I hear you. We were talking about your proposal yesterday and it is possible, although it would take some time to train. But as I said in that thread, just as our species ID AI was trained on observations with millions of IDs made manually, so would this AI, and the community would still have to go through and correct AI mistakes for the next training run to improve things. As to whether we have “enough” annotated observations now to train an acceptable model, that’s something we can look into. But more will pretty much always be better.
@karoopixie interesting experiment, thanks for sharing.