Attractor-Based Fitness Landscapes for Computational Decision Search

Abstract

Managerial decision making involves searches for alternative courses of action, including searches for technological innovations. A substantial stream of computational work on managerial decision making has been based on search using Kauffman's NK landscape model, which represents fitness or payoff values to a discrete set of binary strings. In this paper, we propose a new method for landscape generation, the method of superposition of attractors, in which the fitness landscape is continuous. We introduce the attractor-based (AB) fitness landscape model, the core model based on this method, with parameters specifying the number of attractors and the steepnesses and heights of landscape peaks in the neighborhoods of attractors. We then describe search using this model, consider issues in implementing the search process, and provide an example of applying the model to studying exploration and exploitation. Next, we compare the AB and NK landscape approaches and identify some advantages and disadvantages of the AB approach relative to the NK approach. Advantages of the AB model include more control over the shape of the fitness landscape, applicability to outcomes not arising from intraorganizational interdependence, and visualization. We then consider customizations and generalizations of the model, including applications to coordinated exploration and resource partitioning processes.