MagentA Home

Why Multi-Agent Software?

Business in the context of Global Economy is too complex and ever changing to be modelled or controlled using traditional, rigidly structured software, such as the Enterprise Resource Planning (ERP) system. Elements of the world market are now richly interconnected, making it impossible to draw clear boundaries between them. A growing number of businesses organise themselves into networks of self-contained units, which co-evolve with other businesses, in response to market dynamics. Examples of closely linked but changeable processes occur widely, yet current information systems are unable to respond flexibly and economically to their needs.

Multi-agent systems and their benefits

One way to develop software capable of blending seamlessly into current dynamic business environment is to design it as an agent-based distributed system. The key element of such a system is a software agent, that is, a software object capable of analysing a situation, making decisions and communicating with other agents and users. These capabilities distinguish agent-based systems from traditional, rigidly structured systems, providing them with the important features of autonomy, self-organisation and evolution. In multi-agent software, components of a program are able to act autonomously in performing their tasks; controlling their own activities; initiating a dialogue with a user at unspecified moments and working under conditions of uncertainty. Multi-agent software is particularly suited for dealing with fluid problems such as the allocation of resources and matching supply to demand.

Currently, the capabilities of existing agent-based systems, mainly developed on the basis of Java classes, are still limited and mostly oriented to e-commerce and Internet searching. The development of open multi-agent systems faces a range of serious obstacles, resulting from the lack of universal methodological principles and development tools; the absence of a unified view concerning the principles of organisation and functioning of agents; situation models (ontology); memory structure; reasoning mechanisms and methods of interaction.

To be used on a mass basis, the multi-agent systems need to run on all the standard computer platforms; be networked and distributed; applicable both to corporate networks and the Internet; open for all registered users and be easily, flexibly and dynamically updateable. They should access and apply heterogeneous corporate knowledge; execute a large number of agents and provide a high-speed service.

Moreover, complex multi-agent systems are required to constantly 'live' on corporate servers and/or workstations and continuously participate in the process of problem solving rather to be activated from time to time. Such systems should offer the user the capability of capturing new knowledge and information 'on the fly', without interruption to their work. They should be able to save history, extract new knowledge from it and change their behaviour in response to this newly acquired knowledge.

The development of self-organising and evolving systems goes far beyond the limits of current computer systems, and is closely connected to the notion of emergent intelligence (spontaneous and unprompted) that goes back to concepts developed by M. Minsky in the 1980s.

What are the benefits we can expect?

    Multi-agent software is autonomous, self-organising and evolving, so that it can adapt dynamically to the changing business environment or business strategy rather than to be re-programmed in order to change its functionality.