Intent Inferencing

The fundamental principle underlying intent inferencing is to keep the operator in control, even though the system is able to carry out a series of tasks automatically. Thus, when using this technology, the tasks that the system executes are based on inferences made about the goal the operator is trying to achieve and actions that relate to implementing that goal. The operator does not directly tell the system what to do but rather continues to perform activities. The intelligent system analyzes these activities and makes inferences about the goal of the operator and the best plan to execute to reach that goal. Based on this inference, the system carries out the desired process automatically.

One intent inferencing model, described by Jones et al. (1990), was developed as part of the operator function model expert system. This model, the actions interpreter, dynamically builds a model of operator goals for the current system state and then works to interpret the user's actions in terms of these goals. Each goal is decomposed into a hierarchy of plans, tasks, and operator actions required to fulfill the goal. This representation evolves over time as new information is recorded. The operator function model developed by Mitchell (1987) provides the basis for the action interpreter's knowledge about how events trigger likely operator goals. The actions interpreter has been evaluated in the Georgia Tech Multisatellite Operations Control Center simulation, an interactive simulation that supports simulated satellites and the computer and communications technology used for data capture.

Another framework used to represent intent inferencing is the plan and goal graph (Geddes, 1989; Rouse et al., 1990; Shallin et al., 1993). The plan and goal graph is a task analytic decomposition of the goals and plans for all operators interacting with the system. The top-level nodes in the graph are goals. A goal represents a specific criterion on the state of the system that can be tested by observation. The next level of nodes are plans. Plans involve activities, time frames, the use of resources, and side effects. Plans are decomposed into subgoal, subplans, and actions. Several plans may share common actions and compete for resources. Thus, a key element of the program involves resolving conflicts. The formalism of the relationship between plans and goals guides the decomposition process. The idea is to develop the goal and plan structure for the set of missions the system is expected to perform.

As noted earlier, much of the work in this area has concentrated on military aircraft, specifically the pilot associate program (Banks and Lizza, 1991) and the rotocraft pilot associate (Andes, 1996). Currently, Geddes and his colleagues are working with NASA under the advanced air transportation technology program to demonstrate intent inferencing as an emerging technology that can be used to detect goal and plan conflicts among active participants in free flight scenarios (free flight is discussed in detail in Chapter 9). This research moves from interpreting the intent of one operator to interpreting the intent of several operators.

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement