From the article:
One of the main challenges with deep learning is the need for large amounts of training data, which is slow, difficult, and expensive to collect and label.
Indeed, and maybe too much for the process to be feasible. Taxonomists are in short, and declining, supply. Many large genera have never been subjected to a modern revision, and Museum drawers are full of undescribed species. Take Agra for example: https://www.inaturalist.org/taxa/251408-Agra.
It is not matter of machines “collecting” training data, but of human experts producing them manually - well, intellectually.