Improving the Performance of Satisficing Cognitive Algorithms
We investigate a family of cognitive algorithms that has been proposed
recently by Gigerenzer and Goldstein (1996) to model a kind of human behavior
- known as one-reason decision making - in the task of comparing two objects
as which scores higher on a given criterion based on binary cue information.
How should the cues be ranked in order to achieve the largest number of
correct decisions? We provide a theoretical framework for studying this
question by analyzing the approximation capabilities of satisficing cognitive
algorithms. We introduce an algorithm that has not been considered before
and show that it can be used to improve the performance of any cue-based
algorithm in many cases. We also exhibit a relation between the comparison
task and a class of problems that is studied in the area of machine learning.