Ontology Evolution in Physics

A computer can be programmed to reason automatically by constructing an ontology in which to represent both some knowledge and the rules to derive new knowledge from old. An ontology is a mathematical formalism. Most ontologies are built manually for a particular reasoning task. Successful reasoning depends on striking a compromise between the expressiveness of the representation and the efficiency of the reasoning process. If either the reasoning environment or the goals subsequently change, then the reasoning process is likely to fail because the ontology is no longer well suited to its task. Many modern applications of automated reasoning need to work in a changing environment with changing goals. Their reasoning systems need to adapt to these changes automatically. In particular, their ontologies need to evolve automatically. It is not enough to remove from or add to the beliefs of the ontology. It is necessary to change its underlying formal language. Our group has pioneered work in this new area of research. Our techniques involve diagnosis of faults in an existing ontology and then repairing these faults. In this project we propose to apply and develop our techniques in the domain of Physics. This is an excellent domain because many of its most seminal advances can be seen as ontology evolution, i.e. changing the way that physicists view the world. These changes are often triggered by a contradiction between existing theory and experimental observation. These contradictions, their diagnosis and the resulting repairs have usually been well documented by historians of science, providing us with a rich vein of case studies for the development and evaluation of our techniques. We face some tricky technical challenges in (a) dealing with the large number of choices in diagnosis and repair and (b) filling in some undefined blanks in some of the repair operations. To solve these challenges we propose to compose together a number of diagnosis and repair operations into what we call repair plans. We have already experimented with two such repair plans, which we call Where's my stuff? and Inconstancy . The first works by dividing some stuff into visible, invisible and total stuff. We have applied Where's my stuff? to case studies as diverse as the discovery of latent heat and the speculation of dark matter. The second works by making some stuff dependent on a variable on which it was previously thought not to depend. This plan is being applied to Modified Newtonian Mechanics (MoND -- an alternative to dark energy) and to the gas laws. Our proposal is to extend this pilot study by looking at a much wider range of case studies, developing more repair plans and evaluating their performance on a test set of case studies.We hope to show that a small set of repair plans can successfully account for a large number of case studies. We will use this work as a basis to develop a theory of ontology evolution that we intend to be applicable outwith the Physics domain.