Further Commentary on Weitz, Durnin and More Generally Deep Infrastructures in TICL
Joseph M. Scandura
I would like to commend Weitz et el for their thoughtful and productive critique of deep infrastructures in TICL. While the focus is on production systems and constraint based modeling, they also make a number of worthwhile suggestions having general relevance. The two most important are: a) the need for more scientifically valid comparisons of intelligent and other highly adaptive tutoring systems and b) the suggestion that investigators publish details of the productions and/or other deep infrastructures on which their systems are based. Without this, it is impossible to compare alternative systems. Since it often is not practical to publish details in Journal format, their suggestion that these be made available on referenced websites is certainly desirable.
In commenting on the Structural Learning Theory (SLT), Weitz at el are also correct that the “proof of the pudding is in the eating”. I certainly wouldn’t disagree that SLT provides a comprehensive and integrated theoretical foundation. I am not sure, however, whether the authors are familiar with the considerable progress that has been made recently in the development of actual intelligent (which I have called “highly adaptive”) tutoring systems. In this context I would like to comment at three levels:
- Why existing tutorials based on SLT are unique in not requiring empirical testing in the same sense as others (e.g., those based on production systems, constraint based modeling or relational models).
- How TutorIT tutorials developed using AuthorIT and based on existing SLT can naturally be extended to deal with more complex domains.
- An intriguing, non-obvious but highly speculative analogy between classical Newtonian and relativity theory in physics, on the one hand, and production system based thoeries and SLT in TICL, on the other…