A major difficulty in developing and maintaining very large knowledge bases originates from the variety of forms in which knowledge is made available to the KB builder. The objective of this research is to bring together two complementary knowledge representation schemes: term subsumption languages, which represent and reason about defining characteristics of concepts, and proximate reasoning models, which deal with uncertain knowledge and data in expert systems. Previous works in this area have primarily focused on probabilistic inheritance. In this paper, we address two other important issues regarding the integration of term subsumption-based systems and approximate reasoning models. First, we outline a general architecture that specifies the interactions between the deductive reasoner of a term subsumption system and an approximate reasoner. Second, we generalize the semantics of terminological language so that terminological knowledge can be used to make plausible inferences. The architecture, combined with the generalized semantics, forms the foundation of a synergistic tight integration of term subsumption systems and approximate reasoning models.

There currently exists a gap between the theories proposed by the probability and uncertainty and the needs of Artificial Intelligence research. These theories primarily address the needs of expert systems, using knowledge structures which must be pre-compiled and remain static in structure during runtime. Many Al systems require the ability to dynamically add and remove parts of the current knowledge structure (e.g., in order to examine what the world would be like for different causal theories). This requires more flexibility than existing uncertainty systems display. In addition, many Al researchers are only interested in using "probabilities" as a means of obtaining an ordering, rather than attempting to derive an accurate probabilistic account of a situation. This indicates the need for systems which stress ease of use and don't require extensive probability information when one cannot (or doesn't wish to) provide such information. This paper attempts to help reconcile the gap between approaches to uncertainty and the needs of many AI systems by examining the control issues which arise, independent of a particular uncertainty calculus. when one tries to satisfy these needs. Truth Maintenance Systems have been used extensively in problem solving tasks to help organize a set of facts and detect inconsistencies in the believed state of the world. These systems maintain a set of true/false propositions and their associated dependencies. However, situations often arise in which we are unsure of certain facts or in which the conclusions we can draw from available information are somewhat uncertain. The non-monotonic TMS 12] was an attempt at reasoning when all the facts are not known, but it fails to take into account degrees of belief and how available evidence can combine to strengthen a particular belief. This paper addresses the problem of probabilistic reasoning as it applies to Truth Maintenance Systems. It describes a belief Maintenance System that manages a current set of beliefs in much the same way that a TMS manages a set of true/false propositions. If the system knows that belief in fact is dependent in some way upon belief in fact2, then it automatically modifies its belief in facts when new information causes a change in belief of fact2. It models the behavior of a TMS, replacing its 3-valued logic (true, false, unknown) with an infinite valued logic, in such a way as to reduce to a standard TMS if all statements are given in absolute true/false terms. Belief Maintenance Systems can, therefore, be thought of as a generalization of Truth Maintenance Systems, whose possible reasoning tasks are a superset of those for a TMS.

Rather than discussing the isolated merits of a nominative theory of uncertainty, this paper focuses on a class of problems, referred to as Dynamic Classification Problem (DCP), which requires the integration of many theories, including a prescriptive theory of uncertainty. We start by analyzing the Dynamic Classification Problem and by defining its induced requirements on a supporting (plausible) reasoning system. We provide a summary of the underlying theory (based on the semantics of many-valed logics) and illustrate the constraints imposed upon it to ensure the modularity and computational performance required by the applications. We describe the technologies used for knowledge engineering (such as object-based simulator to exercise requirements, and development tools to build the Knowledge Base and functionally validate it). We emphasize the difference between development environment and run-time system, describe the rule cross-compiler, and the real-time inference engine with meta-reasoning capabilities. Finally, we illustrate how our proposed technology satisfies the pop's requirements and analyze some of the lessons reamed from its applications to situation assessment problems for Pilot's Associate and Submarine Commander Associate.

This paper describes NAIVE, a low-level knowledge representation language and inferencing process. NAIVE has been designed for reasoning about nondeterministic dynamic systems like those found in medicine. Knowledge is represented in a graph structure consisting of nodes, which correspond to the variables describing the system of interest, and arcs, which correspond to the procedures used to infer the value of a variable from the values of other variables. The value of a variable can be determined at an instant in time, over a time interval or for a series of times. Information about the value of a variable is expressed as a probability density function which quantifies the likelihood of each possible value. The inferencing process uses these probability density functions to propagate uncertainty. NAIVE has been used to develop medical knowledge bases including over 100 variables.

We introduce and analyze the problem of the compilation of decision models from a decision-theoretic perspective. The techniques described allow us to evaluate various configurations of compiled knowledge given the nature of evidential relationships in a domain, the utilities associated with alternative actions, the costs of run-time delays, and the costs of memory. We describe procedures for selecting a subset of the total observations available to be incorporated into a compiled situation-action mapping, in the context of a binary decision with conditional independence of evidence. The methods allow us to incrementally select the best pieces of evidence to add to the set of compiled knowledge in an engineering setting. After presenting several approaches to compilation, we exercise one of the methods to provide insight into the relationship between the distribution over weights of evidence and the preferred degree of compilation.

While concept-based methods for information retrieval can provide improved performance over more conventional techniques, they require large amounts of effort to acquire the concepts and their qualitative and quantitative relationships. This paper discusses an architecture for probabilistic concept-based information retrieval which addresses the knowledge acquisition problem. The architecture makes use of the probabilistic networks technology for representing and reasoning about concepts and includes a knowledge acquisition component which partially automates the construction of concept knowledge bases from data. We describe two experiments that apply the architecture to the task of retrieving documents about terrorism from a set of documents from the Reuters news service. The experiments provide positive evidence that the architecture design is feasible and that there are advantages to concept-based methods.

In two recent papers, I have proposed a description of decision analysis that differs from the Bayesian picture painted by Savage, Jeffrey and other classic authors. Response to this view has been either overly enthusiastic or unduly pessimistic. In this paper I try to place the idea in its proper place, which must be somewhere in between. Looking at decision analysis as defeasible reasoning produces a framework in which planning and decision theory can be integrated, but work on the details has barely begun. It also produces a framework in which the meta-decision regress can be stopped in a reasonable way, but it does not allow us to ignore meta-level decisions. The heuristics for producing arguments that I have presented are only supposed to be suggestive; but they are not open to the egregious errors about which some have worried. And though the idea is familiar to those who have studied heuristic search, it is somewhat richer because the control of dialectic is more interesting than the deepening of search.

BPS, the Bayesian Problem Solver, applies probabilistic inference and decision-theoretic control to flexible, resource-constrained problem-solving. This paper focuses on the Bayesian inference mechanism in BPS, and contrasts it with those of traditional heuristic search techniques. By performing sound inference, BPS can outperform traditional techniques with significantly less computational effort. Empirical tests on the Eight Puzzle show that after only a few hundred node expansions, BPS makes better decisions than does the best existing algorithm after several million node expansions

Decision-theoretic control of search has previously used as its basic unit. of computation the generation and evaluation of a complete set of successors. Although this simplifies analysis, it results in some lost opportunities for pruning and satisficing. This paper therefore extends the analysis of the value of computation to cover individual successor evaluations. The analytic techniques used may prove useful for control of reasoning in more general settings. A formula is developed for the expected value of a node, k of whose n successors have been evaluated. This formula is used to estimate the value of expanding further successors, using a general formula for the value of a computation in game-playing developed in earlier work. We exhibit an improved version of the MGSS* algorithm, giving empirical results for the game of Othello.

This paper covers two topics: first an introduction to Algorithmic Complexity Theory: how it defines probability, some of its characteristic properties and past successful applications. Second, we apply it to problems in A.I. - where it promises to give near optimum search procedures for two very broad classes of problems.