The funny thing about this comment is that neural networks are universal function approximators.
The most fundamental essence of what they do is exactly what you say they don't: estimate.
The most fundamental essence of what they do is exactly what you say they don't: estimate.
You can't prove that )))