Those aren't the principles underlying HTML, and I doubt Tim Berners-Lee was thinking about that at all when he designed HTML. Instead, he was trying to design a format which could be read by both machines and humans, which is why "semantic" HTML elements were introduced. In the end, LLMs were what allowed machines to understand web pages, not semantic HTML elements.
If we do want to make this about accessibility, anyone who has worked with screen readers knows that semantic HTML has failed on that front too. Screen readers rarely understand fancy elements. Instead we often have to create obscene DOM structures to get the reader to say something reasonable, which is at complete odds with the ideals of the semantic web.
Isn't that a failure of screen readers, not of semantic HTML? That is, with semantic HTML in place, changes to screen readers themselves can fix the problem by extracting the information that is there; but, without semantic HTML in place, no effort on the part of the developers of screen readers could extract information that wasn't there. And, to put the emphasis differently, if screen readers won't change at all, then there is absolutely nothing that we can do with HTML—except, as you say, to abuse it to achieve what can at best be a fragile and unreliable effect—to fix the problem on that end.
It may, in some sense, be moot because developers of screen readers aren't interested in making those changes, but I think that placing the onus of responsibility there is a more accurate description.
But they cannot. A blind person just cannot make good use of a visual map feature, so there is no reason to make it in pure html for accesibility. You'll have to find other ways to serve them. (I assume mostly with text descriptions of the area)