RULES BASED SYSTEMS
| Collection | International Encyclopedia of Systems and Cybernetics |
|---|---|
| Year | 2004 |
| Vol. (num.) | 2(2) |
| ID | ◀ 2908 ▶ |
| Object type | General information, Methodology or model |
Any system is by necessity of coherence somehow rules- based.
However, there are seemingly at least two different kinds of rules-bases. The first type is supposed to determine strictly the behavior of the system. In A. PREM's words: “Completeness of rules-based systems means the requirement that there is a rule for each possible situation and that the given set of rules correctly solves the problem” (1992, p.1344). In such cases the set of rules is an pre-constructed and rigorously invariable algorithm. This is the general case of mechanisms and can extended with some measure of success only to a limited number of non-mechanical systems as for example chess (as observed by PREM).
Systems based on this class of rules have practically no fault tolerance: “One single missing, inadequate or damaged rule may stop the whole inference process” (Ibid. p.l345).
However connectionism offers us an alternative: networks of rules whose multiple possibilities of interactions are so highly redundant that the potential combinations are practically infinite in relation to the probable span of existence of the system. Such networks are able to construct specific or specialized algorithms to some particular ends without exhausting their capacity to produce new ones. Systems thus equipped are relatively able to adapt themselves to new contexts and to learn new behaviors.
In W. Mc CULLOCH words a rule of this kind is “… a message that could be written many places and read out differently in different context” (1974, p.8).
The point is also made by St. KAUFFMAN (1993, p.394), who observes that rules can be mutually triggered and even “coevolve with other rules by mutation, recombination, and selection”, in biological systems or in problem-solving computer algorithms, as stated by J.H. HOLLAND.