Computer-based test interpretations as expert systems: Validity and viewpoints from artificial intelligence theory

摘要

Principles of artificial intelligence, particularly the notion of expert systems, are applied to computer-based test interpretations (CBTIs), which are programs for transforming psychological test data into interpretive reports. CBTIs are compared against an evolutionary scale ranging from simple algorithmic systems, which execute rule and equations and trigger verbal statements in an invarying sequence, to systems that are capable of learning. This treatment includes user interface requirements, a summary of threats to CBTI validity inherent in various parts of the program, and a review of research directed at alleviating those threats to validity. The state of the industry for CBTI construction is relatively primitive compared to the state of the art for artificial intelligence applications. The implications for the development of improved CBTI designs are discussed such as the use of interactive interfaces where the user can ask questions to which the program can respond, the development of classification systems that become “smarter” as they acquire more information, and follow up with specially designed evaluation studies to determine how well psychological measurements have been transformed into meaningful text.