A Framework for Learning Implicit Expert Knowledge through Observation

There are a number of limiting factors presently constraining the development of a truly intelligent and autonomous machine. The most significant of these is that acquiring expert knowledge continues to be a difficult and time-consuming process. Automated knowledge acquisition techniques have been partially successful in reducing the effort involved in acquiring knowledge from an expert and representing it in a form that can be used by the computer. Most of these techniques, however, focus on the gathering and representation of explicit knowledge. This type of expertise, which includes facts, formulas and rules, makes up most of the expert’s knowledge and is relatively easy to articulate. There is a second type of expertise called implicit knowledge, which includes the more abstract forms of intuition and judgment and is much more difficult to articulate and represent. One way humans learn implicit knowledge is by observing others handle real-life situations and by adapting what we have observed to handle new situations. Most current approaches only address explicit knowledge via query sessions and ignore the implicit expertise altogether. Humans, on the other hand, continually learn and apply both types of knowledge. Yet, intelligent autonomous systems that can operate in an unfamiliar environment require the representation of both types of knowledge. Having machines that can reason and behave in a manner similar to a human expert requires formulating an approach for capturing and representing the implicit knowledge that is commonly applied by experts. This paper describes a framework for automatically learning implicit knowledge by non-intrusionally observing the behavior of a human expert in a simulation of the tasks to be reproduced by the autonomous system. This framework is implemented as a prototype, is rigorously tested and evaluated, and conclusions are derived from the work.