I think that the tendency for users to agree with identifications without independently verifying them (this topic and many others) and the tendency for users to choose the top computer vision suggestion without independently verifying it (this topic and many others) results in a feedback loop that produces many research grade observations of obscure, difficult (or impossible) to identify species that do not have identifiers specifically monitoring them for issues like this to pull the observation back to a higher level. I’ve come across this situation in multiple species, and because they are into the computer vision suggestions the problem never really goes away with these species.
The plans to have the computer vision work with taxa levels other than just species should help with the computer vision problem, but I think something else needs to be done in addition to help with the “automatic agreement” problem.