really good idea.
It's 2059. We're using browsers that render pages created with Dreamweaver and Frontpage using advanced machine learning techniques to determine that <blockquote><font size="+1"> is an entry in the table of contents, while <p><font size=+2> is a level 1 header.
I also used to believe that rigorous semantic tagging was The Way To Do Documents, and it was certainly a useful crutch, but we absolutely need to move beyond that.