Discovering Qualitative Empirical Laws.
Abstract
The process of scientific discovery is a complex interplay of many activities, ranging from the discovery of empirical laws through the construction of structural models to explain those laws. This paper focuses on the discovery of qualitative empirical laws and concepts. The authors primary examples come from the history of chemistry, and their model of the qualitative discovery process is an AI system named GLAUBER. After describing the system and providing some examples of its operation in the domain of chemistry, they consider GLAUBER's relation to some other AI discovery systems that operate on the tasks of conceptual clustering and language acquisition. Given a set of observations, GLAUBER defines abstract classes and formulates laws stated in terms of these classes. The authors approach was driven by examples from the history of early chemistry specifically by the development of the theory of acids and bases. Although the existing version of GLAUBER covers many of these discoveries, it has numerous limitations that should be remedied in future versions of the system. These include the need for improved evaluation methods, the ability to distinguish between unobserved and unsuccessful reactions, and the ability to run simple experiments in order to test predictions. These improvements suggest the need for two additional revisions - methods for the incremental discovery of classes and laws, and a search organization more robust than the current hill-climbing scheme.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 15, 1985
- Accession Number
- ADA159235
Entities
People
- D. H. Fisher
- H. A. Simon
- J. M. Zytkow
- P. Langley
Organizations
- Carnegie Mellon University