The Distributed Arti cial Intelligence Melting Pot Edmund H. Durfee 1 Dept. of Electrical Engineering and Computer Science University of Michigan Ann Arbor, MI 48109
[email protected]
1 This
e ort has been sponsored, in part, by the National Science Foundation under grant IRI-9010645 and Coordination Theory and Collaboration Technology grant IRI-9015423.
Articial Intelligence (AI) has emphasized building \stand-alone systems" that can solve problems with minimal help from other systems (computer or human). These systems have traditionally been brittle, in the sense that they fail miserably when presented with problems even slightly outside of their limited range of expertise. The predominant AI answer to brittleness is to inject more knowledge into a system, including commonsense knowledge, to enlarge the system's range of capabilities. While in the short term such a strategy can overcome particular system limitations, it is in fact a very short-sighted strategy. A more powerful, extensible strategy for overcoming the inherent bounds of intelligence present in any nite AI (or natural) system 64, 83, 84] is to put the system in a society of systems, so that it can draw on a diverse collection of expertise and capabilities in the same way that people overcome the limitations of individuals by banding together into societies that are designed to accomplish what individuals cannot. The ability to exibly team up and coordinate group activities toward individual and collective goals is a hallmark of intelligence. Research in Distributed Articial Intelligence (DAI) concentrates on understanding the knowledge and reasoning techniques needed for intelligent coordination, and on embodying and evaluating this understanding in computer systems. In 1981, the Honorary Guest Editor for this special issue, Professor B. Chandrasekaran, guest edited a special issue of IEEE Transactions on Systems, Man, and Cybernetics that played a critical role in dening the eld of DAI. That special issue collected together articles that reected the seminal ideas in the eld and asked many of the questions that have driven DAI research since. Now, about ten years later, many of those questions have been answered, only to be replaced by new questions. A number of the contributing authors to the 1981 special issue remain active in the eld today, of whom three, Mark Fox, Carl Hewitt, and Victor Lesser, have co-authored papers in this special issue. They have been joined by a new and growing crop of researchers who, for the most part, have stumbled onto DAI from various directions, ranging from distributed computing systems to discourse analysis, from formalisms for representing nested beliefs in agents to cognitive studies of human performance in organizations, from solving inherently distributed problems in applications such as communication network managment to analyzing the evolution of cooperation in populations of articial systems. The wealth of the eld of DAI lies in its interdisciplinary nature, which brings together people with widely dierent perspectives who share a common goal of realizing in computers many of the social capabilities that we take for granted in humans. By refering to DAI as a \melting pot" in the title of this introduction, therefore, I am trying to emphasize the meeting of minds and cultures that takes place among DAI researchers, as well as among DAI systems (in a much simpler way). The challenges of bringing diverse research threads together as part of a concerted DAI endeavor are the same challenges that DAI researchers face in trying to get dierent intelligent systems to work as a team. As a result, DAI researchers must literally practice what they preach.
Contents The goal of this special issue is to capture a small subset of the variety of current work in the eld and to present it in a broad forum|the forum that helped give birth to the eld in the rst place. The contents of this issue should be viewed as representative of DAI,
1
but by no means comprehensive. A comprehensive collection of \classic" DAI papers has been generated by Bond and Gasser 6], and smaller collections of more current eorts have also been published 32, 45]. Surveys of the eld have occurred within these collections and elsewhere 20, 23, 56]. I will not attempt to encapsulate such a survey in this introduction to the issue. However, because the papers for this special issue cover most aspects of the eld, I will use my brief descriptions of each of these papers as an opportunity to point to related work, so as to provide a limited but timely bibliography for the eld. The paper \Distributed Automated Reasoning: Issues in coordination, cooperation, and performance," by MacIntosh, Conry, and Meyer presents an investigation into distributed theorem proving. In this work, each theorem proving agent begins with a subset of the initial axioms, and no agent begins with sucient information to prove a target theorem alone. Given that the agents are loosely-coupled and communicate infrequently, the agents need to be able to decide when they need help from each other, and what help to ask for. Questions of when and where to share information are commonplace in DAI 12, 37, 39, 40, 77, 78, 79, 87]. MacIntosh et al describe heuristics for determining whether progress toward proving a theorem is being made, and methods for formulating requests for more information. The relative simplicity of the theorem proving algorithms (compared to many other AI algorithms) allows the authors to thoroughly analyze the experimental performance of their system. They have systematically identied phenomena that have generally been only intuitively recognized in DAI, such as how duplicate data among agents degrades performance because the agents redundantly explore overlapping areas of the search space. Zlotkin and Rosenschein, in their paper \Cooperation and Conict Resolution via Negotiation Among Autonomous Agents in Non-Cooperative Domains," also employ formal methods for investigating coordination among agents. Their approach takes a decision-theoretic, probablistic view of decisionmaking in multiagent environments 24, 36, 50], where each agent is acting so as to maximize its expected utility. By considering the worths of their goals and the costs of achieving them, the interaction between agents can be characterized as requiring cooperation, compromise, or conict. Moreover, by utilizing probablistic methods, the formalisms the authors develop reveal opportunities to negotiate semi-cooperative deals even in conicting situations. The notion of semi-cooperative deals subsumes other types of deals, and forms the core of the authors' proposed Unied Negotiation Protocol that provides a formal framework for studying negotiation 18, 51, 80, 88]. DAI research on negotiation and coordination has generally emphasized the need for each agent in a multiagent environment to modify what it does to improve group performance. This view typically assumes that the set of interacting agents is xed, and so better coordination requires the agents to adapt appropriately. However, an alternative way to improve group performance is to vary the membership of the group. These approaches are analogous to two ways that many human groups, such as sports teams, improve their performance: through training (a xed collection of agents practices together) and through trading (some group members are replaced to improve team composition). The paper \Controlling Chaos in Distributed Systems" by Hogg and Huberman represents a contribution toward the latter approach. In their work, the authors treat the multiagent environment as an ecosystem, where the composition of successive populations of agents is biased such that agents that perform well in one generation are more likely to propagate to the next. The authors' approach diers from traditional DAI because agents in their system do not individually reason 2
and modify their behaviors they are not necessarily \intelligent" as individuals. Nonetheless, many issues and observations they describe are similar to mainstream DAI. An example is their identication of the need for diversity to ensure robust group performance, which is reminiscent of the need for agent skepticism found in more mainstream DAI research 16]. The contributions of this paper to DAI, therefore, include the rigorous development of an alternative, ecological perspective, and the idea of modifying populations rather than individuals (what I above called \trading" rather than \training") with implications for learning in DAI systems 82]. The paper by Mazer entitled \Reasoning About Knowledge to Understand Distributed AI Systems" also develops a rigorous treatment of distributed computing systems, but its purpose is to improve our understanding of two early approaches that have continued to be inuential in DAI: Contract-Net 86, 87] and blackboards 25, 69]. Mazer points out that most DAI research has focused on identifying, elaborating, and extending methods that lead to coordination between agents, with an emphasis toward experimentally measuring the agents' collective performance. But, Mazer contends, the ability of agents to make decisions that contribute to coherent group activity depends on the knowledge that the agents have. Methods for coordination thus should promote the transfer of appropriate knowledge to relevant agents. Using a temporal, epistemic logic to characterize changing knowledge among distributed systems, Mazer rigorously investigates both the contract-net protocol and blackboard systems, characterizing their knowledge requirements, the role of communication in meeting these requirements, and their limitations. Mazer thus provides tools for explicitly relating research on theories of communication and intention among natural and articial systems 10, 11, 12, 13, 39, 41, 58, 71, 73, 85, 91] with established DAI paradigms. In his invited paper, \A Retrospective View of FA/C Distributed Problem Solving," Lesser also looks back at past DAI research specically, his research on the FunctionallyAccurate, Cooperative (FA/C) paradigm 54]. In overly simple terms, the FA/C paradigm views distributed problem solving as an iterative exchange of tentative partial solutions among agents. By tolerating inconsistent and incorrect information from each other, the agents combine and propagate partial solutions to eventually converge on overall solutions. While FA/C makes intuitive sense, critical questions have cropped up about the nature of inconsistent, concurrent activities, about how to tolerate and resolve inconsistent and incorrect information, and about how to control the formation and propagation of partial solutions to bound the combinatorial complexity of distributed problem solving. In the years since he and Corkill rst introduced the FA/C paradigm, Lesser has guided the development of increasingly complete answers to these questions. In his paper, he provides a more general characterization of FA/C in terms of distributed search (which subsumes the more specic distributed interpretation tasks 57, 92] to which FA/C has typically been applied). He also traces the development of techniques for controlling cooperation and communication among the agents 15, 22]. Finally, he charts some current and future directions for FA/C research 8]. The process of deciding how to coordinate can itself be seen as a distributed problem that agents must solve. For solving coordination problems, DAI researchers have studied approaches involving organizational design 16, 27, 29, 33, 47, 60, 62], and multiagent planning 3, 7, 14, 19, 26, 34, 35, 48, 53, 89, 76], but have treated organizations and plans dierently. In \Coordination as Distributed Search in a Hierarchical Behavior Space," Durfee and Mont3
gomery hypothesize that organizations and plans (as well as schedules) are the same type of entity but dier in their specicity and duration. Building from this hypothesis, they outline the theory that they are developing in which organizing, planning, and scheduling are treated as search problems within the same search space. Besides this encompassing search space, their approach involves a distributed search protocol, local search algorithms, metrics, and control knowledge. They have used their theory to combine concepts from several DAI approaches into a prototype implementation which illustrates the need to balance the level of detail at which coordination is worked out with the costs of coordinating at more detailed levels. Moreover, they show how the same protocol and algorithms that coordinate agents' plans also lead to establishing organizations, providing a possible foundation for an interdisciplinary study of coordination 61]. Recent research in understanding cooperative design and developing knowledge-based systems that collaborate to design a complex artifact has highlighted issues in resolving conicts between design criteria such as cost versus quality 5, 52, 90]. The paper by Klein entitled \Supporting Conict Resolution in Cooperative Design Systems," takes the position that general principles of conict resolution exist, and can be embodied in a computational theory. Klein's work separates generic strategies for conict resolution from domain knowledge about the design problem. When dierent designers' decisions come into conict, the generic conict resolution techniques map the conict situation into a conict class, and instantiate conict resolution plans based on the general advice associated with the conict class. This type of approach is not only useful in resolving conicts among AI systems that are collectively designing an artifact, but they also can be useful support tools for human collaborators. Klein outlines a collaboration support system, through which human designers are alerted to conicts between their design decisions, and are provided with suggestions as to strategies for resolving the conicts. An eort in building computer tools that support human collaboration is also described by Pan and Tenenbaum in their paper entitled \An Intelligent Agent Framework for Enterprise Integration." Considering that many tasks in complex human enterprises are distributed in space and time, Pan and Tenenbaum are developing a computer infrastructure in which a large number of fairly simple agents work together to perform important support tasks|particularly the more routine, well-dened tasks|in an organization. Some of these \intelligent agents" are also designed to act as the portal between people and the computer infrastructure. The authors stress pragmatic issues in making such an infrastructure acceptable to people, such as making each agent understandable to someone who uses the system (and not just to the programmer that built the agent). Because they understand the organizational needs and routines best, system end-users should be able to build and modify agents directly 63]. But while Pan and Tenenbaum have been developing their framework from the \bottom-up" using relatively simple agents initially, their work has led them into the same scientic arena as researchers that have been investigating the use of AI and DAI to support human organizations 9, 17, 46, 61, 70, 74, 75, 81]. In their invited paper entitled \DAI Betwixt and Between: From `intelligent agents' to open systems science," Hewitt and Inman also emphasize the critical role that DAI should play in bridging the gap between what mainstream AI technology can provide for enhancing personal productivity and what open systems science provides for supporting dynamic, networked organizations. Tracing the evolution of their approach from the creation of Actors 4
43, 49] through the development of Open Systems Science 44], the authors draw on a rich body of concepts, ranging from sociology to the mathematics of concurrency. In applying their results to large-scale, complex, organizational computing systems, however, the authors have recognized that Actors must be made scalable. In their ongoing work, they are developing Organizations of Restricted Generality (ORGs) to meet this need. As Gasser has also pointed out 30], the authors note that DAI must embrace social and anthropological concepts if it is to bring AI ideas into human organizations. The Actors formalism has spawned a number of research eorts directed not so much toward addressing the fundamental questions of DAI (such as what knowledge is needed to coordinate well and how should that knowledge be used), but instead toward building tools for implementing DAI systems. Ferber and Carle, in their paper entitled \Actors and Agents as Reective Concurrent Objects: A MERING IV perspective," describe such an eort. The authors point out that other Actor languages have violated aspects of the Actor model in order to make them easier to use and implement, but argue that such violations can be made unnecessary by introducing reection into the language. That is, they introduce meta-level constructs within the language that can be used to tailor the behaviors of basic contructs. The authors also give examples of how their language can support the implementation of DAI techniques for identifying what agents can do and assigning tasks among them. This paper thus presents an eort in developing tools and languages for building DAI systems 4, 31, 42, 59] and for experimenting with DAI systems 55, 67, 68] DAI agents need facilities for representing, maintaining, and propagating what they believe about the world. What makes building such facilities dicult is that slow and potentially errorful communication among the agents makes converging on \common knowledge"| where everyone knows that everyone knows that ... everyone knows fact x|impossible to guarantee 41]. Instead, each agent must maintain a network of what it believes and should revise this network as information propagates among the agents. In \Multiagent Truth Maintenance," Huhns and Bridgeland extend ideas from truth maintenance for use in multiagent domains. Given that achieving globally consistent and well-founded beliefs is problematic (it would imply common knowledge), they dene Local-and-Shared Consistency and WellFoundedness, and develop algorithms to achieve it. Not only do their investigations draw on and extend the small amount of research to date on distributed truth maintainance (and distributed constraint satisfaction in general) 21, 65, 66, 93], but maintaining distributed beliefs is a key technology needed for integrating heterogeneous AI systems within the authors RAD system and elsewhere 2]. In \Distributed Constrained Heuristic Search," Sycara, Roth, Sadeh, and Fox describe an architecture for solving distributed search problems, using both heuristics and constraint satisfaction techniques. Their research has been driven, in large part, by realistic problems in the domain of manufacturing 28, 72]. To deal with the large search space in this and similar domains, they have developed the notion of problem textures (such as contraint tightness and variable goodness) that they use to heuristically order possible choices. Using these heuristics, and an algorithm called Distributed Asynchronous Backjumping to recover from inconsistent distributed search decisions, they have implemented and extensively evaluated their approach. Conry, Kuwabara, Lesser, and Meyer also report on techiques for solving a type of distributed constraint satisfaction problem in their paper entitled \Multistage Negotiation for 5
Distributed Constraint Satisfaction." The specic distributed constraint satisfaction problem they address in their paper involves decentralized management of a communication network 1, 38]. In particular, because a dedicated end-to-end circuit between two points can take numerous routes through the network, nding the proper allocation of channel resources to end-to-end circuits involves searching through the space of alternative assignments, detecting and resolving constraint violations (two circuits cannot share the same channel). Their approach involves a tentative exchange of resource assignments among managers (in a Functionally Accurate/Cooperative manner), so that managers can converge on assignments with minimal exchanges of information. In addition, through careful formalization of the problem, the authors describe techniques for detecting and recovering from overconstrained situations (where all connections cannot be established).
Themes and Directions While my description of the papers in the issue should give a sense of the breadth of research going on, I hope that it also helps show that each paper is related to others, especially those that are physically near it in this collection. The set of papers begins with work on formal models of decisionmaking and of distributed systems, and ows into more formal analyses of past procedural approaches that have been based on social metaphors. The papers build from there into more elaborate couplings of concepts from AI and social science, bringing in issues of supporting human cooperative activity and languages for building distributed intelligent toolsets. Because decisionmaking in large distributed enterprises requires propagating beliefs and constraints, the collection of papers concludes with investigations into treating distributed problem solving as a form of distributed constraint satisfaction. The theme I see underlying all of these papers is that of unifying diverse theories and techniques. For years, DAI researchers have borrowed ideas from dierent elds and examined the capabilities of each idea essentially in isolation. In the special issue ten years ago, many of the papers strove to characterize the eld by compartmentalizing techniques along lines such as completely accurate versus functionally accurate, nearly autonomous versus cooperative, and task sharing versus result sharing. In this issue, the papers are instead striving to highlight commonalities and generalities. As a few examples, the papers talk about unied negotiation protocols about distributed problem solving and coordination as propagation of data and knowledge about common representations and algorithms for organizations, plans, and schedules about generic conict resolution knowledge about the central role of DAI in organizational computing systems and about distributed constraint satisfaction as a general paradigm for many types of distributed problem solving. The emphasis on articulating general principles of coordination, based on the view of cooperative problem solving as distributed search, appears to me to represent a major part of the DAI research agenda for the upcoming years. Although I believe it has emerged independently, DAI's concern with unied theories of coordination seems to be the multiagent parallel of the currently active AI interest in unied theories of cognition. Regardless of the success or failure of developing accepted theories, pursuing them can be an instructive and illuminating exercise. In addition, the practical utility of such theories, as well as of the separate coordination techniques developed in the past, will be increasingly of interest.
6
While DAI research has typically had an experimental emphasis in which new techniques are applied to idealized versions of realistic problems, there is increasing pressure to move from prototypes to real applications. Work toward this objective, involving applications such as computer-supported cooperative work and job-shop scheduling, is described in some of the papers in this issue.
Acknowledgements I have received a tremendous amount of assistance in guest editing this special issue. First and foremost, B. Chandrasekaran has stood behind this endeavor from the very beginning. As Honorary Guest Editor, he has provided me with invaluable advice throughout this entire process|advice on handling authors, on nding and \nudging" reviewers, on establishing criteria for assigning reviewers and making decisions from reviews. Chandra has also played a \hands on" role in the process, including soliciting and forwarding anonymous reviews for papers where I might not be suciently unbiased. Despite his many other commitments, Chandra always found time to answer my mail and follow up my requests. I would also like to thank Andrew Sage, editor-in-chief of these transactions, for giving me the opportunity to bring this collection of papers together, and for being exible about deadlines. Given that the substantial number of papers accepted for this issue represent less than a third of the papers submitted, obviously many reviewers directly contributed to the selection process for this issue. I am indebted to these reviewers, who remain anonymous (some even to me), for the hours of time they spent evaluating the papers and making helpful suggestions to the authors. Finally, I would like to thank the authors of the papers for all of their eorts, often under time constraints, in writing papers that convey the diversity and excitement of the eld. As I said before, DAI researchers must practice what they preach I am grateful to be part of this cooperative distributed problem-solving network.
References 1] Mark R. Adler, Alvah B. Davis, Robert Weihmayer, and Ralph Worrest. Conict-resolution strategies for nonhierarchical distributed agents. In Michael N. Huhns and Les Gasser, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 2] Mark R. Adler and Evangelos Simoudis. Integrating distributed expertise. In Proceedings of the 1990 Distributed AI Workshop, Bandara, Texas, October 1990. 3] C. Applegate, C. Elsaesser, and J. Sanborn. An architecture for adversarial planning. IEEE Transactions on Systems, Man, and Cybernetics, SMC-20(1):186{194, January 1990. 4] R. Bisiani, F. Alleva, A. Forin, R. Lerner, and M. Bauer. The architecture of the agora environment. In Michael N. Huhns, editor, Distributed Arti cial Intelligence, Research Notes in Articial Intelligence, chapter 4, pages 99{118. Pitman, 1987.
7
5] Alan H. Bond. The cooperation of experts in engineering design. In Michael N. Huhns and Les Gasser, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 6] Alan H. Bond and Les Gasser. Readings in Distributed Arti cial Intelligence. Morgan Kaufmann Publishers, San Mateo, CA, 1988. 7] Stephanie Cammarata, David McArthur, and Randall Steeb. Strategies of cooperation in distributed problem solving. In Proceedings of the Eighth International Joint Conference on Arti cial Intelligence, pages 767{770, Karlsruhe, Federal Republic of Germany, August 1983. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 102{105, Morgan Kaufmann, 1988.). 8] Norman Carver, Zarko Cvetanovic, and Victor Lesser. Sophisticated cooperation in FA/C distributed problem solving systems. In Proceedings of the Ninth National Conference on Arti cial Intelligence, July 1991. 9] Ernest Chang. Participant systems. In Michael N. Huhns, editor, Distributed Arti cial Intelligence, Research Notes in Articial Intelligence, chapter 11, pages 311{339. Pitman, 1987. 10] P. R. Cohen and H. J. Levesque. Rational interaction as the basis for communication. In P. R. Cohen, J. Morgan, and M. E. Pollack, editors, Intentions in Communication. MIT Press, 1990. 11] Philip R. Cohen and Hector J. Levesque. Intention = choice + commitment. In Proceedings of the Sixth National Conference on Arti cial Intelligence, pages 410{415, July 1987. 12] Philip R. Cohen and C. Raymond Perrault. Elements of a plan-based theory of speech acts. Cognitive Science, 3(3):177{212, 1979. 13] R. Conte, M. Miceli, and C. Castelfranchi. Limits and levels of cooperation: Disentangling various types of prosocial interaction. In Y. Demazeau and J.-P. Muller, editors, Decentralized AI, volume 2. North Holland, 1991. 14] Daniel D. Corkill. Hierarchical planning in a distributed environment. In Proceedings of the Sixth International Joint Conference on Arti cial Intelligence, pages 168{175, Cambridge, Massachusetts, August 1979. (An extended version was published as Technical Report 7913, Department of Computer and Information Science, University of Massachusetts, Amherst, Massachusetts 01003, February 1979.). 15] Daniel D. Corkill and Victor R. Lesser. The use of meta-level control for coordination in a distributed problem solving network. In Proceedings of the Eighth International Joint Conference on Arti cial Intelligence, pages 748{756, Karlsruhe, Federal Republic of Germany, August 1983. (Also appeared in Computer Architectures for Arti cial Intelligence Applications, Benjamin W. Wah and G.-J. Li, editors, IEEE Computer Society Press, pages 507{515, 1986). 16] Daniel David Corkill. A Framework for Organizational Self-Design in Distributed Problem Solving Networks. PhD thesis, University of Massachusetts, February 1983. (Also published as Technical Report 82-33, Department of Computer and Information Science, University of Massachusetts, Amherst, Massachusetts 01003, December 1982.).
8
17] W.B. Croft and L.S. Lefkowitz. Knowledge-based support of cooperative activities. In Proceedings of the Twenty- rst Annual Hawaii International Conference on System Sciences, volume 3, pages 312{318, 1988. (Published by IEEE Computer Society Press, Catalog Number 88TH0213-9.). 18] Randall Davis and Reid G. Smith. Negotiation as a metaphor for distributed problem solving. Arti cial Intelligence, pages 63{109, 1983. 19] Keith Decker and Victor Lesser. Extending the partial global planning framework for cooperative distributed problem solving control. In Proceedings of the 1990 DARPA Workshop on Innovative Approaches to Planning, Scheduling, and Control, pages 396{407, November 1990. 20] Keith S. Decker. Distributed problem-solving techniques: A survey. IEEE Transactions on Systems, Man, and Cybernetics, 17:729{740, 1987. 21] Jon Doyle and Michael P. Wellman. Rational distributed reason maintenance for planning and replanning of large-scale activities (preliminary report). In Proceedings of the 1990 DARPA Workshop on Innovative Approaches to Planning, Scheduling, and Control, pages 28{36, November 1990. 22] Edmund H. Durfee and Victor R. Lesser. Partial global planning: A coordination framework for distributed hypothesis formation. IEEE Transactions on Systems, Man, and Cybernetics, 21(5):1167{1183, September 1991. (Special Issue on Distributed Sensor Networks). 23] Edmund H. Durfee, Victor R. Lesser, and Daniel D. Corkill. Cooperative distributed problem solving. In Avron Barr, Paul R. Cohen, and Edward A. Feigenbaum, editors, The Handbook of Arti cial Intelligence, volume IV, chapter XVII, pages 83{137. Addison-Wesley, 1989. 24] Kaan Egilmez and Steven H. Kim. Coordination of multiple agents through modal state logic and inuence specications. Technical report, MIT, Laboratory for Manufacturing and Productivity, 1990. 25] Lee D. Erman, Frederick Hayes-Roth, Victor R. Lesser, and D. Raj Reddy. The HearsayII speech-understanding system: Integrating knowledge to resolve uncertainty. Computing Surveys, 12(2):213{253, June 1980. 26] Nicholas V. Findler and Ron Lo. An examination of distributed planning in the world of air trac control. Journal of Parallel and Distributed Computing, 3:411{431, 1986. 27] Mark S. Fox. An organizational view of distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 11(1):70{80, January 1981. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 140{150, Morgan Kaufmann, 1988.). 28] Mark S. Fox. Constraint-directed Search: A Case Study of Job-Shop Scheduling. Morgan Kaufmann Publishers, 1987. 29] L. Gasser. The integration of computing and routine work. ACM Transactions on Oce Information Systems, 4(3):205{225, July 1986. 30] Les Gasser. Social conceptions of knowledge and action: DAI foundations and open systems semantics. Arti cial Intelligence, 47(1-3):107{138, 1991.
9
31] Les Gasser, Carl Braganza, and Nava Herman. MACE: A exible testbed for distributed AI research. In Michael N. Huhns, editor, Distributed Arti cial Intelligence, Research Notes in Articial Intelligence, chapter 5, pages 119{152. Pitman, 1987. 32] Les Gasser and Michael N. Huhns, editors. Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 33] Les Gasser, Nicolas Rouquette, Randall W. Hill, and John Lieb. Representing and using organizational knowledge in DAI systems. In Les Gasser and Michael N. Huhns, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence, pages 55{78. Pitman, 1989. 34] Michael George. Communication and interaction in multi-agent planning. In Proceedings of the Third National Conference on Arti cial Intelligence, pages 125{129, Washington, D.C., August 1983. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 200{204, Morgan Kaufmann, 1988.). 35] Michael George. A theory of action for multiagent planning. In Proceedings of the Fourth National Conference on Arti cial Intelligence, pages 121{125, Austin, Texas, August 1984. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 205{209, Morgan Kaufmann, 1988.). 36] Piotr J. Gmytrasiewicz, Edmund H. Durfee, and David K. Wehe. A decision-theoretic approach to coordinating multiagent interactions. In Proceedings of the Twelfth International Joint Conference on Arti cial Intelligence, August 1991. 37] Piotr J. Gmytrasiewicz, Edmund H. Durfee, and David K. Wehe. The utility of communication in coordinating intelligent agents. In Proceedings of the Ninth National Conference on Arti cial Intelligence, pages 166{172, July 1991. 38] S. Goyal and R. Worrest. Expert system applications to network management. In J. Leibowitz, editor, Expert System Applications to Telecommunications, volume 1, pages 3{44. John Wiley & Sons, 1988. 39] B. J. Grosz and C. Sidner. Plans for discourse. In P. R. Cohen, J. Morgan, and M. E. Pollack, editors, Intentions in Communication. MIT Press, 1990. 40] Barbara J. Grosz and Candace L. Sidner. Discourse structure and the proper treatment of interruptions. In Proceedings of the Ninth International Joint Conference on Arti cial Intelligence, pages 832{839, August 1985. 41] Joseph Y. Halpern and Yoram Moses. Knowledge and common knowledge in a distributed environment. In Third ACM Conference on Principles of Distributed Computing, 1984. 42] Frederick Hayes-Roth, Lee D. Erman, Scott Fouse, Jay S. Lark, and James Davidson. ABE: A cooperative operation system and development environment. In Mark Richer, editor, AI Tools and Techniques. Ablex Publishing Corporation, 1988. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 457{489, Morgan Kaufmann, 1988.). 43] Carl Hewitt. Viewing control structures as patterns of passing messages. Arti cial Intelligence, 8(3):323{364, Fall 1977.
10
44] Carl Hewitt. Oces are open systems. ACM Transactions on Oce Information Systems, 4(3):271{287, July 1986. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 321{330, Morgan Kaufmann, 1988.). 45] Michael Huhns, editor. Distributed Arti cial Intelligence. Morgan Kaufmann, 1987. 46] Michael N. Huhns, Uttam Mukhopadhyay, Larry M. Stephens, and Ronald D. Bonnell. DAI for document retrieval: The MINDS project. In Michael N. Huhns, editor, Distributed Arti cial Intelligence, Research Notes in Articial Intelligence, chapter 9, pages 249{284. Pitman, 1987. 47] Toru Ishida, Makoto Yokoo, and Les Gasser. An organizational approach to adaptive production systems. In Proceedings of the Eighth National Conference on Arti cial Intelligence, pages 52{58, July 1990. 48] Kurt Konolige. A deductive model of belief. In Proceedings of the Eighth International Joint Conference on Arti cial Intelligence, pages 377{381, Karlsruhe, Federal Republic of Germany, August 1983. 49] William A. Kornfeld and Carl E. Hewitt. The scientic community metaphor. IEEE Transactions on Systems, Man, and Cybernetics, SMC-11(1):24{33, January 1981. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 311{320, Morgan Kaufmann, 1988.). 50] Sarit Kraus and Jonathan Wilkenfeld. The function of time in cooperative negotiations. In Proceedings of the Twelfth International Joint Conference on Arti cial Intelligence, August 1991. 51] Brigitte Laasri, Hassan Laasri, and Victor R. Lesser. Negotiation and its role in cooperative distributed problem solving. In Proceedings of the 1990 Distributed AI Workshop, October 1990. 52] Susan E. Lander, Victor R. Lesser, and Margaret E. Connell. Knowledge-based conict resolution for cooperation among expert agents. In D. Sriram, R. Logher, and S. Fukuda, editors, Computer-Aided Cooperative Product Development. Springer Verlag, 1991. 53] Amy L. Lansky. Behavioral specication and planning for multiagent domains. Technical Report 360, SRI International, Menlo Park CA, 1985. 54] Victor R. Lesser and Daniel D. Corkill. Functionally accurate, cooperative distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, SMC-11(1):81{96, January 1981. 55] Victor R. Lesser and Daniel D. Corkill. The Distributed Vehicle Monitoring Testbed: A tool for investigating distributed problem solving networks. AI Magazine, 4(3):15{33, Fall 1983. (Also published in Blackboard Systems, Robert S. Engelmore and Anthony Morgan, editors, pages 353{386, Addison-Wesley, 1988 and in Readings from AI Magazine: Volumes 1{5, Robert Engelmore, editor, pages 69{85, AAAI, Menlo Park, California, 1988). 56] Victor R. Lesser and Daniel D. Corkill. Distributed problem solving. In Encyclopedia of Arti cial Intelligence, pages 245{251. John Wiley & Sons, 1987. 57] Victor R. Lesser and Lee D. Erman. Distributed interpretation: A model and experiment. IEEE Transactions on Computers, C-29(12):1144{1163, December 1980. (Also published
11
58] 59] 60] 61] 62]
63] 64] 65] 66] 67] 68] 69] 70] 71]
in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 120{139, Morgan Kaufmann, 1988.). Hector J. Levesque, Philip R. Cohen, and Jose H. T. Nunes. On acting together. In Proceedings of the Eighth National Conference on Arti cial Intelligence, pages 94{99, July 1990. Douglas J. MacIntosh and Susan E. Conry. SIMULACT: A generic tool for simulating distributed systems. In Proceedings of the 1987 Eastern Simulation Conference, 1987. Thomas W. Malone. Modeling coordination in organizations and markets. Management Science, 33(10):1317{1332, 1987. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 151{158, Morgan Kaufmann, 1988.). Thomas W. Malone. What is coordination theory? In Proceedings of the National Science Foundation Coordination Theory Workshop, February 1988. Thomas W. Malone. Organizing information processing systems: Parallels between human organizations and computer systems. In Scott P. Robertson, Wayne Zachary, and John B. Black, editors, Cognition, Computing, and Cooperation, pages 56{83. Ablex Publishing Corporation, 1990. Thomas W. Malone, Kenneth R. Grant, Franklyn A. Turbak, Stephen A. Brobst, and Michael D. Cohen. Intelligent information-sharing systems. Communications of the ACM, 30(5):390{402, 1987. James G. March and Herbert A. Simon. Organizations. John Wiley & Sons, 1958. C. Mason, R. Johnson, R. Searfus, D. Lager, and T. Canales. A seismic event analyzer for nuclear test ban treaty verication. In Proceedings of the Third International Conference on Applications of Arti cial Intelligence in Engineering, August 1988. Cindy L. Mason and Rowland R. Johnson. DATMS: A framework for distributed assumption based reasoning. In Les Gasser and Michael N. Huhns, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence, pages 293{317. Pitman, 1989. Theresa Moehlman and Victor Lesser. Cooperative planning and decentralized negotiation in multi-reboss phoenix. In Proceedings of the 1990 DARPA Workshop on Innovative Approaches to Planning, Scheduling, and Control, pages 144{159, November 1990. Thomas A. Montgomery and Edmund H. Durfee. Using MICE to study intelligent dynamic coordination. In Proceedings of the Second Computer Society International Conference on Tools for Arti cial Intelligence, pages 438{444, November 1990. H. Penny Nii. Blackboard systems: The blackboard model of problem solving and the evolution of blackboard architectures. AI Magazine, 7(2):38{53, Summer 1986. Sergei Nirenburg and Victor Lesser. Providing intelligent assistance in distributed oce environments. In Alan H. Bond and Les Gasser, editors, Readings in Distributed Arti cial Intelligence, pages 590{598. Morgan Kaufman, 1988. Chisato Numaoka and Mario Tokoro. OpenMind: Conversational model for open systems. In Proceedings of Workshop on Object-Oriented Computing 1990, March 1990.
12
72] H. Van Dyke Parunak. Distributed ai and manufacturing control: Some issues and insights. In Y. Demazeau and J.-P. Muller, editors, Decentralized AI, pages 81{101. North Holland, 1990. 73] H. Van Dyke Parunak. Toward a formal model of inter-agent control. In Proceedings of the 1990 Distributed AI Workshop, October 1990. 74] Michel Roboam, Mark Fox, and Katia Sycara. Enterprise management network architecture| distributed knowledge base support. Technical Report CMU-RI-TR-90-21, Carnegie-Mellon University, Pittsburgh, Pennsylvania, 1990. 75] Michel Roboam, Mark Fox, and Katia Sycara. Enterprise management network architecture| the organization layer. Technical Report CMU-RI-TR-90-22, Carnegie-Mellon University, Pittsburgh, Pennsylvania, 1990. 76] David A. Rosenblitt. Supporting Collaborative Planning: The Plan Integration Problem. PhD thesis, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, 1990. 77] Jerey S. Rosenschein. The role of knowledge in logic-based rational interactions. In Proceedings of the Seventh Phoenix Conference on Computers and Communications, pages 497{504, Scottsdale, AZ, February 1988. 78] Jerey S. Rosenschein and Michael R. Genesereth. Communication and cooperation among logic-based agents. In Proceedings of the Sixth Phoenix Conference on Computers and Communications, pages 594{600, Scottsdale, AZ, February 1987. 79] Jerey S. Rosenschein, Matthew L. Ginsberg, and Michael R. Genesereth. Cooperation without communication. In Proceedings of the Fifth National Conference on Arti cial Intelligence, pages 51{57, Philadelphia, Pennsylvania, August 1986. 80] Arvind Sathi and Mark S. Fox. Constraint-directed negotiation of resource reallocations. In Michael N. Huhns and Les Gasser, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 81] Arvind Sathi, Thomas E. Morton, and Steven F. Roth. Callisto: An intelligent project management system. AI Magazine, 7(5):34{52, 1986. 82] Michael J. Shaw and Andrew B. Whinston. Learning and adaptation in distributed articial intelligence systems. In Les Gasser and Michael N. Huhns, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 83] Herbert A. Simon. Models of Man. John Wiley & Sons, 1957. 84] Herbert A. Simon. The Sciences of the Arti cial. MIT Press, 1969. 85] Munindar Singh. Towards a formal theory of communication for multiagent systems. In Proceedings of the Twelfth International Joint Conference on Arti cial Intelligence, August 1991. 86] Reid G. Smith. The contract net protocol: High-level communication and control in a distributed problem solver. IEEE Transactions on Computers, C-29(12):1104{1113, December 1980.
13
87] Reid G. Smith and Randall Davis. Frameworks for cooperation in distributed problem solving. IEEE Transactions on Systems, Man, and Cybernetics, SMC-11(1):61{70, January 1981. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 61{70, Morgan Kaufmann, 1988.). 88] Katia P. Sycara. Multiagent compromise via negotiation. In Les Gasser and Michael N. Huhns, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 89] Frank von Martial. Interactions among autonomous planning agents. In Y. Demazeau and J.-P. Muller, editors, Decentralized AI, pages 105{119. North Holland, 1990. 90] Keith James Werkman. Multiagent Cooperative Problem Solving Through Negotiation and Perspective Sharing. PhD thesis, Lehigh University, 1990. 91] Eric Werner. Cooperating agents: A unied theory of communication and social structure. In Les Gasser and Michael N. Huhns, editors, Distributed Arti cial Intelligence, volume 2 of Research Notes in Arti cial Intelligence. Pitman, 1989. 92] Robert Wesson, Frederick Hayes-Roth, John W. Burge, Cathleen Statz, and Carl A Sunshine. Network structures for distributed situation assessment. IEEE Transactions on Systems, Man, and Cybernetics, SMC-11(1):5{23, January 1981. (Also published in Readings in Distributed Arti cial Intelligence, Alan H. Bond and Les Gasser, editors, pages 71{89, Morgan Kaufmann, 1988.). 93] Makoto Yokoo, Toru Ishida, and Kazuhiro Kuwabara. Distributed constraint satisfaction for DAI problems. In Proceedings of the 1990 Distributed AI Workshop, Bandara, Texas, October 1990.
14
