Reasoning from Data Rather than Theory

Thecurrent frameworkfor constructing intelligent tutoring systems(ITS) is to use psychological/pedagogical theories of learning, and encode this knowledgeinto the tutor. However,this approachis both expensiveandnot sufficiently flexible to support reasoning that somesystem designers wouldlike intelligent tutors to do. Therefore, weproposeusing machinelearning to automatically derive modelsof student performance. Data are gathered from students using the tutor. These data can comefrom either the current student or from previous users of the tutor. Wehave constructed a set of machinelearning agents that learn howto predict "high-level" student actions, and use this knowledgeto learn how to teach studentsto fit a particular learninggoal. Wediscuss other related workat using machine learning to construct modelswithin ITS. Bycombiningseveral different systems, nearly all of an ITS's decision-makingcould be performedby machine learner derived models.Anopenquestion is the comp...

• Type: Conference
• Year: 2000
• Where: FLAIRS
• Authors: Joseph E. Beck, Beverly Park Woolf
• Keywords: Artificial Intelligence, FLAIRS 2000, Intelligent Tutoring, Intelligent Tutors, Modelsof Student Performance