This paper concerns the empirical basis of causation, and addresses the following issues: 1. the clues that might prompt people to perceive causal relationships in uncontrolled observations. 2. the task of inferring causal models from these clues, and 3. whether the models inferred tell us anything useful about the causal mechanisms that underly the observations. We propose a minimal-model semantics of causation, and show that, contrary to common folklore, genuine causal influences can be distinguished from spurious covariations following standard norms of inductive reasoning. We also establish a sound characterization of the conditions under which such a distinction is possible. We provide an effective algorithm for inferred causation and show that, for a large class of data the algorithm can uncover the direction of causal influences as defined above. Finally, we address the issue of non-temporal causation. 1 Introduction The study of causation is central to the understanding of hum...

I introduce a temporal belief-network representation of causal independence that a knowledge engineer can use to elicit probabilistic models. Like the current, atemporal belief-network representation of causal independence, the new representation makes knowledge acquisition tractable. Unlike the atemproal representation, however, the temporal representation can simplify inference, and does not require the use of unobservable variables. The representation is less general than is the atemporal representation, but appears to be useful for many practical applications. 1

This paper summarizes basic concepts and principles that I have found to be useful in dealing with causal reasoning. The paper is written as a companion to a lecture under the same title, to be presented at IJCAI-99, and is intended to supplement the lecture with technical details and pointers to more elaborate discussions in the literature. The ruling conception will be to treat causation as a computational schema devised to identify the invariant relationships in the environment, so as to facilitate reliable prediction of the effect of actions. This conception, as well as several of its satellite principles and tools, has been guiding paradigm for several research communities in AI, most notably those connected with causal discovery, troubleshooting, planning under uncertainty and modeling the behavior of physical systems. My hopes are to encourage a broader and more effective usage of causal modeling by explicating these common principles in simple and familiar mathematical form. Af...

We address practical issues concerning the construction and use of decision-theoretic or normative expert systems for diagnosis. In particular, we examine Pathfinder, a normative expert system that assists surgical pathologists with the diagnosis of lymphnode diseases, and discuss the representation of dependencies among pieces of evidence within this system. We describe the belief network, a graphical representation of probabilistic dependencies. We see how Pathfinder uses a belief network to construct differential diagnoses efficiently, even when there are dependencies among pieces of evidence. In addition, we introduce an extension of the belief-network representation called a similarity network, a tool for constructing large and complex belief networks. The representation allows a user to construct independent belief networks for subsets of a given domain. A valid belief network for the entire domain can then be constructed from the individual belief networks. We also introduce the partition, a graphical representation that facilitates the assessment of probabilities associated with a belief network. Finally, we show that the similarity-network and partition representations made practical the construction of Pathfinder.

Discovering relations among domain variables from data plays an important role in automating and/or assisting the process of constructing domain models and validating existing ones. An important kind of relation is the temporal association between domain variables. While a straightforward application of correlation analysis may be insufficient for uncovering these relationships, we propose an approach that attempts to identify similarities in the patterns across data sets. These patterns enable us to capture the temporal associations among the variables of a model. 1 INTRODUCTION Modeling temporal associations among a set of variables requires an explicit representation of both covariation and temporal precedence among the variables. Building such a dynamic model is more suitable than a static model for the representation of a dynamic system. Nevertheless, the level of difficulty involved in the construction of a dynamic model increases dramatically over the construction of a static ...

I defend the intuition that a cause raises the probability of its eects, by formulating a dependence principle which expresses this intuition and holds up well against counterexamples in the philosophical literature. There are several options for an account of causality, each of which is based on an intuition about causality. For instance: mechanistic account: this is based on the intuition that there is a physical process linking cause and eect; probabilistic: the intuition is that a cause makes its eects more likely; counterfactual: if a cause were to occur then so would its eects; strategic: bringing about a cause is a good way to bring about its eects. explanatory: a cause is a good explanation for its eects. Ideally, an account of causality would reconcile all these intuitions, but this task has so far proved notoriously dicult. One prime problem is a conict between the mechanistic and the probabilistic accounts | there are several apparent examples in th...

Previous studies have demonstrated that encoding a Bayesian network into a SAT-CNF formula and then performing weighted model counting using a backtracking search algorithm can be an effective method for exact inference in Bayesian networks. In this paper, we present techniques for improving this approach for Bayesian networks with noisy-OR and noisy-MAX relations—two relations which are widely used in practice as they can dramatically reduce the number of probabilities one needs to specify. In particular, we present two space efficient CNF encodings for noisy-OR/MAX and explore alternative search ordering heuristics. We experimentally evaluated our techniques on large-scale real and randomly generated Bayesian networks. On these benchmarks, our techniques gave speedups of up to two orders of magnitude over the best previous approaches and scaled up to networks with larger numbers of random variables.

It is now usual to distinguish between different causal relations. The distinction I shall be more specifically interested in here is between generic causality and singular causality. Generic causality is the relation one refers to in (e.g.) "Exposure to asbestos causes cancer " whereas singular causality is the relation one refers to in (e.g.) "Peter's being exposed to asbestos caused him to develop