Prediction Cannot be Used to Validate a Model

Scientists are not predicting nature, they are predicting the outcome of an experiment.

Those are two VERY different things.

An experiment is an extremely pared-down version of nature with well-defined pieces designed to reveal some version of the expected.

A prediction that agrees with a model has very little information content, as it is almost devoid of surprisal.

Experiments are often little more than reflections of the model itself. The model has chosen to view nature through a very specific lens, and as such, there are but a few measured values that would signal agreement between nature and the model.

But nature does not operate in such low dimensionality, so the scientist must drastically constrain nature artificially for her to produce these values.

By the time the measurement “agrees with prediction” we are observing some residual remnant of nature, forced into an unnatural container, fashioned by the same conceptual structures as the invented model.

Prediction cannot be used to validate a model, only to show that an experiment has successfully constrained nature to the model’s constructs.