How much is control knowledge worth?: A primitive example

摘要

The most basic activity performed by an intelligent agent is deciding what to do next. Usually, the decision takes the form of selecting, from among many applicable methods, one method to try first, or opting to expand a particular node in a simple search. The most primitive case is selecting between two independent alternatives. Below, this case is examined and the value of the control knowledge that makes the decision is determined. Another result derived is the sensitivity of the expected value of control knowledge as a function of the accuracy of the parameters used to make these control decisions.