A Bayesian Interpretation of the Internal Model Principle