Value of information

The negotiation planner is based on the idea of value of information, which is that an agent's expected utility can be increased by gaining information of the other agent's beliefs. This idea comes from decision theory [44]. For example, given a game tree with alternatives $a$ and $b$ at a choice node, and a belief with probability value $p$, the utility of each of $a$ and $b$ would be some function of $p$, since the belief may be used to evaluate a chance node in the subtrees of $a$ and $b$. If (and only if) there is a decision surface between a and b in the belief space defined by $p$, the deciding agent would be interested in the value $p$, since it affects his choice between $a$ and $b$. There may be many such beliefs to be resolved, perhaps leading to a long negotiation dialogue. The negotiation ends when the cost of the remaining negotiation dialogue outweighs the benefit that can be obtained from the information it provides. At this point the agents should begin executing the domain-level plan. It is possible for negotiation to interrupt the domain-level plan once again. If an agent takes an unlikely alternative in the domain-level plan, chance nodes that were previously too remote, due to their weighting, to outweigh the cost of the negotiation dialogue may become more interesting.