My geekiest post to date
Quote:
Originally Posted by Lonnie 
So I'm going to press you here. What is (in your mind) the difference between controllability and accuracy?
|
I'm being a little facetious here, but I do think that small inaccuracies get washed out in the practic loop: try something; observe results; try something a little different; observe.... repeat till convergence. (It ties in with the perception vs reality differences ssh has discussed on occaision.)
So I may think I'm doing thing A to get the results I want, but I'm really doing B. I tweak to improve results, thinking I'm doing A1, but really doing B1, and so on. Eventually, say on A235, I get the results I want, but I'm really doing B235. Now I'm set. To get the same results again, I just have to try to do thing A235 again - I will acually do B235, and get the results I want.
Controllability is a technical term in control theory -- it means that the full range of desired outcomes (dynamic, not just static) can be created from the range of available control inputs.
Observability is the flip side, meaning that the full range of system states (again, dynamic) can be distinguished from each other using the set of available sensors.
One of the techniques for designing a control system is "model based control" where the controller incorporates a mathematical model of the system. It turns out that in many cases, a pretty crude internal model can give acceptable results.
To make the example concrete - suppose I consistently overestimate how much I have my ankle flexed. If the optimal amount is X at some point, I'll find my turns work better if I flex it by one-and-one-half X (because in reality, it is only flexing by X).