Reviewed by
Frederic Amblard and Matthias Mailliard
CNRS-IRIT, Université Toulouse 1 Sciences Sociales
These two books are the first two post-proceedings of a series of three workshops named "Environment for Multi-Agent Systems". The E4MAS workshops, organized together by Danny Weyns, H. Van Dyke Parunak and Fabien Michel, have been apart of the satellite workshops of the AAMAS conference from 2004 to 2006. The workshops original idea was to focus on "the Environment", which is a concept of paramount relevance in multi-agent systems according to methodological works. But it happens that it is not much taken into account when approaching more concrete applications. In this view, the leitmotiv of these workshops was to consider the environment as a first-class entity for the modeling of multi-agent systems. According to one of the E4MAS organizer, the workshop series ended hereafter the third edition when organizers and participants all together felt that it was time to disseminate to a broader audience the idea of considering environment as one of the first component to design while modeling MAS, sometimes even before than the Agent component.Generally speaking, the main pitfall of lecture notes is the lack of homogeneity of the papers. In many cases, such books are not much than a collection of articles written by authors who met around a general topic and they reflect more the individual contributions to the meeting or workshop than the collective exchanges, which sometimes happened during the event. This is unfortunately applicable also to E4MAS post-proceedings, the question of the environment for multi-agent systems being very broad and putting together very different representations as well as preoccupations. Scientists interested in the environmental component for Multi-agent based simulations and researchers interested in the environmental brick for problem solving using multi-agent systems have rather radically different views on the concept (the one trying to render or model a realistic environment for his system, the other designing environment as an expression of much of the problem constraints). To be honest, the lack of homogeneity is not in all, as this collection of articles from different application fields are linked, on the one hand, through methodological articles that have a global focus on the concept and, on the other hand, through the survey article in the first volume which was written hereafter the first workshop. Then, the reader can have an idea of the questions asked from the application domains and gain some perspectives by reading methodological papers that could be applied to different cases. But having a look at the program of the workshop (see: http://www.cs.kuleuven.be/~distrinet/events/e4mas/), where most papers presented can be downloaded, nearly half of the time was dedicated to discussion... And looking at the papers published in the post-proceedings, at the exception of the survey article in the first issue, there is no feedback concerning these discussions.
Before going into much detail, we have to notice that the potential interest of these post-proceedings for the JASSS readers community is limited. Environment is of course a first-order entity of many models when the spatial dimension appears to be important to take into account. But real challenges in this field are only evocated by the post-proceedings:
You would have guess that the E4MAS post-proceedings are much more appropriated for researchers interesting in Multi-Agent Systems rather than Multi-Agent Simulation, even if these latter could find some inspiring works, like conceptual schemes, for taking the environment into account in a generic way.
But even with this particular focus, the E4MAS workshops could be seen as a retreat of distributed solution compared to centralized ones. This observation is not valid for every work presented. But, reading these two first issues, we can observe that there are an increasing number of solutions presented in which more and more complexity is put in the environment and less and less is let to agents' behavior and interactions. The centralized environment gains then in information compared to the agents that look more and more like particles (after all the Particle-Swarm Optimization field is not so far from MAS).
The book is then structured into five parts:
The Conceptual models part is of interest, since the models presented have a part of generality that could be potentially applicable to simulation approaches, even if some of them focus on very specific applications like ubiquitous systems. The first article (by Ferber, Michel and Baez) proposes an extension of the Agent-Group-Role framework, embedding into the Environment (AGRE) a part of the constraints put on the agents' organization. The authors tend to generalize the physical perception/action of an agent to the organization ("an organization is a world where groups are space and where agents perceive and act through their role"), with roles being effectors and sensors in the social world. Such a view could therefore be interesting for organizational design in simulations, at least at an analogical level. The context of the second article (by Chang, Chen, Chien, Kao and Soo) is quite different. The authors aim at proposing an ontology for describing and reasoning about the environment for cognitive (BDI) agents. They suggest to articulate models at different levels: a "reality model" representing the physical laws and properties; a "concept model" that conceptualizes the reality model trying to infer relations among concepts from the observation of physical properties; and finally a "mind model" which reacts to concept model in terms of consequences for the perception (e.g. feelings) and for the action. The last article (by Bandini, Manzoni and Vizzari) deals with a communication model, conceived as a multi-layered system with interfaces among layers, that is spatially dependent for ubiquitous systems.
The Languages for design and specification part is composed of two articles. The first one (by Okuyama, Bordini, and da Rocha Costa) presents ELMS, an XML language to specify MAS objects, their attributes and methods, as a part of the MAS-SOC (Multi-Agent Simulations for the SOCial Sciences) project. It seems very promising, but the example taken looks really far from Social Sciences (just like the first model of Sugarscape). The second article (by Gouaïch, Michel, and Guiraud) presents MIC*, that is, an instantiation of the social part of AGRE focusing on deployment environment.
The Simulation and Environments part consists of two articles. The first one (by Klügl, Fehler, and Herrler) conceptualizes environment as a first order object. The authors present a good description of basic categories of MA-Simulation regarding environment. They also present in a broad view the different kinds of representation of the environment for simulations: topological environment, discrete and continuous ones, GIS-based, and so on. The second article (by Lees, Logan, Minson, Oguara, and Theodoropoulos) deals more with technical aspects, and it is focused on the question of distributed simulations using HLA (High-Level Architecture ).
The Mediated Coordination part is the more interesting part of the book but also the more distant to simulation approaches. It is mostly dedicated to Multi-agent systems used in problem-solving. The global idea behind the three articles presented is to make the environment support the constraints of the problem. Consequently, simple agents evolve in this complex environment, so as to try and make a solution emerging out of this. The first article (by Julien and Roman) proposes a protocol for mediating mobile agents interactions, which depends on the context, with an application to ubiquitous computing. The second one (by Ricci, Viroli, and Omicini) deals with the environmental-based coordination through artifacts in order to endorse the environment for cooperative & social activities. The last one (by Tummolini, Castelfranchi, Ricci, Viroli, and Omicini) illustrates a coordination mechanism by behavioral implicit communication.
The Applications part is of interest for presenting the potentials of the environmental focus in different application fields. The first article (by Brueckner and Parunak) uses "swarming" for the detection of patterns and their classification over a graph. The two others (the first by Parunak, Brueckner, and Sauter, the second by Mamei and Zambonelli) illustrate applications in potential fields, like coordination of vehicles, in the first case, and the Quake 3 game, in the second one.
In the Models, architecture and Design part , the four papers present an interesting evolution compared with the first issue. The first article (by Weins, Vizzari and Holvoet) is dedicated to environments for situated multi-agent systems. The authors propose a three-layer architecture with a particular definition of the environment at each stage, but possibly the same organization: 1) MAS application logic, 2) Software execution platform, 3) Physical infrastructure. The second paper (by Rodriguez, Hilaire and Koukam) presents the holonic modeling approach. Holons are autonomous entities that are apart of the whole. They present an application to the traffic simulation in a firm, considering the environment as an active entity, parts of the environment exchanging vehicles during the simulation. The third article (by Simonin and Gechter) proposes a generic methodology based on the environment to conceive reactive MAS for problem solving. The last article (by Steiner, Leask and Mili) presents interestingly an architecture for MAS Simulation environments, in which the environment is a network of cells, and the different levels of abstraction correspond to a hierarchy of cells. The authors suggest, in order to reduce the coupling between the environment and the agents, identifying the agents duties and responsibilities towards the environment. This results in the DIVA architecture that is dedicated for geographical based simulation system.
The Mediated Coordination part is, as expressed before, the big part of this volume, regrouping eight articles. The first one (by Keil and Goldin) introduces an interesting distinction between indirect interaction through the environment, and the role of the environment in message transport. The authors suggest different classes of interactions (sequential, multi-agent, direct, indirect) and environments (physical, virtual, persistent, amnesic, dynamic, static), which enable to classify MAS and applications. The second article (by Schumacher and Ossowski), entitled "The Governing Environment", could have appeared in a more conceptual part, since it starts from the distinction between objective (inter-agent) and subjective (intra-agent) coordination by underlining the need of infrastructures for objective coordination, and by envisioning environment as a regulating system that would include norms and laws on security, communication, organization, time and history. The third article (by Malucelli, Cardoso and Oliveira) proposes to enrich the MAS environment with institutional services. The proposed environment would be more than a communication mediator with white and yellow pages, including institutional-ontology based services. It would be a kind of normative environment. The following article (by Platon, Sabouret and Honiden) presents an overhearing model of indirect interaction in which the environment entity handles overhearing rather than message broadcasting. The authors illustrate their approach with an application to electronic market. The fifth article (by Klein and Giese) is more methodological, proposing to design the environment first, then agents that are capable to manipulate and perceive the environment. The authors proposed a four step process composed by: a) analysis modeling requirements and the environment; b) social design: designing communities and services; c) agent design: designing agents behaviors and internals; d) deployment: replacing the simulated environment. The following article (by Parunak) presents a survey concerning environments and mechanisms for human-human stigmergy, the coordination of agents through signs they make and sense in a shared environment. The author proposes a taxonomy of stigmergetic interactions and emphasizes many examples through a reading grid composed of the "environment" describing its topology, states and dynamics; the "agent" with its sensors, actuators and dynamics and the "emergent" system behavior. The seventh article (by Mamei and Zambonelli) is dedicated to pervasive computing by augmenting the physical environment through embedded wireless technologies (RFID), using pheromone-based object tracking. At last, (Valckenaers and Holvoet) propose an environmental-based solution to manage complexity in a MAS-based manufacturing control.
The Applications part begins with an application to automatic guided vehicles with collision avoidance (by Weyns, Schelfthout and Holvoet). The second application (by Bandini, Manzoni and Vizzari) aims at capturing visitors' behavior on a website (the environment) and at proposing of new kind of interactions among visitors. The third one (by Messic and Oh) presents the adaptation of agents' roles using polymorphism or specialization as directed by the environment. Polymorphic agents are then capable to adapt roles based on their perceived environment. The last one (by Hellboogh, Holvoet and Berbers) is again an application to automatic guided vehicles.
In conclusion, these workshops are quite interesting as they allow putting attention on an important entity in MAS in general, that is, the environment, which was not much enlightened before. They succeeded, since, in between the two issues, they seem to have regrouped a community of scholars in majority interested in understanding problem-solving issues. Although the proposed solutions, which aim at putting more and more information in the environment rather than in the agents, perhaps is a good one (to find efficient solutions), it looks like a return to centralized solutions in a way. Or, at least, it seems that this focus on environment leaves agents that simply looks like particles.
Anyway, there is something of interest for social simulation researchers in these two volumes. The first interest concerns the conceptual models proposed. Social simulation researchers could find interesting conceptual models for their own applications or at least a source of inspiration in order to find new ways of taking into account the environment in simulations, especially in geography. The second is a second-order interest. The E4MAS workshops point out that focusing on agents' environment is a fruitful solution. In social simulation, the environment of the agent is not much a spatial one, as it is illustrated in the applications parts of E4MAS, but rather an organizational one. Perhaps, a fruitful inspiration for social simulation researchers should be to reason about organization as a first-order modeling entity.
References
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2007