Skip to main content
main-content

Über dieses Buch

Since the first edition was published two years ago, much has been done on extend­ ing the work done on SMART to address new and important areas [3-5,54,79,80, 108-110,116,118-120,122]. In this second edition, we have revised, updated and corrected the existing text and added three new chapters. These chapters provide a broader coverage of the fie1d of agents, and show in more detail how the specific framework described can be used to examine other areas. In Chapter 6, we use the concepts of discovery to apply the framework to autonomous interaction in multi­ agent systems; in Chapter 10 we use it for considering normative agents and sys­ tems; and in Chapter 11 we describe work on an implementation and development environment. As a course text, the book can be considered in different parts, as follows. - Chapter I and Chapter 2 offer a basic introduction to agents and their core com­ ponents. - Chapter 3 and Chapter 4 cover relationships between agents and basic notions of cooperation for multi-agent systems. - Chapter 5 and Chapter 6 introduce sociological agents, which are needed for rea­ soning and planning, and their use in reasoning about communication and inter­ action. - Chapter 7, Chapter 8, Chapter 9 and Chapter 10 each cover different application areas relating to different aspects, inc1uding coordination (through the contract net), agent architecture (through AgentSpeak(L), social dependence networks, and normative systems.

Inhaltsverzeichnis

Frontmatter

1. The Agent Landscape

Abstract
Over the last decade or so, the notions underlying agent-based systems have become almost commonplace, yet were virtually unknown in earlier years. Not only have agent-based systems moved into the mainstream, they have spread beyond a niche area of interest in artificial intelligence, and have come to be a significant and generic computing technology. The dramatic and sustained growth of interest is demonstrated by the increasing number of major conferences and workshops in this very dynamic field, covering a depth and breadth of research that testifies to an impressive level of maturity for such a relatively young area.
Mark d’Inverno, Michael Luck

2. The SMART Agent Framework

Abstract
Though agents are becoming increasingly popular across a wide range of applications, the rapid growth of the field has led to much confusion regarding agents and their functionality. One part of this confusion is that it is now generally recognised that there is no agreement on what it is that makes something an agent. For example, Franklin and Graesser [67] provide evidence of the degree of current difficulties, citing ten different agent definitions of leading researchers. This lack of consensus sets up barriers to the development of an accepted foundation on which to build a rigorous scientific discipline. It can also be argued that this problem, which has resulted in a plethora of different terms and notions, hinders current research since integration and comparison of different approaches is made very difficult.
Mark d’Inverno, Michael Luck

3. Agent Relationships

Abstract
It is the interaction between individual agents by which goals are typically achieved in multi-agent systems [167]. The form of such interaction can range over interleaved actions, combined actions, message-passing or high-level linguistic utterances such as speech acts, depending on the nature of the agents themselves. This suggests that an account of interaction is only possible through an understanding of the agents involved.
Mark d’Inverno, Michael Luck

4. An Operational Analysis of Agent Relationships

Abstract
As elaborated in the previous chapter, fundamental to the operation of multi-agent systems is the concept of cooperation and engagement between individual agents. If single-agent systems can both cooperate and engage others, they can exploit the capabilities and functionality of others to achieve their own individual goals. Once this is achieved, then such systems can potentially move beyond the advantages of robustness in traditional distributed systems in the face of individual component failure since components can be replaced and cooperation configurations realigned. In principle then, the multi-agent system paradigm allows the specific expertise and competence of different agents to complement each other so that in addition to general resilience, the overall system exhibits significantly greater functionality than individual components.
Mark d’Inverno, Michael Luck

5. Sociological Agents

Abstract
Now that the inter-agent relationships that arise from the application of the SMART framework have been identified and analysed, the requisite deliberative qualities of such agents for effective and efficient action in order to gain this benefit can be addressed. An individual agent can achieve this, first through the appropriate use of its own capabilities, and second through the successful exploitation of the capabilities of others.
Mark d’Inverno, Michael Luck

6. Autonomous Interaction

Abstract
While we have considered many aspects of agent systems, especially the configurations that arise in systems of multiple interacting agents, we have not explicitly addressed the issues underlying interaction itself. In this chapter we re-examine what it means for an agent to be autonomous and, more specifically, what that entails for an adequate model of interaction between such agents. As we have discussed, complex environments admit an inherent uncertainty that must be considered if we are to cope with more than just toy problems. In such uncertain environments, agents must be autonomous; an agent can never know in advance the exact effects of its own actions nor of the actions of others. This is of paramount importance, and an agent must therefore be designed with a flexibility that enables it to cope with this uncertainty by evaluating it and responding to it in adequate ways.
Mark d’Inverno, Michael Luck

7. The Contract Net as a Goal Directed System

Abstract
The models defined in SMART provide a structure that can be applied directly to describe and analyse multi-agent systems and theories, and to derive models that highlight relevant aspects. In this chapter, the path from the initial SMART framework to the modelling of such specific systems is completed by describing a mechanism for implementing the contract net protocol.
Mark d’Inverno, Michael Luck

8. Computational Architecture for BDI Agents

Abstract
While many different and contrasting single-agent architectures have been proposed, perhaps the most successful are those based on the belief-desire-intention (BDI) framework. There is a wealth of research that has accumulated on both the formal and theoretical aspects of BDI agents through the use and development of various logics, for example, on the one hand, and on the practical aspects through the development of implementations of BDI agents on the other. Indeed, so many deliberative agent architectures and systems are based on the BDI framework that it is viewed as being as central to single-agent systems as the contract net is to multi-agent systems.
Mark d’Inverno, Michael Luck

9. Evaluating Social Dependence Networks

Abstract
Social dependence networks [155] (SDNs) are structures that form the basis of a computational model of social power theory as originally proposed by Castel-franchi [18]. Essentially, they are taxonomies of social relationships that can be derived from the ‘power’ agents have over one another as a result of their ability to achieve each other’s goals. Based on these taxonomies, social reasoning mechanisms have been developed by which agents can reason about inter-agent dependencies.
Mark d’Inverno, Michael Luck

10. Normative Agents

Abstract
Agents operating in a common society need to be constrained in order to avoid and address conflicts, make agreements, reduce complexity, and, generally, to achieve a desirable social order [28, 30, 31, 124]. This is the role of norms, which represent what ought to be done by a set of agents, and whose fulfillment can be seen as a public good when the benefits that they bring can be enjoyed by the society, organisation or group [20]. Norms thus represent the means to achieve the societal goals of a society. Research on norms and agents has ranged from fundamental work on the importance of norms in agent behaviour [31, 170] to proposing internal representations of norms [29,171], considering their emergence in groups of agents [173], and proposing logics for their formalisation [147, 177].
Mark d’Inverno, Michael Luck

11. act smart: Building a smart System

Abstract
Before the agent paradigm experiences widespread use, there are many issues that need to be resolved [111]. Perhaps most important, however, is the very urgent need for agent development methodologies and agent frameworks to enable developers to implement agent-based systems effectively and quickly. The development of such systems can also provide much needed experience to aid in answering other questions.
Mark d’Inverno, Michael Luck

12. Conclusions

Abstract
This book has described a generally applicable notion of agenthood that is not an attempt to create yet another agent definition, but one that relates to existing concepts and approaches to encompass them. In the definitions provided, we set up a sound conceptual base on which to elaborate more sophisticated definitions and architectures, while at the same time constructing a framework that can accommodate all types of agent, be they non-computational, reflexive, or deliberative.
Mark d’Inverno, Michael Luck

Backmatter

Weitere Informationen