In this section we will describe the method of knowledge integration in more detail. Basically, the process involves a preparatory phase in which a group of systems (agents) is selected and organized into a group that can function together. The group can include systems S1-Sn that are capable of constructing theories from data, responding to external queries and communicating with one another. Here we shall assume that this has already been done and that the system organization is fixed[2].
Having defined the organization of the multi-agent system, it is necessary to determine the overall objectives. That is, for example, one can define which concepts are to be acquired and/or describe in some way the required performance. This is important if we want the multi-agent system to decide when to stop altering its theories. Here we will follow what has been done in the past and let the user control this issue. Consequently, here we will be concerned with, basically, the following three phases:
(1) Generation of independent theories (by consultation or inductive learning),
(2) Competitive characterisation of system<<s theories,
(3) Construction of the integrated theory.
In phase (1) the systems S1-Sn work in an independent manner, and as a result produce theories T1-Tn. Each system involved constructs its own theory on the basis of its own experience. Here Si can represent either a human, or an inductive learning tool. In either case Si will produce theory Ti.
In phase (2) the individual theories are characterized using tests. Without loss of generality let us assume that this is actually controlled by some agent SI. This agent poses a query to all the agents involved, waits for the answers and then proceeds with the next query. Any of the systems S1-Sn could act as SI. The subsystem responsible for characterization of theories is referred to as INTEG3.1.
Phase (3) is dedicated to the issue of constructing one integrated theory (TI) on the bases of the results obtained in phase (2). This task is also done by INTEG3.1.
The three phases mentioned could be followed by two additional ones:
(4) Adoption of integrated theory by one (or more) systems,
(5) Check whether the process should continue, and if so, go to (1).
In this paper we will be concerned mainly with the phases (1)-(3). The issue of how one could construct a 'closed loop system', capable of taking the integrated theory and using it input in further learning will be discussed in a future paper. The next section describes the phases (1)-(3) in more detail.