In machine learning, a learned model should be able to generalize to unseen data. Curve fitting does not have that characteristic.
if you're extrapolating, it'll depend on whether you've assumed the right boundary conditions, but that's true for any model