You are here : 
IAF Calls & Action Lines
Semantic interoperability in agentspace
(Stanislaw Ambroszkiewicz, Institute of Computer Science Polish Academy of Sciences)
Since autonomous software agents are supposed to "live" in cyberspace, they must be intelligent, that is, they must efficiently realize the goals delegated to them by their human masters. To do so they must perceive the world, interact with the world, as well as with other agents and humans. It is clear that a single agent cannot perform efficiently its tasks in a large open world without cooperation with other agents. Sophisticated agent interaction mechanisms and services are needed. During interactions the agents communicate, negotiate, and form organization structures. Hence, along the development of cyberspace, the new world (called agentspace), inhabited by the new intelligent creatures called software agents, is being created. It seems that the process automation in the Web makes the development of agentspace inevitable. Human users are situated at the border of the agentspace and can influence it only through their agents by delegating to them complex and time consuming tasks to perform.
Since the Internet and Web are open distributed and heterogeneous environments, agents and services can be created by different users according to different architectures. Interoperability is crucial to assure meaningful interaction, communication and cooperation between heterogeneous agents and services. The interoperability is not only restricted to interaction and communication, it also comprises semantic interoperability. In order to use services established by different users working in heterogeneous domains, agents have to be capable of acquiring knowledge on how to use those services and for what purposes. Hence, heterogeneous agents must exchange knowledge and understand each other.
It is clear that agents cannot interact meaningfully and understand each other if they have nothing in common. However, interaction interoperability must be assured first, i.e. to understand each other the agents must be able to interact. The idea of our research is to extract the absolute minimum (the very something in common) sufficient and necessary for achieving interoperability in Agentspace. According to our view this absolute minimum is agent interface to environment. The interface consists of three layers:
- I. The first layer, called interaction layer, is responsible for interactions. In the first step of our project it is based on Pegaz, our Mobile Agent Platform. In the final step it should go far beyond OMG MASIF and FIPA standard proposal, and be based on Java Enterprise. It assures interaction interoperability.
- II. The second layer, called representation layer, consists of a formal representation of Agentspace structure. Local events are the basis for the representation. It allows the agents to perceive and reason about the Agentspace in the same way.
- III. The third layer, called language layer, consists of an agent programming language ENTISH. A communication language is builtin, so that the agents can cooperate, form, maintain and reconfigure virtual organizations.
- Presentation slides
- e-mail: sambrosz@ipipan.waw.pl
- URL: http://www.ipipan.waw.pl/mas/stan