Yves Smith has an interesting discussion about the use of models (no, we're not talking about the good looking human ones, it's the boring mathematical ones)in economics. It could be added that even as many economists have a Friedmanite instrumentalistic "it's the predictions that matter, not the realism of assumptions" approach, their models tend to be pretty useless when it comes to predicting future events. Their definition of "predictions" is instead that the models should be able to "predict" past datasets.
Another aspect that could be added is that the users of such "models" fail to grasp the key difference between physics and economics. Models in physics can predict almost all physical events with stunning precision, at least within experimental conditions. Yet as the latest revelation of faster than light particles and previous paradigm shifts (from Aristotelian to Newtonian to Einsteinian to Quantum physics) illustrates, physical scientists appear unable to figure out the underlying causal mechanisms.
By contrasts, good economists can through praxeological reasoning describe the underlying causal mechanisms. But even the best economists cannot make exact predictions about future events.
The reason why economists knows more about the underlying causal mechanisms is that it follows logically from the axiom of action and other indisputable truths applied to specific circumstances. However, because economics can't make controlled experiments and because so many of the specific factor of specific cases are unknown, precise quantitative predictions can't be made-at least not with certainty. Physicists have by contrast very little axiomatic knowledge to go by and must rely on inductive conclusions from experiments. But because they can make experiments they are usually able to make precise predictions within those experiments.
The problem is that most economists believe that they must emulate physics, even though in reality their science is, or should be, radically different. And that is why the make unrealistic models that pretend to be able to make precise predictions, and why they employ unrealistic theories that conform with these models.