A Symmetric View of Utilities and Probabilities

Motivated by the need to reason about utilities, and inspired by the success of bayesian networks in representing and reasoning about probabilities, we introduce the notion of utility distributions, in which utilities have the structure of probabilities. We furthermore define the notion of a bi-distribution, a structure that includes in a symmetric fashion both a probability distribution and a utility distribution. We give several examples of bi-distributions. We also show that every state space with standard probability distribution and utility function can be embedded in a bi-distribution, and provide bounds on the size requirements of this bi-distribution. Finally, we suggest a reinterpretation of the von-Neumann and Morgenstern theorem in light of this new model.

• Type: Conference
• Year: 1997
• Where: IJCAI
• Authors: Yoav Shoham
• Keywords: IJCAI 1997, IJCAI 2007, Probability Distribution, Standard Probability Distribution, Utility Distribution