We're treating statistical correlations in training data as if they constitute understanding

But correlation isn't causation, and pattern-matching isn't comprehension. These distinctions matter