Engineering Societies in the Agents World IV

The goal of this paper is to explore the notion of complex system and, in particular the emergence phenomenon, in order to see which lessons could be learned for both understanding and designing complex software systems. Complex systems are described as sets of non-linearly interacting components making multi-agent systems particularly suitable for modelling and designing such systems. The notion of emergence is explicited and used to derive ways of understanding and designing such complex systems. We conclude by discussing the pros and cons of the emergentist approaches and the research perspectives.

In this paper we discuss some of the ways social order is maintained in animal and human realms, with the goal of enriching our thinking about mechanisms that might be employed in developing similar means of ordering communities of agents. We present examples from our current work in human-agent teamwork, and we speculate about some new directions this kind of research might take. Since communities also need to change over time to cope with changing circumstances, we also speculate on means that regulatory bodies can use to adapt.

How should open agent societies be organized? Should they be democracies, and, if so, what types of democracy? We present three normative models of democracy from political philosophy and consider their relevance for the engineering of open multi-agent systems: democracy as wise rule by an elite; democracy as the exercise of rational consumer choices by voters; and democracy as deliberative decision-making by citizens. We consider the implications of these different models for the design of open systems, in terms of the communications language, the interaction protocol, and the conflict-resolution mechanism used by the agents involved. We also consider the issue of verifiability of the internal semantics of communications languages, and argue that a model of agent democracy based on deliberative democracy provides the basis for a form of verifiability which is stronger than a social semantics.

We outline a model for a society of agents based upon one of the classical theories of sociology: The Weber’s model. We first investigate its links to agent technology through its relationship with the modeling of micro economy and the concept of expected utility. Then, some of the features of an agent society are enhanced by imposing security and validation requirements. We show that a societal concept from Sociology can be implemented through methods of Computer Science and thus is made feasible.

A multiagent system may be regarded as an artificial society of autonomous software agents. Welfare economics provides formal models of how the distribution of resources amongst the members of a society affects the well-being of that society as a whole. In multiagent systems research, the concept of social welfare is usually given a utilitarian interpretation, i.e. whatever increases the average welfare of the agents inhabiting a society is taken to be beneficial for society as well. While this is indeed appropriate for a wide range of applications, we believe that it is worthwhile to also consider some of the other social welfare orderings that have been studied in the social sciences. In this paper, we put forward an engineering approach to welfare economics in multiagent systems by investigating the following question: Given a particular social welfare ordering appropriate for some application domain, how can we design practical criteria that will allow agents to decide locally whether or not a proposed deal would further social welfare with respect to that ordering? In particular, we review previous results on negotiating Pareto optimal allocations of resources as well as allocations that maximise egalitarian social welfare under this general perspective. We also provide new results on negotiating Lorenz optimal allocations, which may be regarded as a compromise between the utilitarian and the egalitarian approaches. Finally, we briefly discuss elitist agent societies, where social welfare is tied to the welfare of the most successful agent, as well as the notion of envy-freeness.

In this paper we aim to describe dynamic aspects of social and collective motivational attitudes in teams of agents involved in Cooperative Problem Solving (CPS). Particular attention is given to the strongest attitude, collective commitment, and its evolution in a course of teamwork. During team action, the collective commitment leads to the execution of agent-specific actions. A dynamic and unpredictable environment may, however, cause the failure of some of these actions, or present the agents with new opportunities. The abstract reconfiguration algorithm, presented in a previous paper, is designed to handle the re-planning needed in such situations in an efficient way. In this paper, the dynamic logic component of the logical framework addresses issues pertaining to adjustments in collective commitment during reconfiguration.

In distributed problem solving with multi-agent systems it is assumed that collective behaviour emerges from interaction among rational agents. The dissemination of mobile agents will lead to open systems. Emergent behaviour in open systems must fulfil common goals based on goals of individual agents. Therefore, goals and tasks need to be coordinated in such a way that the desired collective behaviour emerges. At the same time, the autonomy of agents must be restricted by imposing norms on the system in order to prevent misuse and undesired effects of potential emergent behaviour. The European Privacy Directive is a collection of normative principles that can be implemented in a multi-agent system. A model for enabling emergent and normative behaviour based on social interaction paradigms is presented in this paper.

For large-scale and complex societies of agents monitoring and control are important to both agent designers and society administrators. Specifically, interaction monitoring and termination detection are of importance in optimising performance of a system and keeping users up-to-date on progress. Both monitoring and termination detection are well-studied problems for distributed object systems. In this paper, we investigate how these approaches can be applied to agent systems. We present a novel algorithm, which takes advantage of a monitor’s additional information on partial behaviour specifications of the agents to derive observable termination criteria. We sketch an implementation and qualitatively compare our novel algorithm to the existing distributed systems approaches and propose future experimental work.

Multi-agent systems considered in the paper consist of a finite number of agents, positions of which can be changed by system actions, and of an evaluation function which assigns to each agent a value of its current position (as e.g. the distance from the intended target). The set of all possible values is ordered; the intention of each agent is to reach a position with the minimum value. Any system action can decrease position values of some agents (the winners) and increase those of the others (the losers); consequently, an action execution can create conflicts among its participants (winners and losers); arbitrary resolutions of such conflicts can prevent some agents of reaching their goals. The present paper is aiming to formulate conditions that must be fulfilled by actions to guarantee each agent reaching its intended final situation.

This paper presents an approach to formally link a simple dependency-based coordination model to its related conversation protocol. Hence, by observing conversation among coordinating entities, an external observer is able to recognise valid conversations that do not break the established norms on coordination. This may help in building reliable autonomous agent based systems in open and untrustred environments.

Organisation and coordination are strictly related issues in the engineering of agent societies, and some important issues such as security can be suitable modelled only considering their synergy. Accordingly, in this paper we show how to extend a MAS coordination infrastructure for agent-based systems (namely, TuCSoN) toward the specification and support of MAS organisation. To this end, we adopt the notion of agent coordination context (ACC) as a first-class abstraction that defines the organisation structures and rules, and makes them accessible and manageable at execution time. Then, we show how ACC provides the conceptual framework to model the presence of an agent inside the organisation and the environment, as well as the means to face MAS organisation, coordination, and even security issues in a coherent and uniform way.

A multi-agent organization operating in an open environment has to adapt itself at run-time when the global knowledge of the organization changes unpredictably at run-time. In this paper this problem is named as organizational adaptability and two new patterns called as Ontology Perception and Society Merging are introduced to provide generic and reusable organizational solutions to this problem. Each pattern first defines a unique situation that causes a change in the organizational global knowledge then models the organizations’ behavior in terms of responsibilities and collaboration of participating agents to solve this problem.

The adoption of Multi-Agent Systems for system engineering often requires legacy and human-oriented services to be integrated into agent societies. In turn, this aspect impacts the engineering of interactive systems that involve the cooperation of agents and (human) actors – such as, for instance, workflow management systems. In this context, the coordination model adopted by the multi-agent system infrastructure strongly conditions the design, development and exploitation of available services.

Since its conception, software engineering has been proposing various techniques and technologies addressing the problem of maximizing the reusability of software. The leading edge of such a research is component-based software engineering. This paper reviews agents and multiagent systems from the point of view of reusability and it provides a formalization of reusability that exploits some very basic agent-oriented concepts. In particular, we move from the lesson learned in component-based software engineering and we define a topmost level of reusability in terms of two long-debated concepts: semantic composability and semantic extensibility. Then, we formally define such concepts through goal delegation, and finally we discuss how a concrete, yet simple, agent model can approximate them.

Complexity in software design refers to the difficulty in understanding and manipulating the set of concepts, models and techniques involved in the design process. Agents are sophisticated software artefacts, associated with a large number of features and therefore Agent-Based System (ABS) engineering methodologies involve considerable design complexity. This paper proposes a framework to evaluate ABS engineering methodologies against a number of criteria related to design complexity. The framework is applied to a number of representative ABS engineering methodologies. The strengths and weaknesses of each methodology with respect to the framework aspects are discussed within the context of a case study of a virtual enterprise combining manufacturing and logistics services. The evaluation results are used to motivate and guide further work in the area.

Multi-Agent Systems (MAS) are designed to solve problems, sometimes in unstable environments. Each agent in a MAS can play a set of roles; it may be prone to failure, which may cause a system failure. As a result, agents may become unavailable and some roles may be left unfilled (thereafter called missing roles). Unfortunately, there are not always backup agents ready to replace unavailable ones. In this case, the MAS must reorganize itself in order to achieve its objectives by assigning missing roles to the remaining agents. We propose an agent modelling methodology for Fault-Tolerant Multi-Agent Systems based on MAS reorganization. By taking into account the environment in the modelling phases of the system, we can anticipate different critical situations that could occur in the environment and that could have an impact on the MAS organization. Hence, we can propose an appropriate MAS re-organization so that the system continues to operate correctly. We know that it is impossible to consider all the different situations that can happen in the environment, but we can consider those that are predictable and critical.

Design patterns already proved successful in lowering the development time and number of errors of object-oriented software; now, they are, candidate to play a similar role in the MAS (multi-agent system) context. In this work we describe our experiences in the identification, production and application of patterns for agents. Some patterns are described together with the classification criteria and documentation approach we adopt. Upon them, we base a pattern reuse process that can be considered one of the distinguishing elements of the design methodology (PASSI) we use to develop MAS. Patterns can be applied to an existing agent or used to produce a new one with the support of a specific web based application that can read both the JAVA source code and XMI representation of the agent design documentation. After the successful application of the desired pattern(s), the source code and the design diagrams (usually a structural and dynamic diagram) of the agent can be exported. Some experimental results are reported in order to demonstrate the utility of this approach in automatically producing an interesting percentage of code lines.

The growth of interest in agent-orientation as a new paradigm has introduced the need for developing concepts, tools and techniques for modeling and engineering agent-based software systems. Object technology has been supporting the development of information systems for many years but is now slowly evolving to encompass more recent ideas relating to the concept of “agent”. Integrating agent concepts into existing OO methodologies has resulted in several agent-oriented methodologies, one of which is Agent OPEN. In this paper, we evaluate the existing Agent OPEN description against ideas formulated within Tropos, an agent-oriented software development methodology.

The analysis of an agent or agent communities combining advanced methods of visualization with traditional AI techniques is presented in this paper. However this approach can be used for arbitrary Multi-Agent System (MAS), it was primarily developed to analyse systems falling into Artificial Life domain. Traditional methods are becoming insufficient as MAS are becoming more complex and therefore novel approaches are needed. Our approach builds upon various techniques to deliver means for assessment on multiple levels ranging from single agent to overall properties of an agent community. Our visualization tools suite utilizes novel visualization methods together with traditional AI techniques such as sensitivity analysis and clustering. Among others it offers visualization of many agent’s parameters along time, correspondence between current/previous states (of an agent community), resulting behaviour, grouping of agents based on dominant properties etc. This transparent approach emphasizes MAS dynamics through automatic discovery of its tendency. Agent position inside virtual environment together with overview over the whole time interval adds strong contextual information to analysis. Position in our understanding is not limited to geometrical meaning, but covers also the space of dynamically changing constraints for action selection. A simulated artificial life environment with intelligent agents has been used as a test bed. We have selected this particular domain because our long-term goal is to model life as it could be so as to understand life, as we know it.

Ubiquitous, persistent and pervasive computing systems and their modelling and engineering as agent societies suggest much potential for the application of personal and task profiles. Increasing quantities and types of profiling data are also becoming available. This paper discusses the use of profiles in this context and suggests a new form of profile employment to ease the use of profile data, by increasing convenience and privacy. Familiar dimension reduction techniques map the high dimensional profile to a position in a reduced dimension profile space. However, in this method, each profile owner calculates their position with respect to a number of reference vectors. This method avoids third party involvement and keeps the profile details private. Empirical studies indicate that calculating the whole set iteratively by this method compares favourably with respect to abstraction error when compared to direct calculation.

Agents that behave maliciously or incompetently are a potential hazard in open distributed e-commerce applications. However human societies have evolved signals and mechanisms based on social interaction to defend against such behaviour. In this paper we present a computational socio-cognitive framework which formalises social theories of trust, reputation, recommendation and learning from direct experience which enables agents to cope with malicious or incompetent actions. The framework integrates these socio-cognitive elements with an agent’s economic reasoning resulting in an agent whose behaviour in commercial transactions is influenced by its social interactions, whilst being motivated and constrained by economic considerations. The framework thus provides a comprehensive solution to a number of issues ranging from the evolution of a trust belief from individual experiences and recommendations to the use of those beliefs in market place level decisions. The framework is presented in the context of an artificial market place scenario which is part of a simulation environment currently under development. This is planned for use in evaluation of the framework, and hence can inform design of local decision making algorithms and mechanisms to enforce of social order in agent mediated e-commerce.

The best way to evaluate a new technology such as the agent-oriented programming paradigm is to test it in the real world. In this article, we illustrate how multiagent systems can be deployed to analyse, design and implement a location- and context-dependent information system in a shopping mall. Our goal in this application was to help people by making personalised information available where and when it is needed in a way that disturbs them as little as possible and protects their privacy as much as possible. By employing the VPC communication framework on the KODAMA agent platform, we were able to build a shopping-support system as a collection of interacting, autonomous, flexible agents, with support functions capable of dynamically adapting services to client location and preferences as well as environment changes. Here we will give a close view of the system, examine application scenarios and discuss the pros and cons that emerged from the results of a large-scale experiment.