Abstraction Based Information Technology

ion in AI and CS In the early days of CS, algorithms were often seen as an abstract formulation of procedures. Nowadays, this workshop illustrates well that category theory is associated to an abstract formulation of CS. The example of computer algebra teaches us that such methodologies do not amount to a full abstraction paradigm. Giunchiglia and Walsh introduced theories of abstraction for mechanized reasoning in a context of logic. This was probably the first “reasonable” and complete attempt at abstraction in AI. Giunchiglia and Talcott built on this approach to design the Open Mechanized Reasoning System. Calmet and Homann used a similar approach for computer algebra. The two attempts were combined for symbolic computation systems (OMSCS). Relevant references are found in [1, 2]. Recently there has been a renewal of interest in abstraction in AI through the SARA workshops held at IJCAI 2005 and 2007. The open mechanized approach defines an abstraction as a three levels architecture: Theory, control and environment. One needs first to have a theory (algorithms for instance), then one adds control (a programming language for instance) and finally one considers the environment in which this controlled theory will be active. It is possible to prove that computations do terminate, are unique and correct. Extension to numerical computations has been investigated [3]. Although the Open Mechanized approach looks mainly meaningful for computational processes, its three levels structure is suitable to abstract the concept of multi-agent systems (MAS) when one adopts the Agent Oriented Abstraction (AOA) paradigm. AOA consists of 6 definitions [2]. The first four ones deal