E-butterfly is amassing a large inventory of images, and has developed AI to identify species. But this can only use visible characteristics (much in the tradition of 19th century collectors!) Can we reverse this process and let AI use the e-database of images to construct a representative type specimen? Would this be more useful than the somewhat arbitrary selection of a photograph to illustrate a species. A further thought: when images in field guides were hand-painted then it was HI (Human intelligence) that did the generalizing.
1 Like
Hey @GilesR , that is a great suggestion that definitively would merit and entire PhD, and we love it!
I will bring it out to our team, and see if we can come up with a project to do so. Of course, it might take some time being able to generate something like that but, with my limited knowledge of how AI works, I really think is possible and it would be super useful!
Sorry, I am confused. What does this do for me when I am in the field?
How would you use this?
Now, I submit a photo and AI gives me best guess. What would this new thing do?
Anne Tucker
1 Like
Good question. To me the AI (which is a great help!) is a bit like an expert saying that a butterfly is this or that species, and not explaining why. I’m guessing that the AI has an “ideal” or “average” conception of a species that is updated with each new photo. I think it would be neat to see this expressed as an AI generated image. (It could go in a field guide, for instance, rather than a photograph.)
1 Like
Thanks for your response. That helps.
Anne
1 Like
We have talked quite a lot about how we might, one day, use the computer vision to help teach. For example, it can generate a heat map on a photo showing the most important areas it used for that photo to make a prediction. It could also be used to provide feedback on how we can take better images to help with ID (of course sometimes we only can get what we get!). There are likely many other ways we could use it as an automated tutor.
1 Like