Knowledge Selection Criteria
Whereas traditional epistemologies try to distinguish "true" knowledge from "false" knowledge by postulating one or a few unambiguous "justication" criteria (e.g. correspondence, coherence, consensus), in an evolutionary context we must admit that many different influences impinge on the evolution of knowledge. It is well-recognized from observations of concrete ecosystems and from computer simulations of complex evolutionary systems, that evolution is basically co-evolution of many different systems with complicated interactions, where system A tries to adapt to B, while B tries to adapt to A. This makes it very difficult to formulate fixed and objective criteria distinguishing "fit" systems from "unfit" ones.
However, the definition of knowledge as a vicarious selector helping an organism to survive by anticipating perturbations, to some degree focuses the selection processes on the capacity of knowledge for prediction. Still, there are many different ways in which knowledge can support survival, and the predictive value can generally only be determined indirectly. This brings us to distinguish a number of different, mostly independent categories or dimensions, which each may contribute to the "fitness" of a piece of knowledge for its task, but which are not necessarily mutually consistent. The more a piece of knowledge fulfills each criterion separately, and the more criteria it fulfills in total, the better it is.
In cases where the different criteria contradict each other, no piece of knowledge can ever optimally fulfill all criteria. We rather need to look for a kind of "Pareto" optimality: different local optima, but such that we cannot combine the advantages of the different candidates into a global optimum that is best for all criteria. Whereas classical theories of knowledge would only recognize two categories of knowledge: true or false, more pragmatic epistemologies would rather order models from "more adequate" (precise, reliable) to "less adequate". In the present framework, this ordering is considered to be partial: it is not alway the case that one model is better than another one, and there is generally no best model. However, there are models which are better than other models and thus the theory avoids any "absolute relativism".
The different criteria can be categorized in three "super" classes:
Thus, the criteria embody our "holistic" understanding of knowledge, which avoids reduction to either objective, subjective or social requirements, but acknowledges the roles that each of these realms plays in the spread of a successful piece of knowledge.
- objective criteria, which measure the reliability of predictions: distinctiveness, invariance and controllability
- subjective criteria, which measure the ease with which individual subjects will accept new knowledge: individual utility, coherence, complexity, novelty
- intersubjective criteria, which measure the fitness of the knowledge with respect to the community of carriers: formality, conformity, "infectiousness" or publicity, expressivity, collective utility, authority.
Copyright© 1997 Principia Cybernetica -
Referencing this page
Sep 10, 1997 (modified)
Jul 14, 1995 (created)