McCarthy, J., 1977. Epistemological problems of artificial intelligence, IJCAI77, 1038-1044. Reprinted in Readings in Artificial Intelligence (B. L. Webber and N. J. Nilsson, eds.), Los Altos: Morgan-Kaufmann (1981), 459-465.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....A fundamental requirement for autonomous intelligent agents is the ability to reason about causality, which enables the agent to understand the world to an extent su#cient for acting intelligently on the basis of his or her knowledge as to the e#ects of actions. The qualification problem [ McCarthy, 1977 ] in formal theories for reasoning about actions arises from the fact that generally the successful execution of actions depends on many more conditions than we are usually aware # On leave from FG Intellektik, TH Darmstadt. of. The reason for this unawareness is that most conditions are so ....

....in the formal specification of the action start the engine lest the reasoning agent always has to verify this condition before assuming that the action can be successfully executed. Moreover, it is often di# cult if not impossible to even think of all conceivable disqualifications in advance [ McCarthy, 1977 ] Allowing to assume away all so called abnormal disqualifications by default naturally implies that if further knowledge hints at any such unexpected disqualification, then the previous conclusion that the action in question be qualified needs to be withdrawn. Thus the entire process is ....

[Article contains additional citation context not shown here]

J. McCarthy. Epistemological problems of artificial intelligence. In Proc. of the IJCAI, p. 1038--1044, Cambridge, 1977.

....artificial intelligence o#er among others the advantage of a declarative representation of knowledge. Originally, classical logic was designed and used for the representation of static knowledge. More recently logic has also been applied to model actions, states, and changing situations (e.g. [22, 21, 18, 27, 10, 19]) The very first approaches in this direction revealed some fundamental problems such as the frame problem [22] i.e. the problem of how to represent the invariants of a situation with respect to a given action. To handle this problem J. McCarthy, P. Hayes [22] and C. Green [12] introduced frame ....

J. McCarthy. Epistemological problems of artificial intelligence. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1038--1044, 1977.

....would be an exception (John McCarthy, personal conversation) 3.3 Domain Closure Assumption The domain closure assumption [ 18 ] is the assumption that every object in the universe of discourse is representable by a ground term. The related notion of domain circumscription is defined in [ 12 ] and reduced to the (now standard) predicate circumscription in [ 13 ] The idea of this reduction is to introduce a new kind of atomic formulas, GT (x) expressing that x is representable by a ground term, and postulate 8xGT (x) 7) The meaning of GT can be expressed by the axioms GT (x ....

John McCarthy. Epistemological problems of Artificial Intelligence. In Proc. IJCAI-77, pages 1038--1044, 1977. Reproduced in [ 16 ] .

....actions and their effects. The semantics of an action theory is then defined by an entailment relation that determines what will be true after an action sequence is executed from a given state. For several years, the frame problem [12] the ramification problem [6] and the qualification problem [11] have been at the center of RAC s research. In short, the frame problem is the problem of describing, in a concise way, the non effects of actions, i.e. to express what does not change after an action is executed. The ramification problem is concerned with the representation of static domain ....

J. McCarthy. Epistemological problems of artificial intelligence. In Proceedings Int. Joint Conference on Artificial Intelligence, pages 1038--1044. 1977.

....in the external world as its goals demand and its physical abilities permit . If we consider intelligence as a natural kind, it appears like a phenomenon that AI can model. According to this definition we can say that AI is composed of two parts: the epistemological part and the heuristic part [10]. The epistemological part concerns the representation of the world and its modelisation and shows that AI concerns information machines. The heuristic part concerns the methods to find a solution and to decide what to do. The representation of the world is fundamental for understanding what the ....

J. McCarthy, Epistemological problems of artificial intelligence, Proceedings of the Fifth International Joint Conference on Artificial Intelligence, Cambridge, Massachusetts, 1977, p. 1038-1044. 22

.... the closed world assumption [121] is used, logic programming where the negation as failure rule [15] is applied to implement a form of negation, inheritance hierarchies where properties are inherited by default [97] reasoning about action where the frame problem [94] and the qualification problem [91] have to be solved as well as diagnosis [123] are all examples of areas where reasoning patterns not satisfying the monotonicity property are used. As the list illustrates, nonmonotonic reasoning is one of the most fundamental phenomena in knowledge representation and reasoning. However, the exact ....

J. McCarthy. Epistemological problems of artificial intelligence. In pages 1038--1044, Cambridge, MA, USA, August 1977. -- 157 --

....value of multiple models is that they provide heuristic guidance to the search for plausible rules. Hence, though the DNF model is capable of representing all of these rules, it is not helpful for discovering them. In short, it is episternologically adequate but not heuristically adequate (see [19, 20]) Each model directs the attention of the learning system to a small subspace of the space of all possible DNF VL22 rules. The next section shows how the constraints associated with each model are incorporated into special modelfitting induction algorithms. 6. Architecture and Algorithms Section ....

McCarthy, J., Epistemological problems of artificial intelligence, Proceedings of the Fifth International Joint Conference on Artificial Intelligence (1977) 1038--1044.

....that the preconditions for the executability of the action A(x 1 ; x n ) are determined only by the current situation s, not by any other situation. Our decision to focus on sentences of the form (5) for axiomatizing action preconditions stems from the so called qualification problem [23] and its nonmonotonic formulations. In those cases where closed form solutions have been obtained to the associated circumscription policy (e.g. 20] they have been biconditionals of the form (5) Definition 2.3 Successor State Axiom 1. A successor state axiom for an (n 1) ary relational ....

J. McCarthy. Epistemological problems of artificial intelligence. In Proceedings of the Fifth International Joint Conference on Artificial Intelligence, pages 1038--1044, Cambridge, MA, 1977.

....of the database. The example illustrates the subtleties involved in getting the successor state and or transaction precondition axioms to reflect the intent of a state constraint. These difficulties are a manifestation of the so called ramification (Finger [11] and qualification (McCarthy [34]) problems in artificial intelligence planning domains. Transactions might have ramifications, or indirect effects. For the example at hand, the transaction of registering a student in C200 can be viewed as having the direct effect of causing the student to be enrolled in C200, and the indirect ....

J. McCarthy. Epistemological problems of artificial intelligence. In Proceedings of the Fifth International Joint Conference on Artificial Intelligence, pages 1038-- 1044, Cambridge, MA, 1977.

....presented, PDLA processes exceptions. The idea is as follows. Precepts are always assumed to be correct until enough evidence is available to refute them. This idea of taking as facts only what is provided, and only modifying knowledge as necessary is a form of what McCarthy calls circumscription [McCarthy 77, McCarthy 80, McCarthy 86] In PDLA, enough evidence is defined as follows (function Process Exceptions( above) Any node whose num exceptions exceeds half of the amount of input space covered by its stored pair (i.e. the size of the pair) is removed from the network as being incorrect. The ....

38 McCarthy, J. Epistemological Problems of Artificial Intelligence. In Proceedings of the Fifth International Joint Conference on Artificial Intelligence, 1977, 1038-1044.

....and discuss the prospect of its use for knowledge representation, particularly in formalizing reasoning about actions. 1 Introduction and Motivation Circumscription [9,6] is one of the earliest logical formalisms used for representing common sense knowledge. Since it was proposed by McCarthy [8], many extensions of the original circumscription have been proposed. The main 1 A preliminary version of this work appears in the proceedings of Principles of Knowledge Representation and Reasoning Conference KR 96. Preprint submitted to Elsevier Science 4 June 1998 technique used in ....

J. McCarthy. Epistemological problems of Artificial Intelligence. In Proceedings of IJCAI 77, 1038-1044.

....in order to recover from unexpected action failures. 1 Introduction Intelligent agents in open environments inevitably face the Qualification Problem: The executability of an action can never be predicted with absolute certainty; at any time, actions in the real world may surprisingly fail [13]. Yet it would be irrational, and even impossible in general, for a planning agent to foresee all conceivable reasons for an action to go wrong. Rather, a rational agent needs to devise plans under the assumption that the world will behave as expected. On the other hand, being aware of these ....

John McCarthy. Epistemological problems of artificial intelligence. In Proc. of IJCAI, pages 1038--1044, Cambridge, MA, 1977. MIT Press.

.... them The answer to these questions depends on the preferential logic at hand, and we will look at three preferential logics in which the preference relation is based on set inclusion: minimal epistemic logic (see [HM84] minimal temporal epistemic logic ( En95] and domain circumscription ([Mc77], Mc80] For all three logics, we will characterize the class of downward persistent formulae (which will turn out to be non empty) and prove converses to the fact that a downward persistent formula respects the rule of monotonicity: if a formula respects the rule of monotonicity, then it is ....

....(see [Re80a] This was done by giving a translation of default theories into TEL theories. Minimal consequence can then be used to give the sceptical consequences of a default theory. 4 Domain Circumscription (D CIRC) One of the earliest approaches to nonmonotonic reasoning is circumscription ([Mc77], Mc80] Da80] Et88] a preferential logic based on first order predicate logic. Domain circumscription is one of several forms of circumscription. Semantically, in domain circumscription the intuition is that in an intended model of a formula there should not be more elements than necessary. ....

[Article contains additional citation context not shown here]

J. McCarthy, "Epistemological problems of artificial intelligence", in: Proceedings Fifth International Joint Conference on Artificial Intelligence, 1977, pp. 1038-1044

....problem for this general relevance problem. The traditional solution to the frame problem is to get an agent to ignore things which it decides are irrelevant, but this leads to another part of the whole pudding, a closely related epistemological relevance problem, the qualification problem (McCarthy, 1977). McCarthy illustrates the qualification by means of the cannibals and missionaries problem. Normally, of course, the solution to the cannibals and missionaries problem doesn t involve a bridge half a mile upstream although this isn t explicitly excluded by the description of the problem. The ....

McCarthy, J. (1977). Epistemological Problems of Artificial Intelligence. Paper presented at the Fifth International Joint Conference on Artificial Intelligence (IJCAI'77).

....and the event calculus. 2.4 Problems in Reasoning About Actions 2.4. 1 The Frame, Qualification and Ramification Problems In this section, we briefly discuss three important problems that arise in reasoning about actions: the frame problem [McCarthy and Hayes, 1969] the qualification problem [McCarthy, 1977] , and the ramification problem [Finger, 1986] We will use the example introduced in Section 2.3.1 to illustrate the frame problem. Suppose that we are also given that the car was in the garage initially. We can formalize this fact in the situation calculus by the axiom Holds(InGarage; S 0 ) ....

....said to be monotonic if it follows from T fl that T 0 fl for a superset T 0 of T , where T and T 0 are theories and fl is a formula. A consequence relation is is said to be nonmonotonic if it is not monotonic. The main nonmonotonic formalisms that have been proposed are circumscription [McCarthy, 1977; McCarthy, 1980] default logic [Reiter, 1980] nonmonotonic logic [McDermott and Doyle, 1980] and autoepistemic logic [Moore, 1985] Logic programming, which was proposed around 1974, can also be viewed as a nonmonotonic formalism, though the close connection of logic programming to other ....

[Article contains additional citation context not shown here]

John McCarthy. Epistemological problems of artificial intelligence. In Proc. of IJCAI-77, pages 1038--1044, 1977.

....It has been argued that it is undesirable to explicitly write down effect laws 2(A [ff]C) if C is an indirect effect of ff, and that such indirect effects should be derived from the basic effect law 2(A [ff]C 0 ) together with some static laws relating C and C 0 . 3. The qualification problem [46] is that it is difficult to write down executability laws. The reason is that the number of preconditions of an action is huge (e.g. the preconditions of the action of starting a car) Moreover, it may be undesirable or practically infeasible to check the truth of each of them. It is usually ....

J. McCarthy. Epistemological problems of artificial intelligence. In Proc. of the International Joint Conference on Artificial Intelligence (IJCAI'77), pages 1038--1044, 1977.

....a class of fixpoint formulas was characterized which can be reduced into their first order equivalents. In this paper, we extend the previous work in three ways: 1. We define a general form of domain circumscription which subsumes existing domain circumscription proposals in the literature ([10], 2] 6] 7] and [8] We call the generalization general domain circumscription (GDC) GDC distinguishes itself from other proposals in the following manner. When circumscribing the domain of a theory T , it is permitted to vary arbitrary predicates, functions, or constants, to maximize the ....

....which when circumscribed using GDC are guaranteed to be reducible to equivalent first order theories which are constructively generated as output from extended versions of the original DLS algorithm. Included in this class are theories for which both McCarthy s original domain circumscription [10] and Hintikka s mini consequence [7] are always reducible to first order logic. 3. We characterize a class of theories which, when first circumscribed using GDC and then circumscribed using an arbitrary circumscription policy, are guaranteed to be reducible to equivalent first order theories ....

[Article contains additional citation context not shown here]

McCarthy, J. (1977) Epistemological Problems of Artificial Intelligence, in: Proc. 5th IJCAI, Cambridge, MA, 1977, 1038-1044.

No context found.

McCarthy, J., 1977. Epistemological problems of artificial intelligence, IJCAI77, 1038-1044. Reprinted in Readings in Artificial Intelligence (B. L. Webber and N. J. Nilsson, eds.), Los Altos: Morgan-Kaufmann (1981), 459-465.

No context found.

McCarthy, J. (1977). "Epistemological problems of artificial intelligence". In: Proceedings of the Fifth International Joint Conference on Artificial Intelligence, Cambridge, Massachusetts, 1038--1044

No context found.

McCarthy, J.: Epistemological problems of artificial intelligence. In: Proceedings Int. Joint Conference on Artificial Intelligence (1997) 1038--1044

No context found.

J. L. McCarthy. Epistemological problems of artificial intelligence. pages 1038--1044, 1977.

No context found.

John McCarthy. Epistemological problems of artificial intelligence. In IJCAI77 [32].

No context found.

McCarthy, J.: Epistemological problems of artificial intelligence, MIT Press, Cambridge, MA, 1977.

No context found.

John McCarthy. Epistemological problems of artificial intelligence. In Proc. of IJCAI-77, pages 1038--1044, 1977. Reproduced in

No context found.

John McCarthy. Epistemological problems of artificial intelligence. In Proceedings of the Fifth International Conference on Artificial Intelligence (IJCAI '77), pages 1038--1044, 1977.

First 50 documents  Next 50

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