L. Birnbaum. Rigor mortis: A response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47(1):57--77, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....i.e. two is a links. The above assumptions are widely spread in many fields, for instance: artificial intelligence (under the label logicism [12, 22] situation semantics [7] cognitive science [11] and human computer interaction [8] They can be criticized from many points of view (e.g. [4, 17, 18, 6]) but they will be useful in the following of this paper. Thus, again, I do not take them as established truths, but as useful work hypotheses: for the sake of brevity, I avoid to analyze the (many) philosophical implications of these issues. 2 In this paper I will not distinguish between ....

L. Birnbaum. Rigor mortis: a response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47:57--77, 1991.

....is nothing wrong with classical logics in representing commonsense knowledge; there is, however, a problem with the assumption that to use logic we have to do deduction. David Poole [30] We agree with Poole and Wang [41] that many of the arguments against logical symbolic AI (e.g. Birnbaum [2]) are actually arguments against a more specific target; i.e. classical deduction. We disagree with (e.g. Birnbaum that the obvious alternative to logical AI is some type of situated functional cognition, the nature of which is yet to be explicated (but is explored in [5, 8] Further, ....

Birnbaum, L., Rigor Mortis: A Response to Nilsson's "Logic and Artificial Intelligence" Artificial Intelligence, 1991. 47: p. 5777.

....of empirical sciences. 17] However, the attempt to apply this idea to the semantic study of natural language is not successful [12] It seems that natural language is too subtle and fluid to be put into the frame of model theoretic semantics. Also, it hardly works for non deductive inferences [2, 10], though there are various attempts to extend the theory into more flexible variations by introducing ideas like possible world and multi valued logic [3, 4, 8, 23] The problems in model theoretic semantics are often used as arguments against strong AI . Actually, Searle s assertion that ....

....mind is semantic in his Chinese room argument [16] is directly based on the assumption that all computerized symbol manipulations are bounded to model theoretic semantics, so uninterpreted symbols are meaningless. Model theoretic semantics has been criticized by many authors for its rigidness [2, 10]. However, without a powerful competitor, the solution is far from clear. As McDermott said: The notation we use must be understandable to those using it and reading it; so it must have a semantics; so it must have a Tarskian semantics, because there is no other candidate. 10] Some people ....

[Article contains additional citation context not shown here]

L. Birnbaum. Rigor mortis: a response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47:57--77, 1991.

.... than assuming that a human analyst will behave in a perfectly rational way to create some initial correct design [12] This line is perused further in the situated cognition literature where it is claimed that it is folly to use context independent symbolic assertions to model human reasoning [3, 10, 28, 50]. While many symbolic AI researchers dispute this claim (e.g. 33, 56] it has some following within the KLB community (e.g. Steels [52] and Clancey [10] Our view is not an extreme situated cognition position. We do not reject KLB and KADS out of a disdain for symbolic logics. Like Poole [46] ....

L. Birnbaum. Rigor mortis: A response to Nilsson's "Logic and Artificial intelligence". Artificial Intelligence, 47:57--77, 1991.

....apply this idea to the semantics of natural language runs up against many problems (Ellis, 1993; Lakoff, 1994; Palmer, 1981) It seems that natural language is too subtle and fluid to be put into the frame of model theoretic semantics. Also, this approach works poorly for non deductive inferences (Birnbaum, 1991; McDermott, 1987) despite various attempts to render the theory more flexible by introducing ideas like possible worlds and multi valued propositions (Carnap, 1950; 3. Language 25 Halpern, 1990; Kyburg, 1992; Zadeh, 1986) The problems with model theoretic semantics are often used as arguments ....

....semantic (Searle, 1980) is directly based on the assumption that all computerized symbol manipulations are intrinsically wedded to model theoretic semantics, so that uninterpreted symbols are by definition meaningless. Model theoretic semantics has been criticized by many authors for its rigidity (Birnbaum, 1991; McDermott, 1987) However, without a powerful competitor, the solution is far from clear. As McDermott said: The notation we use must be understandable to those using it and reading it; so it must have a semantics; so it must have a Tarskian semantics, because there is no other candidate. ....

[Article contains additional citation context not shown here]

Birnbaum, L. (1991). Rigor mortis: a response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47:57--77.

.... a domain independent formal language to represent knowledge, and uses formal rules to capture patterns appearing in human reasoning [37] By naming it a Non Axiomatic Logic , I am trying to show that, from the viewpoint of artificial intelligence, the problems of the traditional symbolic AI [3, 25, 34] are not caused by the ideas like formalization , symbolization , logical inferences , and so on, but by the ideas like axiomatization , computation , binary logics , consistent and complete system , and other concepts that explicitly or implicitly assume the sufficiency of knowledge and ....

L. Birnbaum. Rigor mortis: a response to Nillsson's "logic and artificial intelligence". Artificial Intelligence, 47:57--77, 1991.

....We cannot just assume that a machine which has a structure in memory that corresponds in name to a structure in the designer s conceptualization is sufficient for grasping the concept. The structure must play a role in a network of abilities; it must confer on the agent certain causal powers [Bir91]. Some of these powers involve reasoning: being able to use the structure appropriately in deduction, induction and perhaps abduction. But other powers involve perception and action hooking up the structure via causal mechanisms to the outside world. p.10) From this, and the above discussion, ....

L. Birnbaum. Rigor mortis: A response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47(1):57--77, 1991.

.... develop a precise and theoretically sound theory of reasoning is by studying knowledge representations within the knowledge based systems framework (or even within the logic approach to knowledge representation) and any other approach, not following these lines, is a scruffy type of approach (Birnbaum, 1991). However, the approach developed in this thesis suggests an operational approach to studying reasoning, that is nevertheless rigorous and amenable to analysis. We show that this approach efficiently supports a lot more reasoning than traditional approaches and at the same time it matches our ....

Birnbaum, L. 1991. Response to Nilsson's "logic and artificial intelligence".

.... IsA links) The above assumptions are widely spread in many fields, for instance: artificial intelligence (under the label logicism [17, 32] situation semantics [12] cognitive science [16] and human computer interaction [13] They are criticizable from many points of view (see for instance [7, 25, 27]) but they will be useful in the following of this paper for describing the interaction between a user and an information retrieval system. Thus, I do not take them as established truths, but as useful work hypotheses: for the sake of brevity, I avoid to analyze the (many) philosophical ....

L. Birnbaum. Rigor mortis: a response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47:57-77, 1991. 14

....Hofstadter: Something can be computational at one level, but not at another level. 12] On the contrary, traditional computer systems are Turing machines either globally (from experience to response) or locally (from question to answer) Many arguments proposed against logical AI (for example, [1, 20]) symbolic AI (for example, 6] or AI as a whole (for example, 33] are actually against a more specific target: pure axiomatic systems. Designed as a reasoning system, but not a logicist one [25] NARS actually shares more philosophical opinions with the sub symbolic, or connectionist ....

L. Birnbaum. Rigor mortis: a response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47:57--77, 1991.

....a first order predicate logic approach, where the basic semantics of terms is defined through strict deductive inference (modus ponens) only. This leads to difficulties in capturing the semantics and pragmatics of real world concepts, such as polymorphic and prototypical concepts and relationships [11]. A more flexible, procedural semantics is a way out of this problem, although the lack of an explicitly defined semantics may lead to problems. The most common way to realize knowledge based decision support systems has been, and still is, as rule based systems. Many such systems have been ....

L. Birnbaum, Rigor Mortis: a response to Nilsson's "Logic and artificial intelligence", Artificial Intelligence, 47 (1-3) (1991) 57-77.

No context found.

L. Birnbaum. Rigor mortis: A response to Nilsson's "Logic and artificial intelligence". Artificial Intelligence, 47(1):57--77, 1991.

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