Prediction is usually easier than understanding, according to the theory of inverse problems.

Imagine a well defined system that can be observed only in limited ways and conditions, with errors.
Estimates of what is going on inside the system (that are what one might call 'understanding') are
often very sensitive to those errors, and can be made reliable only if measurements are made in a 
wide range of conditions, which is not always possible.

The estimates of the internal parameters may not be very close to 'the real thing' particularly if the
structure of the real thing is also unknown.

The mistake estimates of the internal parameters can be used to predict the behavior of the system
in new conditions, using the forward version of the inward model that created the parameter estimates
from data. 

If you think about it, many errors in the inverse, then forward process will tend to cancel, particularly if
the prediction is for conditions that are not too far from those measured. If they do cancel significantly.
PREDICTION IS EASIER THAN UNDERSTANDING.