The system described in (Gams, 1989) exploits redundant knowledge, and is involved in distributed problem solving. It admits several agents with a learning capability, but these do not really communicate while learning is in progress. Different knowledge bases are taken into account when problems are being solved. As has been shown by Gams, this mode achieves a superior performance when compared to a system containing just one knowledge base.
An important issue in this work is how to combine the opinions of different agents. Generally certain confidence factor is associated with each decision and then some method is used to generate the final decision on the basis of the individual decisions.
We notice that distributed solutions need not necessarily involve weighing opinions of different agents. If agent Ai is capable of dealing with a subset of given problems, and if this agent can be considered "sufficiently reliable", we do not need to worry about redundancy at all. The answer of one agent Ai is sufficient. As in Shannon and Weaver<<s (1964) information theory, the amount of redundancy that is necessary seems to be related to the level of noise present in the data, and the level of uncertainty introduced in its processing. This argument has been put forward by Gams et al. (1990) and is supported by experimental results.