Models aren't Optional

Models get a bad break, and the key is remembering the golden mean: moderation in all things. Macroeconomics, String Theory, Climate studies, all produce highly complicated models that are presented by our best academics as being thoroughly vetted and informative. Yet, for predictions in the real world, they stink. This leads to people saying models are all garbage, and we should all be engineers. Clearly a bad model is worse than no model, but if you are operating in some domain, you have implicitly or explicitly, a model of that domain. in this way, it's simply nice to write it down as clearly as possible to better understand what you are doing.

 I came across the The Good Regulator theorem (1970) by Roger Conant and Ross Ashby. It is stated "Every Good Regulator of a system must be a model of that system".

the theorem shows that, in a very wide class (specified in the proof of the theorem), success in regulation implies that a sufficiently similar model must have been built, whether it was done explicitly, or simply developed as the regulator was improved. Thus the would-be model-maker now has a rigorous theorem to justify his work.

I don't really follow the proof, but I think it's definitely true that to regulate something well, you need a good model of that something. 

Share this post