Limited Rationality

Our algorithm is based on ideas borrowed from limited rationality as defined in Russell and Wefald's book [14] which, in turn, was inspired by the work done by Simon [16]. Russell and Wefald define the agent's thinking as the execution of a series of computational actions, and use metalevel thinking techniques to determine which computational action, if any, the agent should execute next. We have taken this approach and applied it to the problem of using recursive agent models. The decision made by the metalevel thinking is which model, if any, to expand next. The change of domain required us to define the concept of gain since the traditional utility concept was not sufficient for the implementation of a successful algorithm.