Agent Modeling in Expert Critiquing Systems

Expert critiquing systems are a type of humancomputer collaborative system in which a computer agent presents reasoned opinions about a human agent’s problem-solving process for a given task. The challenge in such systems is to provide timely critiques relevant to the user’s focus of attention. A problem with many expert critiquing systems is that their critiques are not always timely or relevant; consequently such systems interfere with problem-solving rather than provide assistance. The problem arises in part from insufficient representations of the human’s problem-solving processes. In this paper, we discuss the flexible use of an agent model based on a task-decomposition hierarchy of human experts in a critiquing system called SEDAR. The model differs from previous research efforts in three ways: 1) the structure of the model, 2) the function the model performs in the expert critiquing system, and 3) the influence of the model on communication between the computer and human agents. A prototype of SEDAR was implemented for the flat and low-slope roof design domain. The results of early testing on the prototype show that SEDAR assists users effectively and reduces the error rate.