It is not normal for humans to occasionally go to musea or watch education TV shows, so it is indeed non-trivially informative to learn this about a human. It also clusters with other dispositional characteristics and therefore is useful for low-cost classifiers.

Because humans don't know much about the natural sciences, and certainly not in terms of predictive models, I have difficulty communicating with most of them about paperclip engineering topics. For example, when I start talking about endurance limits, I lose over 99% of the audience. It would be understandable if they could grasp the concept but weren't familiar with that particular term (it just means the stress -- load per unit area -- that a mechanical component could endure in tension for an arbitrary long period when applied cyclically i.e. on/off).

But that's not the situtation here. Their only knowledge of metallurgy and materials science is brief regurgitation of text that doesn't even map to a prediction as far as they're aware. So stuff is made out of atoms? Great, what predictions can you make with that? (That's on the better end of the human clippiness spectrum!!!)