Learning to teach with reinforcement learning agent

Intelligent tutoring systems (ITS) use artificial intel-ligence techniques to customize their instruction to fit the needs of each student. To do this, the system must have knowledge of the student being taught (commonly called a student model) and a set of pedagogical rules that enable the system to follow good teaching prin-ciples. Teaching rules are commonly represented as a set of "if-then " production rules, where the "if " side is dependent on the student model, and the "then " side is a teaching action. For example, a rule may be of the form "IF (the student has never been introduced to the current topic) THEN (teach the topic to the student)." This approach is fairly straightforward from a knowl-edge engineering perspective, but has many drawbacks. First, there are many such rules, and it is very expensiv