To produce a software product for high schools and the undergraduate curriculum of universities to help students learn about the language and reasoning involved in scientific experimentation. It was theorized that it should be possible to represent the knowledge of a researcher / research assistant using a knowledge-based system. If domain knowledge and knowledge about probabilistic models was included, then it should be possible to do experiment simulation based on this knowledge representation. Dr. Rainer von Königslöw developed an experiment simulation system in the early 70's in Fortran IV for mainframes. The system was used very successfully for some of the earliest interactive learning experiments at Queen's University, and other places. (Reported at conferences around 1976). Graduate students could add new models with help from Dr. Rainer von Königslöw. Graduate students tutored undergraduates in learning about scientific experimentation.
The expert system shell has to be extended and redesigned to allow representing abstract knowledge about experimentation and statistical analysis as well as more concrete knowledge about various experiment domains. Rather than depending on knowledge engineers, the knowledge representation must be simplified so that the models can be developed by teachers and students. In other words, the models must be very easy to develop. To fit the classroom, and to make experiment simulation affordable, the system must run on PCs instead of mainframes. Graphics must be incorporated.
The system focuses on the language and reasoning underlying experimentation, rather than on specific techniques. This supports more abstract learning about science and complements current laboratory and classroom approaches.
The system is student based, and encourages the student to learn through exploration and by making mistakes. The student is empowered by asking an "automated research assistant" to do the actual experimentation. Experiments can be done quickly (20 minutes or so), so that the student can do several, and thus improve over time. Scientific experimentation is represented as an expert system - knowledge-based representation. This allows advanced students to understand the underlying modeling, and to assist in adding models.
Reasoning about experimentation must be represented as expert system based knowledge representation. This includes 3 levels of reasoning:
At the abstract experiment design level, the system has to reason about topics such as surveys, within-group or between-group designs, etc. The reasoning structures must carry forward into reasoning about data analysis.
At the specific domain level, the system has to represent and reason with independent and dependent variables, apparatus, materials, etc. At the specific event level, the system has to support probabilistic reasoning to support modeling different types of behaviours. Special events and special cases that can cause experimental error must be supported (e.g., contamination, overdoses, etc.)
The challenge is to represent the different levels of abstraction involved in experimental design. We also need to represent reasoning with probabilistic models -- and incorporating this kind of predictive reasoning into a goal-oriented, backward-chaining expert system shell. (In this case the model has to analyse the design and predict the data -- in contrast to a trend detecting expert system, where probabilistic trends are analysed from the data and reasoned about.)