Automated identification of species based on images will depend on the species being described and imaged in the first place. The images will have to be fed into whatever AI application emerges--it will not be able to identify a species unless there is an image of it in its memory. Otherwise AI might label as new a species that in fact has been described but for which it does not have an image. It seems to me dubious that a sufficient bank of images can be assembled to make such an application broadly useful. Further, the client who uses it will have to be able to take a quality image of the specimen to be identified. Meanwhile, in the next lab over, an experienced taxonomist is sorting specimens at an order of magnitude faster than the application. By the way, whatever happened to SPIDA-web? I did a Google search for it and found nothing. Could it be that the requirement to have multiple images of each of the tens of thousands of described spider species proved an obstacle to great to overcome?
I believe that SPIDA-Web was a proof of concept project. As I do not recall how many authoritatively identified images were initially fed to the software, but as larger numbers of images (and an increasingly representative sample of the range of variation—I believe we are talking about images of pedipalps specifically) were submitted to the database the accuracy increased to be surprisingly good. You are correct, of course, that it depends on a competent taxonomist feeding reliably identified images to the software; on a competent taxonomist having described the species in the first place; and, of course, users sufficiently well trained to photograph the pedipalp from the correct angle. Although it performed better than I would have imagined, I agree that it is no panacea and second your vote to have more competent taxonomists as the primary investment!
Automated identification of species based on images will depend on the species being described and imaged in the first place. The images will have to be fed into whatever AI application emerges--it will not be able to identify a species unless there is an image of it in its memory. Otherwise AI might label as new a species that in fact has been described but for which it does not have an image. It seems to me dubious that a sufficient bank of images can be assembled to make such an application broadly useful. Further, the client who uses it will have to be able to take a quality image of the specimen to be identified. Meanwhile, in the next lab over, an experienced taxonomist is sorting specimens at an order of magnitude faster than the application. By the way, whatever happened to SPIDA-web? I did a Google search for it and found nothing. Could it be that the requirement to have multiple images of each of the tens of thousands of described spider species proved an obstacle to great to overcome?
I believe that SPIDA-Web was a proof of concept project. As I do not recall how many authoritatively identified images were initially fed to the software, but as larger numbers of images (and an increasingly representative sample of the range of variation—I believe we are talking about images of pedipalps specifically) were submitted to the database the accuracy increased to be surprisingly good. You are correct, of course, that it depends on a competent taxonomist feeding reliably identified images to the software; on a competent taxonomist having described the species in the first place; and, of course, users sufficiently well trained to photograph the pedipalp from the correct angle. Although it performed better than I would have imagined, I agree that it is no panacea and second your vote to have more competent taxonomists as the primary investment!