One reason why Distributed AI (DAI) technology has been deployed in relatively few real-size applications is that it lacks a clear and implementable model of cooperative problem solving which specifies how agents should operate and interact in complex, dynamic and unpredictable environments. As a consequence of the experience gained whilst building a number of DAI systems for industrial applications, a new principled model of cooperation has been developed. This model, called Joint Responsibility, has the notion of joint intentions at its core. It specifies pre-conditions which must be attained before collaboration can commence and prescribes how individuals should behave both when joint activity is progressing satisfactorily and also when it runs into difficulty. The theoretical model has been used to guide the implementation of a general-purpose cooperation framework and the qualitative and quantitative benefits of this implementation have been assessed through a series of comparativ...

Coordination, the process by which an agent reasons about its local actions and the (anticipated) actions of others to try and ensure the community acts in a coherent manner, is perhaps the key problem of the discipline of Distributed Artificial Intelligence (DAI). In order to make advances it is important that the theories and principles which guide this central activity are uncovered and analysed in a systematic and rigourous manner. To this end, this paper models agent communities using a distributed goal search formalism, and argues that commitments (pledges to undertake a specific course of action) and conventions (means of monitoring commitments in changing circumstances) are the foundation of coordination in all DAI systems. 1. The Coordination Problem Participation in any social situation should be both simultaneously constraining, in that agents must make a contribution to it, and yet enriching, in that participation provides resources and opportunities which would otherwise ...

We describe representation, inference strategies, and control procedures employed in an automated conversation system named the Bayesian Receptionist. The prototype is focused on the domain of dialog about goals typically handled by receptionists at the front desks of buildings on the Microsoft corporate campus. The system employs a set of Bayesian user models to interpret the goals of speakers given evidence gleaned from a natural language parse of their utterances. Beyond linguistic features, the domain models take into consideration contextual evidence, including visual findings. We discuss key principles of conversational actions under uncertainty and the overall architecture of the system, highlighting the use of a hierarchy of Bayesian models at different levels of detail, the use of value of information to control question asking, and application of expected utility to control progression and backtracking in conversation.

The acquisition of new goals: the case of norms It is generally acknowledged that norms and normative action emphasise autonomy on the side of decision. But what about the autonomous formation of normative goals? In a recent paper (Dignum & Conte 1997), the treatment of goalacquisition in the Agent Theory (AT) literature was found inadequate, some formal rules for goal-generation have been proposed, and the role of social inputs in the acquisition of new goals has been emphasised. Here, we intend to continue that work, by including norms among the social inputs to one's goals, and by extending the goal-generation rule to the case of normative goals. The general question then is, how and why do autonomous agents form normative goals? The answer to this question goes back to a former paper by some of the authors (Conte & Castelfranchi 1995), where a typology of reasons for accepting norms has been explored in analogy with goal-adoption. Here, however, the formal instruments worked ou...

this paper is inspired by philosophical work, it is squarely motivated by the concerns of building intelligent systems that are capable of collaborative behavior, either with a user, or with other such systems. Still, we hope that the paper sheds light on philosophical issues, and treats the subject of joint action at a sufficiently precise level to be illuminating of problems that any philosophical account needs to confront. An important consequence of focusing on joint actions, rather than solely on individual actions, is the opportunity to rethink related theories. In particular, we claim that speech act theory will need to be recast in light of joint action theory since many of the basic illocutionary acts (e.g., requests, promises) are intimately involved in eatablishing, monitoring, and discharging joint activities. However, despite this tight relationship, no existing speech act theory provides guidance on this connection. This paper takes a first step in the direction of linking speech act theory and joint action theory by showing how various speech acts can be used to form and disband teams. It is by now commonplace to observe that joint action is different from a collection of individual actions, even if they are coordinated. Agents can be acting in a coordinated fashion, as in ordinary automobile traffic, but not be acting together. Conversely, agents can be acting together, but not be coordinated except at the start and end of their joint action (e.g, see [36]) The key property distinguishing joint or collaborative action from mere coordinated action is the joint mental state of the participants. The best way to explore what this mental state must be is to imagine a joint action going astray. Our favorite example is driving in a convoy, versus ordinary traff...

. The cooperation knowledge level is a new computer level specifically for multi-agent problem solvers which describes rich and explicit models of common social phenomena. A cooperation level description (called joint responsibility) is developed to describe how participants should behave during interactions in which groups of agents collaborate to solve a common problem. The utility of this model is highlighted in the real-world environment of electricity transport management in which agents have to make decisions using partial, imprecise views of the system and cope with the inherent dynamics of the environment. In such situations the tracking of social action becomes a primary consideration; joint responsibility provides evaluation criteria and the causal link to behaviour upon which such assessment can be based. Keywords: Multi-Agent Systems, Distributed AI, Joint Intentions, Knowledge Level 1. Introduction Sophisticated problem solving is based upon knowledge. In advanced syste...

In many industrial applications, large centralized software systems are not as effective as distributed networks of relatively simpler computerized agents. For example, to compete effectively in today's markets, manufacturers must be able to design, implement, reconfigure, resize, and maintain manufacturing facilities rapidly and inexpensively. Because modern manufacturing depends heavily on computer systems, these same requirements apply to manufacturing control software, and are more easily satisfied by small modules than by large monolithic systems. This paper reviews industrial needs for Distributed Artificial Intelligence (DAI), giving special attention to systems for manufacturing scheduling and control. It describes a taxonomy of such systems, gives case studies of several advanced research applications and actual industrial installations, and identifies steps that need to be taken to deploy these technologies more broadly.

Systems composed of multiple interacting problem solvers are becoming increasingly pervasive and have been championed in some quarters as the basis of the next generation of intelligent information systems. If this technology is to fulfill its true potential then it is important that the systems which are developed have a sound theoretical grounding. One aspect of this foundation, namely the model of collaborative problem solving, is examined in this paper. A synergistic review of existing models of cooperation is presented, their weaknesses are highlighted and a new model (called joint responsibility) is introduced. Joint responsibility is then used to specify a novel high-level agent architecture for cooperative problem solving in which the mentalistic notions of belief, desire, intention and joint intention play a central role in guiding an individual's and the group's problem solving behaviour. An implementation of this highlevel architecture is then discussed and its utility is il...

this paper we focus on the solutions we are providing for the outer layer of the architecture. They are embedded into a domain independent COOrdination Lan3 guage (COOL) that provides services for defining distributed agent configurations, managing communication, defining and managing structured interactions amongst agents, external software integration and in context acquisition and debugging of coordination knowledge. As these solutions impact on the way agents manage change by information distribution and conflict resolution, we also address these aspects showing how the coordination service supports these tasks. The paper is structured as follows. In section 2 we review the work in Distributed Artificial Intelligence from several perspectives and define our research goals. As the subsequent presentation of our tools is carried out in the context of our main application, the agent-based integration of the supply chain of manufacturing enterprises, we continue in section 3 with presenting this application domain. Section 4 deals with the main subject of the paper, the components of the coordination language. We illustrate the language throughout with examples from the supply chain. Section 5 then deals with the coordination knowledge acquisition service that allows users to extend and debug coordination knowledge on-line. To show how the coordination system is integrated with other reasoning tasks in the Agent Building Shell, in section 6 we review two other services of the architecture that make use of the coordination framework, cooperative information distribution and cooperative conflict management. In the end, we discuss some related approaches and provide concluding remarks.

Abstract. To scale agent technologies for widespread use in open systems, agents must have an understanding of the organizational context in which they operate. In this paper we focus on the issue of task valuation and action selection in socially situated or organized agents – specifically on the issue of quantifying agent relationships and relating work motivated by different sources. 1