While I don't believe it's hard to see that perspective and be cognizant of it, I don't quite agree with that sentiment. Categorization itself has an immense level of usability. E.g. building steps to simulate a real human with AI. Even if like the below poster says, the spectrum may be more continuous than discrete, it is perhaps much faster to try to advance in a particular field with categorization than solely relying on unsupervised learning.