Pearl, J. [1988]. Embracing causality in default reasoning, Artificial Intelligence 35: 259--271.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....steps. The causal and evidential rules warranting the various inference steps will be left implicit. Given the choice for a modus ponens approach, one question immediately arises: must causal knowledge be represented as cause e#ect rules or as e#ect cause rules In terms of Pearl [10], do we want to represent causal knowledge as causal rules or as evidential rules The causal model of Figure 2 suggests that we need both kinds of rules, since we both need to do explanation, i.e. derive causes from effects, and prediction, i.e. derive effects from causes. For instance, we must ....

....a causal rule. More generally, in a modus ponens approach causal knowledge should be represented as follows: each time we want to do a prediction, we must write a causal rule, while each time we want to do an explanation, we must write an evidential rule. However, caution is required here: Pearl [10] shows that if the modus ponens approach is applied to a mixture of causal and evidential rules, not all modus ponens inference steps may be made. Consider the following case PC P where P Q is an evidential rule and the subscript in PC indicates that P has been derived from a causal rule. ....

Pearl, J. [1988]. Embracing causality in default reasoning, Artificial Intelligence 35: 259--271.

....would be minimal, given the di erences in emphasis, and in mathematical machinery. It appears that Ge ner s proposal was inspired by Judea Pearl s investigation of the distinction between causal and non causal grounds in general default reasoning and probabilistic reasoning, as developed in [45] and many subsequent publications. Consideration of the relationship of UCL to this body of work is beyond the scope of this note. In recent years, many researchers have put forward proposals for causal theories of action and change [12,10,4,2,33,38,49,22,48,21,30,39,50,51,18,19,5,16,52] In ....

Judea Pearl. Embracing causality in default reasoning. Arti cial Intelligence, 35:259-271, 1988.

....can be viewed as a noninertial fluent. Other examples of this kind can be found in [ Giunchiglia and Lifschitz, 1998 ] p. 628. 15 6 6 PQ PQ 0; A 0; A PQ P Q 0; A 0 A Figure 2: Transition system described by (13) 17. This view has been developed by several authors, including Pearl [1988], Geffner [1990] Lin [1995] and McCain and Turner [1995] In application to the language B, this understanding of causal laws suggests using the syntax caused L if caused F instead of L if F . 18. This is similar to the principle of universal causation [ McCain and Turner, 1997 ] according ....

Judea Pearl. Embracing causality in default reasoning. Artificial Intelligence, 35:259--271, 1988.

....and causal conditions in standard decision theory has been set right in newer systems of causal decision theory. See, for example, Gibbard and Harper, 1981] and [Lewis, 1986a] 3 it does not follow that the switch can be opened by turning off the light. 4 In the AI literature, Judea Pearl [1988] has emphasized the importance of the distinction between causal and non causal grounds in general default reasoning. Some of the difficulties encountered in using state constraints for determining the indirect effects of actions were recognized by Ginsberg and Smith [1988b] Lifschitz [1990] and ....

Judea Pearl. Embracing causality in default reasoning. Artificial Intelligence, 35:259--271, 1988.

.... assumption based truth maintenance systems ; see (Benferhat et al. 4] More generally it would be interesting to develop a logical framework where it is possible to express both weighted causal rules associating manifestations to disorders and weighted evocation rules (in the sense of Pearl [18]) associating plausible disorders to manifestations, and perform local reasoning tasks with this kind of hybrid knowledge. ....

Pearl J. (1988) Embracing causality in default reasoning. Artificial Intelligence, 35, 259271.

....that whatever follows from the explicitly described effects of an action and the state constraints is an indirect effect. Conceptually, this is a mistake one which rests on a confusion of causal and non causal grounds. Technically, it leads to unintuitive results. In the AI literature, Judea Pearl (1988) has emphasized the importance of the distinction between causal and non causal grounds in general default reasoning. Some of the difficulties encountered in using state constraints for determining the indirect effects of actions were recognized by Ginsberg and Smith (1988) Lifschitz (1990) and ....

Pearl, J. 1988. Embracing causality in default reasoning.

....as a primitive notion, Shoham [ 1990 ] and Iwasaki and Simon [ 1986 ] attempt to derive it from an acausal theory. In particular, Iwasaki and Simon consider deriving the causal relations from a set of acausal equations. It is not clear if this approach can be ported into the situation calculus. Pearl [ 1988 ] argues about the need for a primitive notion of causality in general default reasoning. This paper obviously echoes the same theme. In fact, the title of this paper follows that of [ Pearl, 1988 ] 7 Conclusions We have argued that acausal state constraints like (1) are not adequate for ....

....set of acausal equations. It is not clear if this approach can be ported into the situation calculus. Pearl [ 1988 ] argues about the need for a primitive notion of causality in general default reasoning. This paper obviously echoes the same theme. In fact, the title of this paper follows that of [ Pearl, 1988 ] 7 Conclusions We have argued that acausal state constraints like (1) are not adequate for representing the indirect effects of actions, and proposed a solution using causal rules like (9) By embracing causality, we are able to use only simple nonmonotonic formalisms for solving the frame, ....

Judea Pearl. Embracing causality in default reasoning. Artificial Intelligence, 35:259--271, 1988.

....a collection of single causal relationships, each of which only relates two particular effects, accounts for both several indirect effects caused by a direct one and also indirect effects in turn causing further indirect effects. To illustrate the latter, consider the relationship 3 See [Pearl, 1988a] for a general discussion on the different nature of causal compared to evidential implications. A change of light to light causes a change of light detector to light detector , provided detector activated is true. in addition to the one above. Since we do not expect the designer of a formal ....

Judea Pearl. Embracing causality in default reasoning. Artificial Intelligence Journal, 35(2):259-- 271, 1988.

....Away Keeping track of the dependency or causal structure among events is critical in uncertain reasoning. One fundamental reason is the inherent asymmetry between predictive (or causal) reasoning, from cause to effect, and diagnostic (or evidential) reasoning, from effect to cause. Pearl [9] clearly illustrates this asymmetry with the sprinkler example, depicted in Figure 1. Either A, it rained last night, 1 or B, the sprinkler was on last night, could cause C, the grass is wet. C could in turn cause E, the grass is cold and shiny, as well as F , my shoes are wet. ....

....hypothesis, even though the possibility of simultaneous sprinkling and rain is allowed. a a Rain last Rain last night night b b Sprinkler Sprinkler on on e e Cold and Cold and shiny shiny f f Shoes Shoes wet wet c c Grass wet Grass wet Figure 1: Causal diagram for the sprinkler example [9]. This common and intuitively compelling pattern of reasoning is called explaining away, because one cause explains the observed effect and so reduces the need to invoke other causes. This qualitative pattern of reasoning is entirely compatible with Bayesian inference when probabilistic influences ....

[Article contains additional citation context not shown here]

Judea Pearl. Embracing causality in default reasoning. Artificial Intelligence, 35:259--271, 1988.

....causes(a; s 0 ; stuffy) as wanted. The real issue is that if the vents are blocked, that brings about stuffiness causally, whereas in the other direction, stuffiness only suggests that the vents are blocked evidentially. Causal and evidential implications have different logical behaviours [ Pearl, 1988 ] It does not work to use the same connective to encode both, whether the connective is understood inside a monotonic or a nonmonotonic logic. One could still object that stating causal relationships explicitly is less parsimonious. However this difference is unimportant: computationally an ....

Judea Pearl. Embracing causality in default reasoning. Artificial Intelligence, 35(2):259--271, June 1988.

....0 6j= it rained and Delta 0 [ fsprinkler was ong j= it rained. In fact, we shall see later that causally inconsistent databases are also not uncommon in ATMS implementations of diagnosis systems and Dempster Shafer reasoning, thus leading to counterintuitive results (Laskey Lehner 1989; Pearl 1990). Given the importance of causal consistency, and given the tendency to generate causally inconsistent databases, we shall concern ourselves in this paper with formalizing this notion in order to support do sprinkler was on wet ground is it rained Figure 1: A causal structure. injured ....

....causal network is guaranteed (by satisfying causal independence) to protect us from conclusions that clash with our causal understanding of the domain. The importance of this property is best illustrated by an example that uses ATMSs to implement Dempster Shafer reasoning (Laskey Lehner 1989; Pearl 1990). Specifically, the Dempster Shafer rules, wet ground :7 Gamma it rained and sprinkler was on :9 Gamma wet ground, are typically reasoned about in an ATMS framework by constructing the database, Delta = wet ground a 1 oe it rained sprinkler was on a 2 oe wet ground; and ....

[Article contains additional citation context not shown here]

Pearl, J. 1988a. Embracing causality in default reasoning.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC