Full disclosure, I didn’t read every post here, there seems to be a lot of this type of discussion lately.
However, it looks like many people are discussing the details of how to implement such a feature, or whether it’s possible.
I question the why. The mere existence of unidentified observations isn’t as much of a problem that needs to be addressed by automation. It is much more a low-barrier-of-entry opportunity for many would-be indentifiers to become engaged. Very gentle first steps on an identifier’s learning curve.
With the introduction of ML/AI/CV, identification by humans who are not experts is becoming increasingly obsolete. The human’s role is more and more reduced to making judgement calls in ambiguous cases, but even that will fade over time. Automatically identifying left-over unidentified observations would only increase the oft-lamented barrier for beginners, “how could I possibly help, I’m not an expert.”
Of course subject experts remain valuable for identification. I have this nagging thought, though, that today’s experts did not acquire their expertise with the help of AI/etc.