A challenge to moth-ers everywhere: Could AI show us what we’re missing?

ObsIdentify and similar nature ID apps can be frustratingly flaky. Anyone who has used them regularly will have seen some spectacularly confident wrong answers. I had a warbler in a tree that one app thought was a harbour seal and another obvious beetle that it flagged as some kind of American rodent! Yet there is another side to the story: sometimes it is astonishingly good, particularly when used to confirm an identification you already know almost for sure rather than to make one from scratch. That raises a question that deserves much more serious scientific attention. Could these AI ID tools actually be seeing differences between some species, such as micro-lepidoptera, that we have not learned to spot? British micro-moths, and indeed, the macro-moths, contain numerous species pairs and complexes that, for practical recording purposes, are routinely aggregated in records. We know that some are genuinely separable externally only with difficulty, using genital dissection or having…