The machine vision on iNaturalist is great, but it's not necessarily adequate for distinguishing between insect species -- there can be details that are difficult to photograph, let alone identify from a casual user's photograph.
And we still need training data (i.e. identified by the expert taxonomists) for future work.
I have been at talks by their devs. They candidly admit it will never meet the needs (accuracy) to address questions at this level of specificity. Some things will be extremely successful, many (most) will not.
Think of biodiversity as a curve, with a long tail. Will AI on poorly taken images specimens work for the bell? Probably. Will it actually get at the numbers at the tail? Almost certainly not. This is largely because 1) getting at the tail requires intimate knowledge of where to find that biodiversity (the vast majority of iNaturlaist pics are shockingly close to civilization, where diversity may not be) and 2) intimate knowledge, often of internal features or other non-imagable data, so that one can actually record data that fits in the tail.
And we still need training data (i.e. identified by the expert taxonomists) for future work.