J. Doyle. Artificial intelligence and rational self-government. Technical Report CMU-CS88 -124, Computer Science Department, Carnegie-Mellon University, Pittsburgh, Pennsylvania, 1988.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....procedures that can mechanize and optimize this process. The capability of a system to reason about its own decision making component in order to estimate the value of continued deliberation has been generally referred to as meta reasoning [Batali, 1986; Davis, 1980; Dean and Boddy, 1988; Doyle, 1988; Genesereth, 1983; Horvitz, 1987; Russell and Wefald, 1989b] Meta reasoning, or reasoning about reasoning, can be used in various ways in order to improve the performance of a system: by selecting the most appropriate base level reasoning procedure in any given situation, by controlling a base ....

....evaluation of alternative computations is an internal problem that is solved in the same way using meta meta reasoning. The difficulty with such a uniform meta level architecture is that any attempt to maximize the expected utility of the agent leads to an infinite regress problem [Batali, 1986; Doyle, 1988; Russell and Wefald, 1989b] This problem arises as a result of optimizing a process that involves self reference and recursive evaluation of internal computations without any justified way to truncate this recursive process and maintain overall optimality. However, limiting the number of ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CMU-CS88 -124, Computer Science Department, Carnegie-Mellon University, Pittsburgh, Pennsylvania, 1988.

.... 1987, Haugh, 1987, Kautz, 1986, Lifschitz, 1987a, Lifschitz, 1987b, Morris, 1988, Pearl, 1988, Shoham, 1988b, Baker, 1989 ] plan interaction (e.g. Sussman, 1973, Tate, 1975, Sacerdoti, 1975, Vere, 1983 ] and meta planning (e.g. Russell and Wefald, 1989, Brooks, 1991, Brooks, 1986, Doyle, 1988, Pollack and Ringuette, 1990, Ingrand and Georgeff, 1990 ] We have not sought to build an optimal planner, not even a state of the art planner; there are manyways to make the plannermore sophisticated. Our aim has been first and foremost to couch planning in a fully time situated framework; ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

.... this strength is already contained in the preferences and or the utility function if we accept its cardinality (which rises more problems than brings solutions, see Pomerol and Barba Romero, 1993, for various aspects of the question and references) Another modelling of intention, suggested by Doyle (1987), consists of translating the strength of intention into priorities. It is patent that priority reasoning has given good results in many intelligent systems, but the distinction between priorities and preferences is not clear. This is similar to the discussion about the respective roles of ....

....but also confidence and the value of the arguments exchanged are also worthwhile to consider. We restrict ourself in this paper to personal, isolated decision. Anyway, goals in AI are unaware of the richness and relevance in that matter of multiattribute preference theory. Actually, the idea of Doyle (1987) is that we should diagnose two states before a decision : the state of the environment and, among all the possible mental attitudes of the agent, the agent s state which is a subset of his mental attitudes. A mental attitude comprises : beliefs, desires and intentions. Beliefs are expressed by ....

Doyle J., 1987, "Artificial intelligence and Rational Self-government", working paper, Pittsburgh Carnegie Mellon University.

....lead to less elegant formalisms, it also surely is requisite if AI is to get fully out of the blocks world and into the real world. This research was supported in part by Army Research Office Award DAAH049510628, National Science Foundation Award IRI 9210906, and National Science Foundation Award IRI 9311988. CONTENTS ii Contents 1 Introduction 1 1.1 Motivation . 1 1.2 Related work . 5 ....

....We use RTA(x; y; i) to state that object x is referred to as y prior to step i. In particular if Names(x; y; j) then RTA(x; y; k) for k j, trta(y; i) is used an abbreviation for: xRTA(x; y; i) the unique thing referred to as y prior to step i , itself a non ambiguous reality term. Figure 16 gives a brief sketch of the evolution of reasoning we have in mind. In the figure we use M , BL, j, and j to abbreviate Married, BrokenLeg, john, and john respectively. Also j1 is used to abbreviate the expression trta( j; 2) i.e. j1 = xRTA(x; j; 2) namely the unique thing referred to as ....

[Article contains additional citation context not shown here]

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

....does not have a mechanism to override its commitment to its current plan. Whether limiting the planner to examining only a subset of possible goal orderings is rational depends on whether the opportunity cost of not considering other possible orderings is offset by the savings in computation time [Doyle, 1988] . The plan generator can also include domain specific methods for generating plans. For the Hero domain, a method was added for inserting a new goal when the currently executing action involves carrying an object from one location to another. An on the way plan is created in which the robot ....

Jon Doyle. Artificial intelligence and rational self-government. Technical Report CMU-CS88 -124, Carnegie Mellon University, March 1988.

....that the criteria given by the different reasoning procedures can not be unified. To clarify the matter, it must be considered that each reasoning criterion provides implicitly a partial preferential ordering of extensions, given a set of axioms because incomparability may arise among extensions [4]. Then, the universal reasoner should provide a global ordering of the extensions, such that the answers to queries are sentences of the most preferred maximal extensions. Doyle argues for an aggregative process to obtain the global ordering. This process should obey the following conditions, ....

Doyle,J. 1988. Artificial Intelligence and Rational Self-government. TR CS-88-124, Carnegie-Mellon University Computer Science Department.

....notion of deliberation that is central to decision theory, which he calls Type I Rationality, and Type II rationality, which takes into account the time cost of deliberation. The need to trade decision quality for timeliness is also been a concern in several other AI projects [Dean and Boddy 1988,Doyle 1988,Horvitz et al. 1989,Russell and Wefald 1991,Dean et al. 1993] 2 The main competing alternative is to effectively eliminate execution time reasoning, by compiling into the agent all decisions about what to do in particular situations [Brooks 1991,Kaelbling and Rosenschein 1990,Schoppers ....

Jon Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie Mellon University, Pittsburgh, PA, 1988.

.... by compiling into the agent all decisions about what to do in particular situations [1, 6, 12] This is an interesting endeavor whose ultimate feasibility remains an open question, but we and others believe that in complex domains, the exclusive use of compilation techniques is impractical [8, 9, 14]. An alternative is to design agents that perform explicit reasoning at execution time, but manage that reasoning by engaging in meta level reasoning. Within the past few years, researchers in AI have provided theoretical analyses of meta level reasoning, often applying decision theoretic notions ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88124, Carnegie Mellon University, Pittsburgh, Pa., 1988.

....to computation at any point, and about which computations will be most effective. This premise is shared by several other researchers in AI, and has been the source of a number of projects that have, until recently, developed more or less independently. Doyle s rational psychology project [10, 11] is based on the idea that computations, or internal state changes, are actions to be chosen like any other actions. He has applied this idea to clarify the notions of belief, intention and learning. Horvitz [24, 25] was also an early AI contributor to the study of rational choice of computation, ....

....that is, we are most concerned with changes to the internal state that are made in the service of selecting the next action in the real world. Other computational actions, such as inductive learning and compilation, should be covered in any general theory of metareasoning (cf. Doyle s analysis [11]) In principle, their utility can be defined in terms of their effect on the complete sequence of future actions, and by replacing single actions by sequences in our analysis we can evaluate this utility; but to attempt to derive practical consequences from such an analysis appears premature at ....

[Article contains additional citation context not shown here]

Doyle, J. (1988) Artificial Intelligence and Rational Self-Government. Technical report no. CMU-CS-88-124, Computer Science Department, Carnegie-Mellon University, Pittsburgh, PA.

....static environment. As I have already noted, dynamic environments stress an agent further, and so amplify the need for control of reasoning. Explicit control of reasoning for agents in dynamic environments has been explored, again primarily using The Uses of Plans 9 the tools of decision theory [5, 16, 18, 34, 35, 39, 61, 62]. In addition to these theoretical investigations, programming systems have been built to support the implementation of explicit meta level reasoning processes. The Procedural Reasoning System (PRS) typifies the approach [23, 24] Using PRS, a system designer can encode both object level and ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie Mellon University, Pittsburgh, PA, 1988. The Uses of Plans 39

....the Seventh Conference on Uncertainty in Artificial Intelligence, Los Angeles, CA, pages 151158. Morgan Kaufmann, San Mateo, CA, July 1991. Also Stanford CS Technical Report KSL 91 33. of that agent s behavior. Recent work by several investigators has addressed such tradeoffs in reasoning systems [Doyle, 1988, Horvitz, 1988, Boddy and Dean, 1989, Russell and Wefald, 1989, Breese and Horvitz, 1990] We constructed the Protos system to experiment with the use of metareasoning procedures to control inference approximation methods [Horvitz et al. 1989a] Protos determines the length of time it should ....

Doyle, J. (1988). Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University.

....best performance. Making control decisions rationally raises the problem of infinite regress, since trying to control the cost of making rational control decisions by means of additional rational control decisions creates a tower of deliberations, each one concerned with the level below (as in (Doyle, 1980)) 11 Thus striking a balance between control and reasoning computations means taking effort expended at all these levels into account. In practice, the deliberative information available at higher levels but unavailable at lower ones vanishes as one ascends the reflective tower, and most ....

....We need to investigate how to integrate these theories in useful ways that recognize that meaning, possibility, utility, and probability must all be evaluated with respect to changing purposes and circumstances. 31 Doyle Acknowledgments This paper is an extended version of an invited talk (Doyle, 1990) presented at AAAI90. I thank Ramesh Patil, Peter Szolovits, and Michael Wellman for reading drafts, Rich Thomason for lending me some of his notes, Tom Dean, Othar Hansson, Eric Horvitz, Barton Lipman, Andrew Mayer, Stuart Russell, Joseph Schatz, and David Smith for valuable discussions, and the ....

Doyle, J. 1988a. Artificial Intelligence and Rational Self-Government. Technical Report CS-88-124, Computer Science Department, Carnegie-Mellon University, Pittsburgh, PA.

.... in the field of artificial intelligence; the fifteenth anniversary of the appearance of the original nonmonotonic logic (McDermott Doyle, 1980) and the tenth and fifth anniversaries (respectively) of the appearance of my mathematical monograph (Doyle, 1983c) and my foundational monograph (Doyle, 1988), from which the present paper derives and upon which it improves, and looking back from these anniversaries has led me to include some personal interpretations of their history in this introduction. c fl1993 Jon Doyle. All rights reserved. Doyle 1.1 Nonmonotonic reasoning Though the reader ....

....that any mental entities were inherently contradictory with any others. At the same time, it followed logic in aiming to describe reasoning agents regardless of computability, in seeking to describe the mind of God as well as the mind of Man. The present exposition also draws on my later monograph (Doyle, 1988), which provided better development of the motivation, better forms for some of the concepts, and better notation. This paper omits treatment of many important topics addressed in (Doyle, 1983c) and (Doyle, 1988) in their numerous sequelae, and to some extent in (Doyle, 1980) We have omitted ....

[Article contains additional citation context not shown here]

Doyle, J. (1988). Artificial intelligence and rational self-government. Tech. rep. CS-88-124, Carnegie-Mellon University Computer Science Department.

No context found.

Doyle, J., 1988a. Artificial intelligence and rational self-government, Pittsburgh: Carnegie Mellon University, Computer Science Department.

....that simultaneously balances mental resources against each other and against physical resources, and ffl Developing logics of preferences that make acquisition and specification of planning preferences convenient. This paper focusses on the latter three issues; for our approach to the first, see (Doyle 1988; 1992) Reason maintenance for replanning Replanning in an incremental and local manner requires that the planning procedures routinely identify the assumptions made during planning and connect plan elements with these assumptions, so that replanning may seek to change only those portions of a ....

....of different reason maintenance systems. But if we are to achieve the efficiency required for revising large plans, reason maintenance must be redesigned to make these choices rationally whenever possible. Accordingly, we developed formal foundations for the theory of rational belief revision (Doyle 1988; 1991) But to really achieve efficiency, the RMS must do more than choose rationally among assumptions in backtracking. It must in addition be much more incremental than the original architectures, which were based on making unbounded (potentially global) optimizing computations that in some ....

Doyle, J. 1988. Artificial intelligence and rational selfgovernment.

....set of goals with the value of achieving another. Consequently, the primary means used to guide the planning and replanning process have been purely structural (how many goals achieved) or probabilistic (how likely the goals are achieved) Recent efforts on decision theoretic planning and acting [8, 9, 11, 19, 31, 37] have attempted to employ utility information, but most planning methods in use remain fairly insensitive to this information. Even in those efforts, the focus is on optimizing the non computational properties of the plan without regard to the computational resources required to do so. Work by ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

....of different reason maintenance systems. But if we are to achieve the efficiency required for revising large plans, reason maintenance must be redesigned to make these choices rationally whenever possible. Accordingly, we developed formal foundations for the theory of rational belief revision [9, 10]. But to really achieve efficiency, the RMS must do more than choose rationally among assumptions in backtracking. It must in addition be much more incremental than the original architectures, which were based on making unbounded (potentially global) optimizing computations that in some cases may ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

....the more traditional context, the unified theory may be transferred back to the case of default inference. To effect the transfer, we view default reasoning as rational inference, basing our treatment on the original presentation of this view in [Doyle 1983a] and on the refined presentation in [Doyle 1987]. In particular, we view default reasoning as rational inference in two ways: taking the purpose or selection of default rules as rationally guided, and taking the meaning or interpretation of default rules as rational choice. We discuss these in turn. 2 Rational evaluation of defaults Thinking ....

.... there, for example, natural applications in the society of mind for the iterative voting schemes of Tiedman and Tullock [1976] See also [Mueller 1979] Acknowledgments Though intended to stand on its own, this paper is an exposition of some of the material contained in two much longer works, [Doyle 1983a, 1987]. I thank Jaime Carbonell, Ronald Loui, Joseph Schatz, Richmond Thomason, and Michael Wellman for valuable comments and ideas. ....

Doyle, J., 1987. Artificial intelligence and rational self-government, Pittsburgh: Carnegie Mellon University, Computer Science Department.

....as the strongest argument for acceptance. Pragmatic efficiency strikes me as too narrow a criterion unless we interpret efficiency in its most general sense as economic rationality taking into account the impact of beliefs on the agent s situation as well as computational costs. My papers [1, 2, 3, 4, 5] provide more details on my view of rational belief and inference, generally taking a coordinate free approach in which one may consider alternative ways of reading beliefs out of mental states. This permits, for example, interpreting specific beliefs and inference in any of the ways discussed ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88124, Carnegie-Mellon University Computer Science Department, 1988.

No context found.

Doyle, J., 1988a. Artificial intelligence and rational self-government, Pittsburgh: Carnegie Mellon University, Computer Science Department.

....always exist. It is not clear that this assumption is necessary, or that the definitions here are the best possible. How to deal with the case in which no alternatives are maximally preferred is an interesting question for future study. It would appear that the notion of rational representation [10, 11], in which consistent subsets of inconsistent preferences are chosen to temporarily represent the inconsistent sets, should play a role here. The generality exhibited in the definitions of rational contraction and revision goes against the usual presupposition of epistemologists that knowing more ....

J. Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

....several authors. The most prominent example is Minsky [40] who explicitly models thinking as the aggregate activity of many small mental agents. In the context of nonmonotonic reasoning, Borgida and Imilienski [3] appeal to committee decision making as a metaphor for default inference, and Doyle [5, 7, 8] presents nonmonotonic reasoning from a group decision theoretic perspective. Related views of thinking can be found in economics, philosophy, and psychology [43, 44, 58, 59] The central tenet of Minsky s Society of Mind [40] is the rejection of the single self viewpoint in favor of a mind made ....

Doyle, J. Artificial intelligence and rational self-government. Technical Report CS-88124, Carnegie-Mellon University Computer Science Department, 1988.

No context found.

Doyle, J., 1988. Artificial intelligence and rational self-government, Pittsburgh: Department of Computer Science, Carnegie Mellon University, TR CMU-CS-88-124.

....reason maintenance systems. But if we are to achieve the efficiency required for revising large plans, reason maintenance must be redesigned to make these choices rationally whenever possible. Accordingly, we have begun to develop formal foundations for the theory of rational belief revision [ Doyle, 1988; Doyle, 1990 ] and are developing techniques for encoding probabilistic and preferential information within the RMS and methods by which the RMS can use this information to backtrack in a rational manner. In this, we build on techniques for qualitative representation of probabilistic ....

Jon Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

....which abstracts away some of the inessential details of the motivation in order to achieve a theory covering a variety of representational systems. A more complete treatment of these ideas, including applications of the formalism to describing some aspects of AI architectures, may be found in [ Doyle, 1988 ] which improves on earlier treatments in [ Doyle, 1983a, Doyle, 1983b ] The full paper will contain a comprehensive treatment of the theory. Rationality and constitutions Constitutions interact with the notion of rationality in several ways. The primary way is that constitutions restrict ....

Jon Doyle. Artificial intelligence and rational self-government. Technical Report CS-88-124, Carnegie-Mellon University Computer Science Department, 1988.

First 50 documents

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