Simulating work behavior

TL;DR: In this article, an agent model is run under the control of a comparator process both in a forward-looking mode to generate predictions and in an explanatory mode to compare predictions generated by the agent model with actions of the user.

Abstract: Methods and apparatus for modeling behavior. A computer model includes an agent having one or more beliefs, a world model, and facts. Behavior is modeled by frames including workframes modeling a time-consuming portions of an activity or thoughtframes modeling non-time-consuming portions of an activity or information processing. Detectables model acquisition of, and response to, information during a defined portion of an activity. A detectable tests facts to form beliefs of the agent, which can affect continued performance of the activity. The model is run by selecting for each time one of the frames to be the working frame; maintaining a context of active frames representing activities that are underway at any time; performing the working frame until it completes or another frame is selected to be performed in its place. Another frame may be selected when a detectable effect causes an impasse in the working frame or completes or aborts the working frame, or a higher-priority frame interrupts the working frame. In another aspect, the invention provides the services of an intelligent agent. An agent model models a user with a general component modeling an idealized agent having the role or tasks of the user and a situation-specific component describing the state and the beliefs of the user. The agent model is run under the control of a comparator process both in a forward-looking mode to generate predictions and in an explanatory mode to compare predictions generated by the agent model with actions of the user.