PROBLEMS IN DECENTRALIZED DECISION MAKING AND COMPUTATION

Abstract

2with the case in which a single serial processor is to execute a sequence of instructions, in order to evaluate a desired result (value).In decentralized computation , the same goal is achieved by a set of processors operating in parali« and exchanging partial results.Parallel computation is advantageous in many situations, because the desired final result may be evaluated much faster, or becai the input data of the computation are physically distributed (for example, if the computation consists of statistical processing of data acquired by physically disti sensors).Interesting problems arise in this context because good parallel algorit can be very different from simple adaptations of good serial algorithms.Concerning decision making problems, especially in human organizations, it is often the case that distinct decision makers have different objectives, which are also different from the objective of the organization.This may lead to conflict i to situations best addressed by game theory.We will restrict, however, to situati in which: a) There is a well-defined organizational objective.b) The individual decision makers are either physical processors (so that no interest may be ascribed to them) or they may be treated -for the purpose of analysis -as if they were processors with predictable behavior.For example, a human decision maker may make decisions motivated by his perceived selfinterest; however, if an analyst knows the perceived self-interest of that decision maker, he may be able to predict his behavior.From that point on, ^elf-interests become irrelevant: as far as analysis is concerned, the human decision maker may be modelled as a processor, operating in a specific way.*In this context, the terms "parallel" or "distributed" computation are often usedSince S is an "easily computable" set as a function of its arguments, we can see that the centralized counterpart of DS is a trivial problem.So, any difficulty inherent in DS is only caused by the fact that information is decentralized• * Throughout this Chapter an overbar stands for negation.Also, Λ stands for logical "and" and V stands for logical "or".