Artificial Intelligence and Automatic Programming in CAI

This paper discusses generative computer-assisted instruction (CAI) and its relationship to Artificial Intelligence Research. Systems which have a limited capability for natural language communication are described. In addition, potential areas in which Artificial Intelligence could be applied are outlined. These include individualization of instruction, determining the degree of" accuracy of a student response, and problem-solving, A CAI system which is capable of writing computer programs is described in detail. Techniques are given for generating meaningful programming problems. These problems are represented as a sequence of primitive tasks each of which can be coded in several ways. The manner in which the system designs its own solution program and monitors the student solution is also described. There is currently significant interest in generative systems for computer-assisted instruction (CAI). A generative system has the capability to both generate and solve meaningful problems. It must also be able to monitor a student's solution, determine to what extent the student is correct, and provide pertinent remedial feedback. Generative CAI systems free the course author from having to develop several alternative presentations of the same or similar material. In addition, the course author does not have to specify correct and incorrect answers and their consequences as the system can determine all of these on its own. These systems often have the capability to reply to student questions and allow him to divert the instruction to areas that interest him. DataBase Systems There have been a few CAI systems designed which are oriented around a structured network of facts. Wexler describes a system which combines generative CAI with frame-oriented CAI in that the course-author must specify certain question formats. The system generates parameters for these formats and searches the data base to determine the correct answer. It generates remedial feedback from its network if the student's solution is incorrect. This feedback can either be in the form of a correct statement based on the student's response or a trace of the steps required to retrieve the correct answer. On command from the student, it will generate statements from its database which conform to prespecified patterns. Carbonell describes a system, SCHOLAR, in which the generation of several types of questions is done entirely by the system from the semantic network. Also, the system has the capability to interpret a variety of student questions and generate appropriate responses. As in the Wexler system, the semantic network must be …