A knowledge-based system is an AI system in which knowledge domains are represented and reasoned with. A relatively new development is that of agent-based systems in which components called 'agents' reason autonomously with their knowledge and take their own decisions to act.

AI research in ATIA

At ATIA we aim at developing a combination of knowledge-based and agent-based systems for applications in, for example, the medical domain.

We aim at devising advanced knowledge-based systems that support the use of heterogeneous knowledge sources and in which different computational and logical methods to reason about this knowledge are integrated. The knowledge has to be described explicitly in a format that is accessible to computers. The facts and the relationships between these facts, together with the structure of the facts and relations will need to be captured in a formalized manner such that algorithms can have access to the information in this knowledge. The domain knowledge can then be used by computers to compute new relationships or infer missing or inconsistent 'facts'.
To construct domain models and knowledge-based applications, Protege software is used by developers and the academic community. Protege is a free open-source platform which implements knowledge-modeling structures and actions that support the creation, visualization, and manipulation of ontologies in various representation formats. To allow the integration and/or linking of several domains, basic structures will have to be defined which make it possible to combine different information in a unified and consistent manner.

We propose to do this using a component-based architecture, called a (heterogenous) multi-agent system (HeMAS). The components are called agents, which are software entities that reason with knowledge, make decisions and come to a conclusion. When devising
these systems we have to take into account various requirements such as those from modern software engineering in order to render a system that is e.g. adaptive, modular, reusable, extensible, efficacious, and user-friendly. Since the resulting system is complex, dealing with complex forms of reasoning, it should also be able to explain results to the user
in user-friendly terms.
Attention should also be given to the aspect of heterogeneity. Since the agents may use their own knowledge representation (ontology) and reasoning method, the overall architecture of the MAS should be able to connect various different agents in a convenient way. To this end it seems reasonable to devise a uniform interface for communication with the agents.

The main objective of the knowledge representation facet is that knowledge is described explicitly, yet domain experts will remain in control of the knowledge, that is to say, the human expert can have the overview of the correctness of the knowledge. At least two success dimensions can be distinguished: the amount of knowledge that can and will be represented, and the human experts' ability to overview the knowledge.