The Stanford University Formal Reasoning Group led by Professor John McCarthy proposes research in artificial intelligence relevant to high performance knowledge bases and planning.
Concepts of Logical AI, included with this proposal, describes many concepts important in our work. Especially relevant to HPKB and planning are the logical theory of context, elaboration tolerance, logical ontologies of action and methods of non-monotonic reasoning.
Recent work in knowledge representation has developed new ways of representing information. This information previously was lost, as it could not be represented. In particular, theoretical developments in uncertainty, knowledge gathering, reasoning about actions, and reasoning about contexts, allow us to capture more of the planning process on computers than was previously feasible.
The major innovations that we intend to develop concern how plans adapt in the light of new information. In order to assimilate arbitrary new information, a system must be able to understand information represented in different formats, and must be able to understand the assumptions, biases, and context of the information. In order to take these features into account, we must first represent them, and then make them available to the system in an efficient way. Our formal theory of context allows us to explicitly represent these facts.
The group intends to develop foundational theories that enable HPKB technology, in particular, so that it can be applied to planning. The formal theory of context will allow the integration of KB's, developed by different groups, at different times, and the representation of the many relationships between KB's. We expect these inter-relationships to grow in importance, as larger and more detailed knowledge bases are created. In order to reason successfully with multiple large KB's, information on how they are related is vital. We also see a formal representation of contexts playing an important role in intelligent information retrivial, and information understanding.
All knowledge bases grow and develop. Elaboration tolerance, is a measure of how easily a KB can be developed to meet changing circumstances. Understanding what makes representation elaboration tolerant is a focus of the Formal Reasoning Group.
Real world knowledge is uncertain, and often comes as ``best guesses'' or defaults. Using this knowledge successfully therefore necessitates default, or non-monotonic reasoning. A key area of real-world knowledge is reasoning about action and change. Representing non-monotonicity, and reasoning about change requires ontologies, and formalisms that correctly mirror how the common sense world works. The Formal Reasoning Group intends to continue its research into modeling and characterizing reasoning about change, defaults, and common sense knowledge.