I think that's because we're in the very early stages of a transitional period. The end goal is something like the Star Trek computer, free-form natural language queries ending in desired actions. I think this approach of filing off all the rough edges (namely, handling every possible unique way of phrasing a specific intent) is a dead end.
What we have nowadays is more like a voice-input command line, with all of the stiltedness and attention to detail when crafting your input that requires.
Re:Amazon’s plan for Alexa
by buzzkillington on Monday November 11, 2019 @01:56AM
(#21502761)
A voice input command line is exactly what we need.
Unfortunately people want to pretend that natural language parsing can be solved, which, at least, for English is impossible: https://en.wikipedia.org/wiki/Time_flies_like_an_arrow;_frui...
All of these commands printed out on a list somewhere would probably be consistent, and make sense. But you don't get to see that list. Google and others want to present these systems as magical AI helpers that you just converse with, as if they had any more intelligence than a touch tone menu system circa 1990.
Knowing how these things work, it just strikes me as hubris. The UX isn't good because they can't admit that the whole system isn't as good as they say it is or want it to be, and this makes them opaque and frustrating to interact with.