Agenthood

Following the theory of two player games [47], it is assumed that two agents each select a strategy from a number of alternatives available to each. Different combinations of strategies are assigned different utility values with respect to each agent. A mechanism is required whereby each agent chooses the strategy that maximises its expected utility.

The game is organised into turns, in which each agent selects and reveals its strategy to the other. An assumption that simplifies the decision process is that of perfect information, where each agent finds out the strategy chosen on the previous turn before making its choice. This is reasonable since in most dialogues the environment is immediately observable to the other agent, and most agents wait until they have recognised the other's last act before proceeding with their one. Without this assumption, there are games where one agent cannot settle on one particular strategy, since if it did settle either of them, there would be a worse alternative which the other could choose. The agent is therefore forced to adopt a randomised "mixed-strategy" [47]. However, with the assumption, the choosing agent may use a simpler process of just taking the maximum utility strategy at each node in a game tree.