Showing papers by "Munindar P. Singh published in 1998"

Agent communication languages: rethinking the principles

Abstract: Agent communication languages have been used for years in proprietary multiagent systems. Yet agents from different vendors-or even different research projects-cannot communicate with each other. The author looks at the underlying reasons and proposes a conceptual shift from individual agent representations to social interaction.

Verifying Compliance with Commitment Protocols

Abstract: Interaction protocols are specific, often standard, constraints on the behaviors of autonomous agents in a multiagent system Protocols are essential to the functioning of open systems, such as those that arise in most interesting web applications A variety of common protocols in negotiation and electronic commerce are best treated as commitment protocols, which are defined, or at least analyzed, in terms of the creation, satisfaction, or manipulation of the commitments among the participating agents When protocols are employed in open environments, such as the Internet, they must be executed by agents that behave more or less autonomously and whose internal designs are not known In such settings, therefore, there is a risk that the participating agents may fail to comply with the given protocol Without a rigorous means to verify compliance, the very idea of protocols for interoperation is subverted We develop an approach for testing whether the behavior of an agent complies with a commitment protocol Our approach requires the specification of commitment protocols in temporal logic, and involves a novel way of synthesizing and applying ideas from distributed computing and logics of program

A customizable coordination service for autonomous agents

Abstract: We address the problem of constructing multiagent systems by coordinating autonomous agents, whose internal designs may not be fully known. We develop a customizable coordination service that (a) takes declarative specifications ofthe desired interactions, and (b) automatically enacts them. Our approach is based on temporal logic, and has a rigorous semantics and a naturally distributed implementation.

Cognitive agents

Abstract: Several researchers have proposed using cognitive concepts as a semantic basis for agent communications (M.N. Huhns and M.P. Singh, 1997). One of the leading candidates for such a semantics is based on Arcol, the communication language used within Artimis. Interestingly, this application (not only of Arcol, but also in general) appears extremely misguided. The intentional concepts are well suited to designing agents, but are not suited to giving a basis to a public, standardizable view of communication. A challenge for using the cognitive concepts is that although they are natural in several respects and can guide implementations, full blown implementations that try to be faithful to every aspect of the model can end up being computationally demanding. As the cognitive concepts are put to use in real applications, the principles for simplifying the implementations will emerge. In any case, because of their naturalness to humans, the cognitive concepts are here to stay, and we will do well to consider them in the design of our agents.

Developing formal specifications to coordinate heterogeneous autonomous agents

Abstract: We have been developing an approach for the distributed coordination of heterogeneous, autonomous agents. This approach takes as input (a) agent skeletons, giving compact descriptions of the given agents in terms of their events that are significant for coordination, as well as (b) relationships among the events occurring in these skeletons. A natural question is how may the skeletons and relationships be produced in the first place. Parunak (1996) proposed a methodology for designing multiagent systems based on Dooley graphs (1976) from discourse analysis. We show how with a few key modifications, Dooley graphs can also be used to generate the skeletons and relationships required for coordination. This combines the benefits of an intuitive methodology with a formal and distributed framework for developing multiagent systems from autonomous agents.

Workflow agents

Abstract: Software agents as user agents, resource agents, and brokers may be able to enhance usefulness of workflow applications. Workflow technology is important to network computing because workflows exist naturally wherever distributed resources are interrelated. The problem with current workflow technology is that it is often too rigid. The lack of freedom accorded to human participants causes workflow management systems to appear unfriendly. As a result, they are often ignored or circumvented. This rigidity also causes productivity losses by making it harder to accommodate the flexible, ad hoc reasoning of human intelligence. Another challenge is that system requirements are rarely static. Software agents promise to address these challenges. The roles of greatest interest to a workflow setting are user agents, resource agents, and brokers. When a workflow is constituted in terms of distinct roles that agents can instantiate, the agents can be set up to respect the constraints of their users and resources. User agents negotiate with one another and with resource agents to ensure that global constraints are not violated and that global efficiencies can be achieved. Agents can include functionality to identify different kinds of exception conditions and react appropriately, possibly by negotiating a special sequence of actions. More importantly, agents can learn from repeated instances of the same kinds of exceptions. With this learning ability, agents can process the updated set of constraints when system requirements change.

Personal assistants

Abstract: Already there are simple personal agents to help with some of our shopping. There are agents to track stocks in our portfolios, advise us on how to use particular software products, and arrange meetings within corporate workgroups. However, none of these agents takes more than one aspect of our activities into account, nor do they adapt easily to our preferences. Personal assistants, on the other hand, are agents that can represent individuals on the Web. They help users in their day-to-day activities, especially those involving information retrieval, negotiation, or coordination. A personal assistant might schedule a meeting and then, based on the meeting location, find the nearest babysitting service or the ATM with the lowest transaction fee. The paper discusses the interaction framework, negotiation, profiles and standards.

Lazy updates of profiles in a system of communication devices

Abstract: Profiles in personal communication devices are updated by a method called lazy updates or demand-driven updates. In a lazy update, information is transmitted between users of personal communication devices and updates are performed only when the users communicate for other purposes. Users and service providers may update their profiles as often as they wish. Then, when a call or service request is initiated based on a given profile, the initiator of the call or service request ensures that the profile that is being used is sufficiently up-to-date for the present purpose and, if it is not, updates the profile. Lazy updates can occur when communication occurs for reasons other than solely updating the information. Examples of such updates include modification of a user's schedule or changes in the services that a service provider offers. Updates that affect future communication, such as changes in address information, telephone numbers or email addresses, happen much less frequently. However, updates such as these must occur at the time the update occurs, rather at some indefinite time in the future. Such updates are typically sent to certain designated key recipients, such as to individuals in frequent contact with the user or to selected services providers.

The intentions of teams: team structure, endodeixis, and exodeixis

Abstract: Teams arise in a number of important multiagent appli- cations. Several theories of intentions for teams have been proposed. By and large, these theories tend to model team intentions exclu- sively on the basis of mental concepts, and fail to acknowledge the internal structure of teams. We present a formal theory of intentions for teams that consid- ers the structure of teams explicitly. In this context, we distinguish between exodeictic and endodeictic intentions, which are conceptu- alized as pointing outward or inward from a team. These concepts are formalized in a framework that models the structure of teams in terms of their members' commitments and coordination require- ments. In this way, our approach combines mental and social con- cepts in a principled manner. We describe some postulates concern- ing intentions and structure, and give technical results establishing or falsifying these postulates with different definitions.

All agents are not created equal

Abstract: As the technology advances, we can expect the development of specialized agents to be used as standardized building blocks for information systems. Two trends lend credence to such a prediction. First, software systems in general are being constructed with larger components, such as ActiveX and JavaBeans, which are becoming closer to being agents themselves. They have more functionality than simple objects, respond to events autonomously, and, most importantly, respond to system builders at development time, as well as to events at runtime. Moreover, there is a move toward more cooperative information systems, in which the architecture itself plays an important role in the effectiveness of the system, as opposed to traditional software systems where effectiveness depends on the quality of the individual components. These architectures are generating a set of standardized agents. Architectures based on standardized agent types should be easier to develop, understand, and use. Perhaps most important of all, these architectures will make it easier for separately developed information systems to interoperate.

Multiagent Systems in Information-Rich Environments

Abstract: Information-rich environments are the open environments that characterize most of the modern applications of computing technology. The applications include ubiquitous information access, electronic commerce, virtual enterprises, logistics, and sensor integration, to name but a few. These applications differ from conventional database applications not only in the nature and variety of information they involve, but also in including a significant component that is beyond the information system per se: the creation, transformation, use, and ultimate fate of information. The environments are typified only by the large amounts and varieties of information they include, and whose effective and efficient management is key to the above applications. Multiagent systems (MAS) are an important paradigm for building complex information systems, especially cooperative ones. We describe how cooperative information system architectures have evolved a set of common types of computational agents. We also describe two approaches that address complementary aspects of MAS construction.

Intelligent agents IV : agent theories, architectures, and languages : 4th International Workshop, ATAL'97, Providence, Rhode Island, USA, July 24-26, 1997 : proceedings

Abstract: Panel: Methodologies for multi-agent systems.- Designing embedded agents to optimize end-user objectives.- Agent architecture as object oriented design.- Implementation of a cooperative agent architecture based on the language-action perspective.- Toward the specification and design of industrial synthetic ecosystems.- Bidding mechanisms for data allocation in multi-agent environments.- Distributed storage of replicated beliefs to facilitate recovery of distributed intelligent agents.- A customizable coordination service for autonomous agents.- A behavior-based approach to reactivity and coordination: A preliminary report.- On explicit plan languages for coordinating multiagent plan execution.- Social comparison for failure detection and recovery.- Multi-agent coordination through coalition formation.- A formal specification of dMARS.- A framework for argumentation-based negotiation.- Agent modelling in MetateM and DESIRE.- Semantics for an agent communication language.- Formal semantics for an abstract agent programming language.- Intentional agents and goal formation.- A layered agent calculus with concurrent, continuous processes.- Approximate reasoning about combined knowledge.- On the epistemic foundations of agent theories.- Facilitating open communication in agent systems: The InfoSleuth infrastructure.- Competition for attention.- Analysis and design of multiagent systems using MAS-CommonKADS.- Multi agent systems as open societies - a design framework.- TKQML: A scripting tool for building agents.

Semantical considerations on intention dynamics for BDI agents

Abstract: The BDI paradigm is a powerful means for constructing intelligent agents in terms of their beliefs, desires, and intentions. For this paradigm to bear its full potential, it must incorporate considerations from rationality. This paper develops a set of postulates for intelligent agents who deliberate about their intentions and actions. However, even simple postulates can lead to paradoxical results when formalized naively. We propose an approach based on temporal possibility and action that avoids those problems. This approach incorporates a formal model based on branching time in which a probabilistic analysis of choice can be captured. In this manner, the intuitions of the BDI paradigm can be reconciled with those of rational agency.

Applying the Mu-Calculus in Planning and Reasoning about Action

Abstract: Planning algorithms have traditionally been geared toward achievement goals in single-agent environments. Such algorithms essentially produce plans to reach one of a specified set of states. More general approaches for planning based on temporal logic (TL) are emerging. Current approaches tend to use linear TL, and can handle sets of sequences of states. However, they assume deterministic actions with all changes effected solely by one agent. By contrast, we use a branching model of time that can express concurrent actions by multiple agents and the environment, leading to nondeterministic effects of an agent’s actions. For this reason, we view plans not as sequences of actions, but as decision graphs describing the agent’s actions in different situations. Thus, although we consider single-agent decision graphs, our approach is better suited to multiagent systems. We also consider an expressive formalism, which allows a wider variety of goals, including achievement and maintenance goals. Achievement corresponds to traditional planning, but maintenance is more powerful than traditional maintenance goals, and may require nonterminating plans. To formalize decision graphs requires a means to “alternate” the agent’s and the environment’s choices. From logics of program, we introduce the propositional mu-calculus, which has operators for least and greatest fixpoints. We give a semantics, a fixpoint characterization, and an algorithm to compute decision graphs.

The Bases of Effective Coordination in Decentralized Multi-Agent Systems

Abstract: Coordination is a recurring theme in multiagent systems design. We consider the problem of achieving coordination in a system where the agents make autonomous decisions based solely on local knowledge. An open theoretical issue is what goes into achieving effective coordination? There is some folklore about the importance of the knowledge held by the different agents, but the rest of the rich agent landscape has not been explored in depth. The present paper seeks to delineate the different components of an abstract architecture for agents that influence the effectiveness of coordination. Specifically, it proposes that the extent of the choices available to the agents as well as the extent of the knowledge shared by them are both important for understanding coordination in general. These lead to a richer view of coordination that supports a more intuitive set of claims. This paper supports its conceptual conclusions with experimental results based on simulation.

Agent jurisprudence

Abstract: The agent metaphor comes packaged with a number of powerful abstractions. Some of these are psychological, such as beliefs, knowledge, and intentions-abstractions that were traditionally studied in AI. However, there are a number of other abstractions that the agent metaphor brings to the fore. Of these, one has been emphasizing the social abstractions. Close cousins of the social abstractions are the ethical and legal abstractions. These too are being recognized as increasingly important in developing agents that are not only sociable, but also well behaved.

Agents on the Web: Anthropoid Agents.

Abstract: In the study of agents on the Internet, we often ascribe to them human qualities, such as beliefs and intentions. These qualities are best understood as metaphors that give developers a way to talk about and design the capabilities and applications of agents. Despite all the progress in computing, users have been slow to accept the technology. They have often accepted what was thrown at them, but only under economic duress. Bringing the technology closer to their emotional needs might ease this resistance. So how can we put a human face on computing? Maybe by putting an animated face on it! Thus, an interface may consist of an agent, which has an explicit presence (for example, as an on-screen animated figure) and appears to have a personality. In computer human interfaces, especially for education and commerce where a large variety of people must deal with computers, an anthropoid agent might be more inviting. Depending on the situation, the agent might appear shy, friendly, stern, or knowledgeable. For example, people might better accept advice offered politely by a shy agent, or heed warnings uttered seriously by a stern agent. And they might be more likely to purchase goods or services offered by a friendly, knowledgeable agent that could appear empathetic to their needs.

Anthropoid agents

Abstract: In the study of agents on the Internet, we often ascribe to them human qualities, such as beliefs and intentions. These qualities are best understood as metaphors that give developers a way to talk about and design the capabilities and applications of agents. Despite all the progress in computing, users have been slow to accept the technology. They have often accepted what was thrown at them, but only under economic duress. Bringing the technology closer to their emotional needs might ease this resistance. So how can we put a human face on computing? Maybe by putting an animated face on it! Thus, an interface may consist of an agent, which has an explicit presence (for example, as an on-screen animated figure) and appears to have a personality. In computer human interfaces, especially for education and commerce where a large variety of people must deal with computers, an anthropoid agent might be more inviting. Depending on the situation, the agent might appear shy, friendly, stern, or knowledgeable. For example, people might better accept advice offered politely by a shy agent, or heed warnings uttered seriously by a stern agent. And they might be more likely to purchase goods or services offered by a friendly, knowledgeable agent that could appear empathetic to their needs.

Agents on the Web: Personal Assistants.

Abstract: Already there are simple personal agents to help with some of our shopping. There are agents to track stocks in our portfolios, advise us on how to use particular software products, and arrange meetings within corporate workgroups. However, none of these agents takes more than one aspect of our activities into account, nor do they adapt easily to our preferences. Personal assistants, on the other hand, are agents that can represent individuals on the Web. They help users in their day-to-day activities, especially those involving information retrieval, negotiation, or coordination. A personal assistant might schedule a meeting and then, based on the meeting location, find the nearest babysitting service or the ATM with the lowest transaction fee. The paper discusses the interaction framework, negotiation, profiles and standards.

Agent information contracts within virtual private networks

Abstract: Virtual enterprises are being built on an Internet backbone, for which security provisioning is an obvious first concern. However, additional services need provisioning as well. For instance, contracted arrangements for quality inspection, notification of delay, and other such coordinations can be provided by semi-autonomous agent-based services. We introduce information contracts, which specify the commitments of ContractAgents to each other. Based on human-readable contract templates with fill-in-the-black terms, ContractAgents manage the negotiation stage between user specialists and then operationalize and ensure these commitments during enactment. Building on the idea of an agent-based sphere of commitments (SoComs) to overcome weaknesses in extended transaction processing and traditional workflow management, we show how ContractAgents also need to be positioned within the security architecture of virtual private networks. The standardization of contracts, SoCom servers, and operationalized commitments can then provide a basis for compliance checks by trusted third parties and oversight authorities.

Using Agents to Achieve Process Coherence

Abstract: ions for Composite Activities Information management in the large involves three main concerns, which must be addressed by any approach for constructing solutions: • data integrity and flow: correctness of data and how it is conveyed from one party to another • organizational structure: how the various parties relate to each other • autonomy: how the autonomy of the different parties is preserved. Table 1 summarizes the major abstractions for programming composite activities. Database transactions are the most rigorous, and require that only correct data ever be visible [11]. This entails that outputs be released only when a transaction completes. Thus, the producer of data is restricted, but the consumer has full autonomy, unless it also is a transaction. Spheres of control (SoCs) release results early, but may undo and redo the consumers—thus, consumers have no autonomy [8]. Extended transactions release results liberally, but restrict the autonomy of their components by requiring compensating subtransactions to undo the effects of data that are invalidated [8]. Workflows ignore the integrity aspects, but capture the data flow required by specific applications [10]. They allow autonomy, but are not flexible. Spheres of commitments are discussed below. Support Provided (columns) Integrity & Autonomy Organizational 1 Ericsson New Concepts Group, Research Triangle Park, NC. 2 IBM Intelligent Agents Group, Research Triangle Park, NC. 3 Dept of Computer Science, North Carolina State University, Raleigh, NC.

Agents on the Web: All Agents are Not Created Equal.

Abstract: As the technology advances, we can expect the development of specialized agents to be used as standardized building blocks for information systems. Two trends lend credence to such a prediction. First, software systems in general are being constructed with larger components, such as ActiveX and JavaBeans, which are becoming closer to being agents themselves. They have more functionality than simple objects, respond to events autonomously, and, most importantly, respond to system builders at development time, as well as to events at runtime. Moreover, there is a move toward more cooperative information systems, in which the architecture itself plays an important role in the effectiveness of the system, as opposed to traditional software systems where effectiveness depends on the quality of the individual components. These architectures are generating a set of standardized agents. Architectures based on standardized agent types should be easier to develop, understand, and use. Perhaps most important of all, these architectures will make it easier for separately developed information systems to interoperate.

Agents on the Web: Cognitive Agents.

Abstract: Several researchers have proposed using cognitive concepts as a semantic basis for agent communications (M.N. Huhns and M.P. Singh, 1997). One of the leading candidates for such a semantics is based on Arcol, the communication language used within Artimis. Interestingly, this application (not only of Arcol, but also in general) appears extremely misguided. The intentional concepts are well suited to designing agents, but are not suited to giving a basis to a public, standardizable view of communication. A challenge for using the cognitive concepts is that although they are natural in several respects and can guide implementations, full blown implementations that try to be faithful to every aspect of the model can end up being computationally demanding. As the cognitive concepts are put to use in real applications, the principles for simplifying the implementations will emerge. In any case, because of their naturalness to humans, the cognitive concepts are here to stay, and we will do well to consider them in the design of our agents.

Implementing Multi-Agent Systems: Languages, Frameworks, and Standards

Abstract: As multi-agent systems research nears the end of its second decade, researchers have moved beyond stand-alone, one-off systems and have begun to create the software infrastructure for quickly creating new, highly interoperable systems. However, the desire for seamless interoperability ('open' systems) brings with it the push to standardize on agent communication languages and related agent service facilities. Although a host of theoretical and practical controversies surround the specification of such standards, some shared, guiding principles have emerged.

Implementing Multi-Agent Systems: Languages, Frameworks, and Standards (Extended Abstract)*

Abstract: As multi-agent systems research nears the end of its second decade, researchers have moved beyond stand-alone, one-off systems and have begun to create the software infrastructure for quickly creating new, highly interoperable systems However, the desire for seamless interoperability ("open" systems) brings with it the push to standardize on agent communication languages and related agent service facilities Although a host of theoretical and practical controversies surround the specification of such standards, some shared, guiding principles have emerged Many researchers focus their work on the internal structures of agents (architecture), while others focus more on the activities and structures between agents (organization) The first principle is the realization that most principles can be recast at both the intraand inter(architectural and organizational) levels The second principle is a drive toward the reuse of models and particular behaviors both conceptually, in reusable implementations, and in the creation of generic infrastructure components The third principle revolves around how to structure individual or multi-agent knowledge and behaviors that includes techniques for composition, layering, and abstraction The final guiding principle is the centrality of the interoperability question After all, what would a multiagent system be unless there are in fact multiple agents interoperating with one another?! Methodologically, research in the area of agent frameworks and languages has been pursued in both a theory-driven and application-driven manner Often there can be significant interaction between theory and practice, as in the development of BDI-based systems (Bratman 1987; Rao & Georgeff 1995) Although there are as yet no complete solutions, multi-agent system development methodologies will be important for the software engineer-