I’m not an expert in this field. What are the likely root causes for this happening?

The most obvious answer that no one wants to mention is that there just genuinely is some similarity between the two categories which is stronger than the similarity between others.

While I don't have a source, it seems clear that with sufficient training data they could do better at avoiding this mistake. This should not be hard.

Perfect accuracy in image classification is an unsolved problem. So yea, it's hard.

If Google and Facebook have both failed with the billions they have spent on the problem, I think we can say yes, it is that hard.

Obviously the people at google had thought of having more training data.