"modeling" is different.) If we think about oversimplified, or too abstract, disconnected form reality models and especially those probabilistic models, when they are trying to infer the next immediate outcome based on the probability distributions of from the past (which is completely wrong - all you could infer form probability distributions is what could happen on average), then, it is a total disaster. Look at these loses in finance and profound failures in humanities.

Models has to be "just right", like every molecular arrangement in the Nature.

But the skill of "extracting correct models from some aspects of reality" is really important.