Eh I wouldn't say that. You can selective breed certain plants to look extremely similar to other plants over the course of even a single year. As in, you would be able to trick a human being as well. If AI is making the same mistakes as a human, I don't really think it's a problem with the model as much as a limitation of the data

I mean taxonomists themselves have often been proven incredibly wrong once phylogenetic studies have been conducted. There's simply major limitations to identifying a species by it's physical characteristics alone. Not to mention you don't get to, for example, smell, touch, taste, or watch something over time. Really it's incredible how accurate they even are already