ABSTRACT
Suppose that N agents must jointly select one from a finite set A of possible acts. Agents can be human decision-makers, knowledge sources sharing inferences and expertise via a blackboard, or expert systems distributed throughout a complex system and exchanging messages via an internal communication network. Unless all N of them agree on a choice from A within a certain amount of time, a 'disagreement outcome' occurs by default. After an act has been chosen by consensus, 'nature' chooses a state from a set S of possible states. A consequence from a set C of possible consequences then results. In the simplest causal model, the consequence corresponding to a particular act—state pair is uniquely determined and the causal correspondence between choices and consequences can be represented by a consequence function c: A × S ⇒ C giving the consequences for each agent of each act-state pair (a, s). Elements of C are typically represented as N-ary 'payoff vectors' for the agents (Rosenschein and Breese, 1989). Let c(a, s) denote the consequence obtained when the group chooses act a in A and 'nature' chooses act s in S. Many apparently more complicated decision problems can be reduced to this 'normal form' by suitable definition of A and S; see, for example, Luce and Raiffa (1957, especially Chapter 3) for a classic treatment of this framework for decision analysis.
