next up previous contents
Next: Negotiation Up: Previous Work Previous: Agents and Multi-Agent

Conflict Resolution

Klein [1991] suggests a model of conflict resolution having a hierarchy of conflicts with the most abstract conflicts at the top and most concrete conflicts at the leaves. There is a corresponding hierarchy of resolution strategies. This structure is shown in figure 2.1. Klein's hierarchy has domain dependent and domain independent conflicts and their associated resolution strategies. The nodes higher in the hierarchy represent the domain independent conflicts while the lower nodes become more domain dependent. The conflict hierarchy first differentiates all conflicts into two broad categories. The first is two agents giving incompatible specifications for a design component. The second category is one agent criticizing specifications asserted by another agent.

Figure 2.1: Hierarchy of conflicts and strategies

Klein's model of agents, as shown in figure 2.2 all have their own design knowledge and conflict resolution knowledge. The agents have differing design expertise, but their conflict resolution knowledge is identical, consisting of the two hierarchies mentioned above.

Figure 2.2: Components of agents

The operation of the conflict resolution components of the agents in the Klein model, shown in figure 2.3, are as follows. Conflicts that occur are matched to the most specific conflict type on the hierarchy. To able able to do this matching, agents need to gather information about the current conflict situation. Agents get this information by sending queries to each other using a query language. Then the corresponding strategy on the hierarchy of resolution strategies is used to solve the conflict, probably after some specialization which again involves queries among agents. If this strategy fails, the strategy represented by the parent node of the current strategy is used. Since this strategy is more general, it is applicable to a wider class of conflicts. The process ends either when the conflict is resolved or when the strategy represented by the root node is reached and there is not an agreement between the conflicting agents, in which case there is a failure. An example of a general strategy is: If two plans for achieving two different agents' goals conflict then find an alternate way of achieving one goal that does not conflict with the other agent's plan for achieving its goal [Klein & Lu 90] [Klein & Lu 91].

Figure 2.3: Operation of conflict resolution component

The SINE work has taken a small step in the direction of this model. Work presented in this thesis has many more similarities to Klein's work as it defines a similar conflict hierarchy. The main differences are that the conflict hierarchy presented in chapter 5 does not have any domain dependent nodes, and the agents, as explained in section 4.3 all have their own unique conflict resolution knowledge.

Another interesting approach to conflict resolution is Sycara's Situation Assessment Packages (SAPs) [Sycara 87]. SAPs are information structures that contain, among other things, a description of a problem solving situation, expectations, reasons for expectation violations, and warnings for failures.

next up previous contents
Next: Negotiation Up: Previous Work Previous: Agents and Multi-Agent

Ilan Berker
Thu Apr 27 16:25:38 EDT 1995