Evolutionary Development of Knowledge Systems: a theoretical and empircal study in the framework of the Principia Cybernetica Project.

Aim: This project purports to model the development of new knowledge systems, starting from an evolutionary-systemic analysis based on the recombination and selection of rules. A list of theoretically motivated and (mostly) empirically measurable selection criteria is proposed. The adequacy of the model can be tested by analysing the computer-supported construction of a conceptual system in the Principia Cybernetica Project.

Background: My long-term, individual research project should be situated as part of the Principia Cybernetica Project (PCP) [10,19,42,49,53], of which I am one of the editors (together with V. Turchin and C. Joslyn). PCP is a computer-supported collaborative attempt to develop an integrated evolutionary-systemic philosophy or world view. Its contributors are distributed over several continents. They keep contact mainly by electronic mail (mailing list PRNCYB-L), annual conferences, and a Newsletter. PCP focuses on the clear formulation of basic concepts and principles of the cybernetic approach. This leads to a semantic network of nodes and links, implemented as hypertext on the World-Wide Web system [19,73]. The PCP philosophy is based on the self-organization of multi-level systems through variation and natural selection [42,11,21].

Theoretical assumptions and objectives: I would like to apply this selectionist theory to the development of knowledge. The aim is to develop an integrated theory of the variation and selection of knowledge, based on Campbell's evolutionary epistemology, Holland's classifiers and genetic algorithms, Edelman's neural Darwinism, Dawkins' memes, and basic principles from cybernetics and systems theory [47]. Knowledge systems can be modelled as sets of connected production rules. Production rules consist of distinctions (A, B) [4,6,7], connected in the form of an elementary prediction ("if A, then B") [13,6,44]. Recombination of rules produces a variety of potential knowledge systems. I have developed a preliminary list of fitness criteria or selection criteria [50], which determine the likeliness that a newly generated piece of knowledge will be accepted, maintained and reproduced, in an environment consisting of other knowledge systems, carried by the same or different individuals. The project aims to further refine, enrich and test the adequacy of these criteria.

Objective criteria: distinctiveness (detectability of effect) and invariance (of effect over time, situation, person, ...). These criteria determine the generality and reliability of causal predictions made by a knowledge system [4,24,25].

Subjective criteria: utility or contribution to individual fitness [45](survival and reproduction) of the carrier of knowledge, and coherence (ease with which new knowledge is assimilated into already established knowledge, cf. Thagard, 1992).

Intersubjective or social criteria: expressivity (ease with which knowledge can be expressed in language), formality [13,17,26,28](degree to which the interpretation of that expression is unambiguous, i.e. independent of context), contagiousness [16](degree to which a carrier of knowledge is motivated to transmit that knowledge), consensus (number of carriers of a piece of knowledge).

Finally, there is the requirement of minimal complexity [12,44,40,51].

Empirical tests: Most of these criteria can be empirically measured. F. Van Overwalle and I have designed a number of psychological experiments that check the role of distinction invariance in de induction of causal explanations [24,25,41]. In these experiments several variant or invariant causal situations are presented to experimental subjects, after which their answers to questionnaires or computer-controlled tasks are statistically analysed. Thagard (1992) has formulated a number of principles, and has implemented them in the computer program ECHO, which computes the coherence of hypotheses with a given conceptual network. This model has been used to reconstruct and explain a number of historical scientific revolutions. These principles are related to a number of principles which I have formulated concerning de reduction of cognitive complexity in similar networks [44], but these approaches should be further integrated. J.M. Dewaele and I have discovered that the average formality of knowledge, expressed in natural language, can be measured by means of the frequencies of different word categories (nouns, adverbs, etc.) in the text [23,26,27]. This measuring instrument should be further refined and tested on different types of material. It is probably possible to develop a similar linguistic measure for the expressivity, viewed as "information density" of a text. Consensus can be easily operationalized through a questionnaire which asks participants in how far they agree with a given proposal.

Implementation through Computer Networks: The Principia Cybernetica Project (PCP) proposes an excellent source of material to which the above methods can be applied. Indeed, all knowledge produced within PCP is stored on computer in a structured hypertext format, including author's name, date, version and additional data, and can thus be easily analysed [73,19]. Moreover, the use of electronic networks makes the exchange of knowledge between the participating scientists much faster and easier. The net effect is that the processes of scientific discussion and theory construction are "magnified" for the observer.

The PCP conceptual network can be used via the World-Wide Web hypermedia system over the global Internet. At present, the PCP server [73] contains some 500 documents. On average, every day some 40 different users, from all parts of the world, consult in total some 300 nodes on the server. Soon, users will not only be able to passively read information, or to send it by electronic mail, but also to interactively add new ideas, by filling out electronic forms. Thus, different authors will be able to elaborate or criticise preliminary texts, and to link new arguments, producing a network of arguments, counter-arguments, and provisional syntheses.

One could then measure the formality or expressivity of a contribution, the degree to which it is coherent with already present propositions, the complexity of the resulting cognitive system, or the degree of consensus about it. On the other hand, a detailed history of the developments will be maintained. The comparison of history with measuring results will then make it possible to test in how far the proposed selection criteria provide adequate explanations or predictions of the long-term developments of the cognitive system, and to complement the theory by practical observations. Finally, the theories of knowledge development elaborated in that way may be implemented in the form of computer programs that make the further co-operative development of knowledge more efficient [13,36].


[1,2,3...] numbers in square brackets refer to items in my publication list (appendix A)