Multiagent Engineering: A New Software Construction

Preface 1 Multiagent Engineering: A New Software Construction Paradigm Multiagent systems have a long academic tradition. They have their roots in distributed problem solving in Artificial Intelligence (AI) from where they emerged in the mid-eighties as a distinctive discipline. Research in multiagent systems owes much to the work of Rosenschein on rationality and autonomy of intelligent agents, the European MAAMAW workshop series, and last but not least the famous readings of Bond & Gasser (1988) and Jacques Ferber´s book on multiagent systems (1991). It gained further by a public discussion via the Distributed AI mailing list in summer 1991, when the pioneers of the field compared in much detail the concepts of distributed problem solvers to multiagent systems. Within only five years, a new exciting field of research had been established. Now, 15 years later, the field has matured to a degree that allows the re- sults of academic research to be passed on to practical use and commercial exploitation. This potential coincides with a need for much larger flexibil- ity of our IT infrastructure in light of its highly distributed character and extreme complexity, but also the global character of the business processes and the large number of business partners due to outsourcing and speciali- zation. Many experts claim that multiagent systems are the right software technology for the needed IT infrastructure at the right time. The appeal has much to do with the broad perspectives of multiagent systems research. 1. The sociological perspective considers agents as members of a collec- tion of agents (groups) of a more or less formally established agent or- ganization, or of an agent society. In that perspective, agents have ac- quaintances, exhibit social behaviors, and establish social relationships. Important themes are collective action, joint intentions, and group be- haviors. 2. The AI perspective is concerned with intelligent behavior on the individ- ual and on the collective level. It studies strategies for collective prob- lem solving, and is particularly interested in cooperative planning algo- rithms, negotiation protocols, distributed representations of knowledge, and collective forms of temporal and spatial reasoning. All these are re- VI Preface quired in physically distributed systems, where the nodes exhibit to a substantial degree locally autonomous behavior. 3. The economic perspective claims that bounded rationality, limited physical resources, and incomplete knowledge require agents to behave “economically rational.” That is, the agents need goals and objectives, they need appropriate decision-making rules, they may display strategic behaviors, and they must be capable of coping both with intra-agent (e.g., role) conflicts, and with inter-agent conflicts. Important contribu- tions come from game theory and decision theory. 4. Finally, the application perspective developed, often experimentally, ap- plications in a wide range of fields. Examples can be found in mechani- cal engineering, robotics, production planning and control, in business, the natural sciences (e.g., simulation of ant colonies, of fish populations, etc.), and in the social sciences. Once we start to exploit the field of multiagent systems in the practical world we pass from research to engineering. This book – Multiagent Engi- neering: Theory and Applications in Enterprises – intends to support the passage by focusing on the technology of multiagent systems as a power- ful, productive tool in the hands of software engineers satisfying extremely complex, dynamic new applications in enterprises. This book, thus, concentrates on a fifth, the engineering perspective. After introducing the basic concepts of agent technology it gives an over- view of two powerful multiagent systems in promising application sce- narios, production logistics and health care. It then turns to the technical details of multiagent technology all the way from (early) requirements en- gineering to system testing. The evaluation section summarizes the main results and presents new and interesting insights from a legal perspective, which, of course, is highly relevant if enterprises decide to employ multi- agent technology in order to improve their competitiveness through ex- tremely adaptive new enterprise information systems architectures. 2 Who Should Read This Book, and Why? Why would we claim that multiagent systems satisfy extremely complex, dynamic new applications in enterprises? The thinking goes that the vari- ous aforementioned perspectives in their combination make multiagent technology – and by extension the software systems relying on them – ex- tremely flexible. Then when would we need flexibility? Certainly when the applications and their environments are highly dynamic, i.e. change on short notice and in many and often unforeseen ways. Multiagent systems Preface VII should be capable of quickly adapting the behavior of the local subsystems (agents) as well as of the overall (multiagent) system, of exchanging coop- eration protocols, of integrating new agents or withdrawing older agents. To turn promise into reality is foremost an engineering task. Engineer- ing must take place on two levels: The individual agent as the basic entity and the multiagent system as the collaborative effort of several agents gov- erned by an intelligent interplay between these two levels. And since agent technology goes beyond traditional object-oriented or component-oriented software, it raises new challenges for the software engineer. First of all, then, the book has been written for software engineers. They will find answers to questions such as: What are the relevant new features of this new tool, compared with other recent software technologies like object orientation and component technology? Which software develop- ment challenges can and should be addressed by it? How shall the software development process be organized? What are the concrete benefits and are they really worth the risks coming with each new technology? What are the “hidden costs” of the technology that need to be considered? The last two questions are not merely technical. Rather, they express concerns of business managers as well. They must ultimately decide how well multiagent technology can contribute to their business objectives, whether their IT systems must indeed cope with rapidly changing produc- tion technologies, enterprise processes, supply chains, business models, markets, and customer demands, and whether multiagent technology seems to be a solution superior to today’s enterprise software. The book is thus a unique source of knowledge about multiagent engi- neering, starting from its basics and then considering all phases in the software lifecycle from early requirements engineering through system de- sign and implementation to testing and illustrating these in the framework of two applications with a high degree of dynamics. 3 A Reader’s Guide The book has to keep in mind that it addresses a diverse readership ranging from business managers to software engineers and that even software en- gineers are a heterogeneous group exhibiting different levels of knowledge in software technology, artificial intelligence, and multiagent systems. The book starts with a management summary that gives a brief over- view of the most important issues and their solutions. As the title suggests, the chapter intends to put the business manager in a position to decide whether the investment in agent technology is worth the desired benefits VIII Preface and which issues deserve closer inspection by the IT specialists in the en- terprise. Part I addresses both the business manager and the software engineer new to agent technology. It gives a concise introduction to multiagent technology and lays the foundation for the remainder of the book. In particular, its last chapter presents operational definitions for the different types of flexibility of multiagent systems. Parts II and III, Agent.Enterprise and Agent.Hospital respectively, have been written for readers primarily interested in the application of multiagent technology in important business domains. They illustrate to the software engineer how one builds large multiagent systems and at the same time demonstrate to the business manager the benefits of agent technology or, to some degree, how difficult it sometimes can be to be absolutely sure when one technology is superior to another one. Each part starts with an overview and is followed by further chapters that give more details on the design, implementation, and evaluation of the multiagent system. Since both systems take a fairly strict supply chain/logistics perspective, their designs can easily be compared. It is thus recommended that readers primarily interested in manufacturing information systems focus on Part II, but they may also gain useful insights from Part III and vice versa for readers interested in new approaches to health care information systems. Part IV concerns itself with the methodical underpinning for engineering multiagent systems software. Its readers should primarily be software architects, software designers, and software engineers. This part should be understood even without knowing the details in Parts II and III. On the other hand, after having gone through Part IV it may well be worth the reader’s effort to go through Parts II and III as illustrative examples for applying the methods, protocols, architectures, and tools introduced in Part IV. Finally, Part V evaluates multiagent technology and thus presents interesting new insights from different evaluation perspectives. The reader of this part should be familiar with the basics of multiagent technology as gained from Part I and Chapters II.