François CholletVerified account

A good model isn't a description, it's an explanation. An accumulation of observations does not explain anything.

While I agree it’s nice to have explanations, it’s not always possible. Why does an apple drop to the ground? Gravity (without gravitons) is a description of how this works, not an explanation, but it’s a damn good model.

Gravity is an explicative model (even though it does not answer the "why"): it explains the *how* and can be used to produce new predictions and simulations. A descriptive model would be the kind of model of the solar system we had in the 13th century, based on observation.

Most of science is based on explicative models. That's what makes them good models, and that's what makes science effective. But some fields rely more on descriptive models (medicine, biology, neuroscience...). Which is still useful for specialized purposes (like neurosurgery).