Automated Reasoning with Situation Theory

In the late 80’s I read Situations and Attitudes by Jon Barwise and John Perry (published in 1983). I was fascinated. I spent the next few years immersed in a project to create a computational realization of some of the ideas in that book. This project became my Computer Science Master’s thesis Automated Reasoning with Situation Theory: A Novel Approach to Beliefs and Perception (1992).

The main products of that research were a mathematical formalism for infon logic, a belief logic, and FELIX, a theorem prover that implemented these logics. FELIX was implemented in Prolog using MacProlog32 on a Mac II and later ported to XGP. (I think it’s too big of a system to run in ProscriptLS.)

From the introduction of the thesis:

Computationally modeling and analyzing intelligence (CMAI) is the large issue toward which this thesis is directed. Situation theory (ST) provides the philosophical and logical basis for the approach of this work. There are many issues in CMAI. Many of these issues have in common various problems in knowledge representation (KR). This thesis provides an application of situation theory to some of these common knowledge representation problems, particularly in the area of reasoning about beliefs.

The central claim of this thesis is that situation theory is superior to classical logic as a foundation for knowledge representation in artificial intelligence. This claim is elaborated in the following four hypotheses:

  1. A version of situation theory can be defined which has a characterizing logic (an “infon” logic) similar in form and expressivity to classical first order logic.
  2. There is a semi-decision procedure for this new infon logic, and a theorem prover can be devised which implements it. Further, many of the techniques of automated theorem proving developed for classical logic can be applied to automated theorem proving in this new infon logic.
  3. This new version of situation theory and the associated theorem prover is appropriate as a knowledge representation and reasoning system for theories of perception and belief.
  4. Theories of perception and belief as defined by their embeddings in the new version of situation theory provide a better account of human reasoning than classical logic-based computational approaches to perception and belief.

Situation theory is a recently developed theory, primarily concerned with understanding information and meaning among people. It has seen only very limited application to computational problems in natural language understanding. Thus, one of the contributions of this thesis is to further develop the computational application of situation theory. Another contribution is a new set of problems for situation theory to address – what are the inference rules appropriate to reasoning in situation theory? Also, this thesis is intended to contribute toward a computational system which is as good at intelligent behavior as a typical person is, neither better nor worse.