> Unicode is definitely a prescriptive standard. Code point 65 is a capital A because Unicode says so -- any such standard is necessarily prescriptive.
You're missing the point. Unicode is descriptive like a dictionary is descriptive: it describes — and thereby locks into place and somewhat formalizes (really: ossifies) — existing usage.
If you prefer, rather than "prescriptive" vs "descriptive", you can pretend I said "proactive" vs "reactive." Unicode as a standard reacts to existing out-of-Unicode usage of glyphs, by defining into existence Unicode code-points to allow the encoding of those glyphs. (Or rather, to allow encoding of any of the equivalence-class-hypersphere of glyph-configuration-space described by the name of the codepoint, as that codepoint. Those particular codepoints have been defined through usage, whether the Unicode Consortium likes it or not.)
But even that's not quite accurate, because, especially with the modern forced assignments by Apple/Google, Unicode in fact is purely describing — documenting — codepoint assignments that the OEMs decided on and implemented unilaterally, ahead of standardization. (I.e., Apple/Google now just allocate new codepoints for emoji and assign them glyphs all on their own — and Unicode then must play catch-up, including those codepoints in the standard only after they're already shipping on real devices and thereby effectively already "locked in" in their meanings.)
> Because they don't serve the same semantical purpose. Instead you could have introduced -- for example -- a simple vector graphics format.
They quite clearly do serve a semantic purpose:
• Screen readers can describe an emoji — and smart ones can even use an emoji at the end of a sentence to add emotional color to their reading of the sentence. This would not be possible if emoji were just vector images.
• LLMs have particular trained associations on what a given arbitrary Unicode codepoint should relate to. It would be prohibitively difficult for them to form the same associations between regular word tokens, and the huge sequence of tokens that collectively represents a vector image.
• Search engines can index emoji just like any other text. No search engine that I know of could embed a vector image in its fulltext index in a useful way.
• If a font doesn't represent an emoji, people can still copy-and-paste the unrepresentable-codepoint-glyph rendering of the codepoint into their OS's character map to get a Unicode-standardized description of the codepoint. This wouldn't be true if emoji were just arbitrary vector images.
Also, if you wanted to oust some glyphs from Unicode in favor of embedded vector images, where would you stop? If U+1F60A "Smiling Face with Smiling Eyes" shouldn't be in Unicode, should U+2660 "Black Spade Suit"? How about U+21F6 "Three Rightwards Arrows", or U+2713 "Check Mark"? U+2766 "Floral Heart" / U+2042 "Asterism"?