1. an agent responding to a query should attempt to do so in a way that reduces the need for future queries.
2. Multiple agents responding to a single query need to provide legitimately different responses. A military planner responding to a crisis needs to understand and evaluate his options at a qualitative level.
3. Software agents solving problems need to understand the limitations of the solutions they propose.

The key insight is the recognition that all of these concepts can be described in terms of the distribution of problem solutions in the overall search space. Solutions tend to cluster; the larger the cluster in which any particular solution appears, the more robust the solution. Solutions appearing in different clusters are qualitatively different; limiting factors correspond to situational changes that move one outside of a solution cluster entirely.

The realization that these concepts refer to properties of the solution space leads immediately to another insight: ideas such as robustness are best understood not as properties of a planner/scheduler, but of the solutions that it produces. We will not speak about a planner being robust, but of a planner producing robust plans. From an operational point of view, the fact that a plan is capable of surviving a variety of situational changes or opposing responses (i.e., is robust) is far more important than the details of the computational architecture used to generate that plan.

IWCM Project Page
AI Center Home Page
SRI International Home Page