Case-based planning in an exploratory discovery system

It has been argued that the case-based reasoning method combined with case-based planning can be used to improve a machine discovery system (R. Oehlmann, D. Sleeman and P. Edwards, 1992). In contrast to the case-based approach, current machine discovery systems typically use experimental results to generate explanations without using previous experience. Furthermore, machine discovery systems do not employ strategies which support the controlled exploration of different options for experimentation. In the authors´ implementation, IULIAN, case-based planners were used to generate self-questions, answers, and experiments, thus combining the two approaches. They argue that: the reasoning process represented by self-questions and answers can be planned and the index of a question plan should utilise information about previous failures of the reasoning process. The authors present a top level view of the IULIAN system and discuss their approach to planning the reasoning process. They use the example of question plans to describe important aspects of IULIAN´s memory organisation and the related indexing problem. They also evaluate the approach and discuss various options for future work