The way I think of it, iNaturalist at some point in their history chose to let unknowns exist, and they are now still choosing to let unknowns exist. If they wanted to get rid of existing unknowns they could automatically apply the computer vision suggestion. If they wanted to prevent new unknowns they could require a basic broad ID upon upload (something that I first started thinking when I joined iNaturalist six years ago, and still think, although I acknowledge problems with the idea.) They have done neither of these things, therefore they either think unknowns are not a problem, or they think the potential solutions have too many downsides.
A lot of people promote unknowns as the place where new identifiers should go to practice identifying. In my early years on iNaturalist I used to spend time identifying unknowns, because that was what others suggested I do. After a while I came to the conclusion doing so isn’t an efficient use of my time. I know more about plants anyway. And why put so much work into a category iNaturalist could, but doesn’t, choose to eliminate. But the real thing that stopped me from identifying unknowns is that they generate endless notifications. Put a broad ID on something, and for years you’re going to be notified any time someone refines the ID, even if your ID isn’t actually “wrong.”