Algorithms for rational thought : an innate-language approach
Bowen, Paul Davis
MetadataShow full item record
A class of algorithms for rational thinking is presented. These algorithms systematically generate hypotheses in a syntactically rich formal language (the "innate language"). Thus thinking is portrayed as, in a restricted sense, a linguistic activity. Ryle has argued that thinking cannot be linguistic because language presupposes thought, which would lead to an infinite regress. The present theory can escape this, however, because the algorithms themselves are the only presupposition of innate-language thinking, and they presuppose no more than any algorithm does, i. e., a set of states and transition rules. As the hypotheses are generated, they are evaluated in three ways. First, they are tested for explicit contradictoriness. Second, they are tested for incompatibility with sentences known to be true from experience. This latter sort of sentences are very limited in content; they are true by virtue of being extensionally equivalent to a statement to the effect that they are believed, and, as such, have no invariant meaning. They are not translatable into conventional (i. e., public) language. Conversely, all conventional-language reports are theory-laden. These first two tests make an absolute decision to accept or reject an hypothesis, but the third test assigns a numerical "degree of belief" to hypotheses on the basis of fruitfulness, novelty, and simplicity. The same degree of belief may be assigned to each of any number of incompatible hypotheses. Also, subtle contradictions may escape the testing procedure. It is therefore quite possible for rational beings on this model to have erroneous beliefs. Indeed, since experience is not in general exhaustive enough to decide between rival hypotheses, there are no absolute "right" or "wrong" views for a rational being to hold, given a certain experiential biography. In fact, there is no guarantee that a rational being will ever arrive at the truth of a given matter; but then, this is a realistic picture of the situation of a rational being, who can after all never know that he does not inhabit a universe which is not connected with respect to causal chains. Since there are an infinite number of algorithms in the class under consideration, there is a certain relativism to rational thought. I argue that this is not a vicious relativism because convergence between any two rational beings is always possible, regardless of difference in algorithm or previous experience. I also argue that the theory is transcendentally consistent, i. e., it does not contradict its own presuppositions.