Excellent (and nuanced) point. Truth is, all principles are "data-driven" (or more precisely data-informed).

However, what distinguishes principles from observations is they are a distillation of observations into a fundamental assumption/assertion about how the world works which are then used to create a rule-based model. This is in contrast to using data directly to train models, which cuts out the middle step of generating assumptions.

While the line between these two types can be blurry, it is one way we can think about this stuff.