Despite the changes in technology and in educational focus there is still an ongoing desire for educational and training systems to tailor their interactions to suit the individual learner or group of learners: for example, by being able to deal appropriately with a wider range of background knowledge and abilities; by helpfully limiting the scope for the learner to tailor the system; by being better able to help learners reflect productively on the experience they have had or are about to have; by being able to select and operate effectively over a wider range of problems within the domain of interest; by being able to monitor collaborative interchanges and intervene where necessary; or, most tellingly, by being able to react sensibly to learners when the task they are engaged on is inherently complex and involves many coordinated steps or stages at different levels of granularity. Individualising instruction in an effective manner is the Holy Grail of ITS work and it is taken as an article of faith that this is a sensible educational goal.

This paper explores the question of how much educational difference the “AI” in an ITS system makes compared either to conventional classroom teaching or to conventional CAI methods. One criterion of educational effectiveness might be the amount of time it takes students to reach a particular level of achievement. Another might be an improvement in achievement levels, given the same time on task. So the paper surveys the recent past for ITS systems that have been evaluated against unintelligent versions or against traditional classroom practice and finds cause for optimism in that some of the techniques and solutions found can be applied in the present and the future.