Perspectives: a relativistic approach to the theory of information

Abstract

This thesis is concerned with the elucidation of the structure of three basic cognitive functions. Firstly, an organism must be able to make distinctions between different aspects of its environment if it is to respond selectively. This is classification. Secondly, it must be able to anticipate conditions in other parts of its environment. If an organism at x is to anticipate that the condition c holds at y then, at the very least, the information that c holds must be accessible from x. Hence anticipation depends on a flow of information. Thirdly, an organism must be able to recognize uniformities across different parts of the environment. This is individuation. We propose that each of these functions can be understood in terms of a primitive ability of `seeing' the world from a perspective. In contrast to the possession of a conceptual scheme, or mastery of a language of thought, a characteristic of an organism's ability to adopt perspectives is the additional ability to shift from one perspective to another. In the thesis we first propose a theory of classification. Its usefulness in categorizing different classificatory systems, like taxonomies, state systems and attribute-value structures, is demonstrated in the Appendix. We then study two approaches to characterizing the flow of information. One, due to Dretske (1981), is based on conditional probabilities. The other, due to Barwise and Perry (1983), is based on the Situation Theoretic idea of a constraint. Our theory of perspectives takes ideas from both accounts: from Situation Theory, the distinction between information supported and information carried by a situation, and from Dretske, the implicit relativity to an information channel. We give a rudimentary account of the individuation of objects as predictive regularities across situations. Properties of objects individuated in this way are characterized as shifts in perspective which preserve the predictive regularity. Finally, we consider a more concrete model of information flow (called a world system, Rosenschein (1989)) in which environmental conditions are understood in terms of possible state distributions over locations and times. We generalize his model and show how information channels offer a more sensitive account of information flow than the one induced by the global notion of possibility. Information channels are then used to construct perspectives within a world system.