Does tutoring really have to be intelligent?
Wendy E. Mackay · 2001
This experiment was designed to determine whether or not tutoring is more effective if it is relevant to the user's current problems. The experimental design presented identical tutoring advice to pairs of subjects: advice was directly relevant to one subject (as determined by a human researcher monitoring a pre-specified task) and effectively random to the other.The quantitative and qualitative results were strikingly different. On one hand, subjects learned almost all tutored commands, regardless of their relevance to their immediate activities, and rarely learned commands that had not been tutored. On the other hand, subjects were very enthusiastic about relevant tutoring and were frustrated when it seemed random: they felt it was an irritating interruption. An intelligent rule-based tutor may be unnecessary for effective learning if users can control the tutoring environment.