A bird ID-ing algorithm will totally miss that a bird has 3 eyes or 2 heads as long as the markings are ok. It has no clue how birds work.
I mean... this is what happened when image recognition algorithms tried to caption another algorithm's attempt at generating "an image of a girl eating a large slice of pizza".
Spoiler: they saw nothing wrong.
Generated image from: arxiv.org/abs/1711.10485
Spoiler: they saw nothing wrong.
Generated image from: arxiv.org/abs/1711.10485
Also state-of-the-art algorithms will do things like this
Azure's image recognition algorithm has difficulties with sheep
Here's Visual Chatbot weighing in on an AI-generated image of "a herd of cows is grazing in the grass".
Visual chatbot: demo.visualdialog.org
Image from supp. data of this paper: arxiv.org/abs/1711.10485
Visual chatbot: demo.visualdialog.org
Image from supp. data of this paper: arxiv.org/abs/1711.10485
Visual Chatbot sees nothing wrong with the other algorithm's "a herd of sheep grazing on a lush green field".
Visual chatbot: demo.visualdialog.org
Image from supp. data of this paper: arxiv.org/abs/1711.10485
Visual chatbot: demo.visualdialog.org
Image from supp. data of this paper: arxiv.org/abs/1711.10485
1/2 
1/2 
