Computational Asymmetry in Strategic Bayesian Networks

Among the strategic choices made by today’s economic actors are choices about algorithms and computational resources. Different access to computational resources may result in a kind of economic asymmetry analogous to information asymmetry. In order to represent strategic computational choices within a game theoretic framework, we propose a new game specification, Strategic Bayesian Networks (SBN). In an SBN, random variables are represented as nodes in a graph, with edges indicating probabilistic dependence. For some nodes, players can choose conditional probability distributions as a strategic choice. Using SBN, we present two games that demonstrate computational asymmetry. These games are symmetric except for the computational limitations of the actors. We show that the better computationally endowed player receives greater payoff. In competitive arenas such as computational finance, web search and advertising, Internet security, and AI challenges, adoption of computer programs is an important strategic choice made by economic agents. One limiting factor to the adoption of these programs is the computational resources available for processing. This introduces a new potential for economic asymmetry, analogous to information asymmetry: computational asymmetry. Actors may be symmetrical in all respects except for their capacity to compute. Intuitively, we might guess that more computationally powerful players will often be better off. Under what conditions is this the case, and how can we model situations where computational strategies are in play? Algorithmic game theory offers a strong foundation for analyzing this phenomenon, but we believe a new theoretical apparatus would assist us. In order to demonstrate the potential effect of computational asymmetry on economic agents, we develop a new kind of game specification: a Strategic Bayesian network (SBN). Strategic Bayesian networks are an extension of