This technology relies on developing a very capable learning and reasoning agent shell, called Disciple-RKF, that can be taught directly by a subject matter expert how to solve various problems in a way that resembles how the expert would teach a person. The knowledge base of a Disciple-RKF agent is structured into an object ontology and a set of task reduction rules. The object ontology represents the objects from an application domain and can generally be imported from existing knowledge repositories such as CYC. The task reduction rules use these objects to represent the problem solving knowledge of the subject matter experts.

To rapidly develop an integrated knowledge base, a knowledge engineer and a subject matter expert first import and/or develop an object ontology to be shared by all the subject matter experts. Then each subject matter expert directly teaches a personal Disciple-RKF agent how to solve problems. Finally, the knowledge bases of the personal Disciple-RKF agents are integrated into a larger knowledge base that represents the collective problem solving expertise of the subject matter experts. This integrated knowledge base can also be exported into the CYC knowledge repository.

Disciple-RKF is a complex system composed of a suite of innovative tools for the end-to-end development of an integrated knowledge base. It includes the following tools to be used by a subject matter expert with no or very limited assistance from a knowledge engineer: the scenario elicitation tool (to elicit specific object descriptions through a natural language interaction with the subject matter expert), the modeling tool (to express expert's reasoning in natural language, based on the task reduction paradigm), the task formalization tool (to transform the informal natural language expression of a problem solving task into a formal logical expression), the task learning and refinement tool (to learn and further refine a general task description starting from a specific task), the rule learning tool (to learn a general task reduction rule from a specific example of task reduction), the rule refinement tool (to improve a previously learned rule), the interactive problem-solving tool and the autonomous problem-solving tool (to perform problem-solving tasks through task reduction), and the ontology learning and refinement tool (to learn and further improve new object concepts and features). Disciple-RKF also includes the following tools to be used jointly by a knowledge engineer and a subject matter expert: the ontology import tool (to develop an initial object ontology by reusing object descriptions from CYC or other knowledge repositories), the knowledge base export tool (to translate the developed knowledge base and to incorporate it into the CYC knowledge repository), and several ontology browsers and editors (to develop, visualize and navigate the object ontology with the support of intelligent assistants).

The development of Disciple-RKF benefits from the results of an AFOSR funded project aimed at creating a theory of mixed-initiative knowledge base development that integrates human and agent reasoning, to take advantage of their complementary knowledge, reasoning styles and computational strengths. It also benefits from the collaboration with the Center for Strategic Leadership of the US Army War College on the development and experimental use in several courses of a knowledge base and an agent for the identification and testing of strategic center of gravity candidates.