Cristina's Written Questions

Marek (Experimental Design)

In a recent paper, available from Marek, "Influence of Tobacco Marketing and Exposure to Smokers on Adolescent Susceptibility to Smoking", Journal of the National Cancer Institute, 87(20)1538-1545, October 18, 1995, Nicola Evans, Arthur Farkas, Elisabeth Hilpin, Charles Berry and John P. Pierce describe a study evaluating the influence of tobacco advertizing and promotion and exposure to smokers on never-smoking adolescents' susceptibility to smoking. The authors conclude that tobacco marketing may be a stronger influence in encouraging adolescents to initiate the smoking uptake process than exposure to peer or family smokers or sociodemographic variables including perceived school performance. Please, look critically at the study and discuss its strengths and weaknesses. Are the authors' conclusions justified? Are there any major problems with the experimental procedure? If yes, what improvements would you suggest?
Demonstrating or disproving causation is one of the main foci of empirical studies. Statistics is the main source of tools for analyzing data and making scientific decisions. Paradoxically, statistics does not seem to be concerned with the question of causation. "Correlation does not mean causation" is often the only reference to causality (if there is any at all!) that can be found in a typical statistics textbook. Explain what correlation does mean and what it tells us in both experimental and non-experimental settings with respect to causation.
Nancy Cartwright, a computer scientist by training, has designed and built an intelligent tutoring system that "talks" with adolescents about alcohol, tobacco smoking, drugs, and sex. It's goal is to teach the adolescents how to make well-informed, prudent decisions related to risky behavior. The system's interface is very fancy and includes text, graphics, voice, and video. Nancy's intention is to make her tutoring system available on the World Wide Web (through web pages equipped with web scripts). She believes that her system will be a very effective teaching tool. Unfortunately, she has neither training, nor experience to subject her belief to a rigorous empirical test. Provide a reasonable interpretation of what "effective" might mean in this context (Nancy is quite unclear about it herself) and propose a set of studies that will test the effectiveness as you defined it.

Kurt VanLehn (ITS and Cognitive Skill Acquisition)

There is ample evidence that Mastery Learning (as in Bloom, 1984) works. However, the older explanations for why it worked made the "faculty psychology" assumption that different students had different innate mental faculties or capacities. Thus, some students were "slower" and had to be given more training time to reach mastery than"faster" students. John Anderson has proposed a more modern explanation, namely that mastering the prerequisites of a new skill reduces the working memory load when learning the skill. However, neither the faculty explanation nor the working memory load explanation are at the level of cognitive processes. For instance, Anderson, myself and several others have hypothesized learning processes (e.g., Pups-analogy, EBLC) that construct new knowledge. These operate in the context of an overall problem solving process, such as means-ends analysis. Can you hypothesize an explanation for the mastery learning effect at this cognitive process level? That is, how does mastering (as opposed to not mastering) the prerequisites of a skill affect the cognitive processes responsible for acquiring that skill?
An old and controversial issue is, Can learning be taught? That is, can poor learners be taught skills that make them into good learners? Review the evidence bearing on this question, considering both cognitive and motivational facets. If you would like to confine yourself to just the papers in the reading list, you may do so. Feel free to propose your own theory.
It is often said that the only complete ITS is Burton's West tutor. Although this may be an extreme view, it is true that West had more pedagogical rules (e.g., don't interrupt too often; only tutor when there is a clear advantage to learning the new rule) than subsequent tutors (except for Stevens and Collin's Why, whose implementation was never completed). Why is this? Is it due to the West task domain, the pedagogical philosophy, or what? To sharpen your analysis, imagine what a model tracing tutor for the West game would be like, and whether it would be as effective as Burton's tutor. What can you conclude in general about the relationship between the amount of knowledge in the pedagogical component and the particular tutorial application?

Martha Pollack (Planning)
- In their introduction to the recent special issue of "Artificial Intelligence" on planning (Vol. 76, Nos. 1-2), the editors make the following claim:
  
  It may be inevitable for the field of planning to split into even smaller subfields, each with its own domain of interest (manufacturing, deliberation scheduling, logistics planning, etc.) After all, there may not be much in common between designing the behavior of a robot and designing the behavior of a military logistics organization. It would be a pity to see this happen, but if what is gained is a set of elegant and powerful theories coupled with useful implementations replacing the current elegant but weak theories coupled with toy systems, then maybe it will be worth it. (p.13)
  Comment on this claim, in light of the reading you did. Do you agree with it, and if so, what role do you see for the "core" planning systems and algorithms you read about?
- There have been two rather diverse sets of reactions to Ginsberg's idea of "approximate planning". Some researchers believe that his notion is a radical step in the right direction, while others have suggested that all he has done is develop another formalization of the basic planning algorithms that have been in the literature for the past 20 years. Justify one or the other of these views.
- Explain the roles of probabilistic reasoning in plan recognition, and their relationship to the roles of probabilistic reasoning in plan generation.
R. Michael Young < myoung+@pitt.edu>