General Intelligence and Hypercomputation
Selmer Bringsjord · 2009
From its revolutionary arrival on the behaviorismdominated scene, the information-processing approach to both understanding human intelligence, and to the attempt to bestow such intelligence on a machine, has always proceeded under the assumption that human cognition is fundamentally computation (e.g., see (vE95) in connection with CogSci, and (Hau85) in connection with standard AI).But this approach can no longer leave it at that; it can no longer rest content with the coarse-grained creed that its allegiance is to the view that intelligence consists in processing information.There are two reasons: First, AGI is now on the scene, insisting, rightly, that general intelligence should be the focus.Second, ours is an age wherein the formal science of information processing takes quite seriously a detailed mathematical framework that generates a difficult and profound question: "What kind of information processing should those interested in general intelligence take the mind to rest upon, some form of standard, Turing-level computation, or that and hypercomputation?"My question is not to be confused with: "Which form of Turinglevel computation fits best with human cognition?"This question has been raised, and debated, for decades. 1 A recent position on this question is stated by (LEK + 06), who argue that Turing-level neurocomputation, rather than Turing-level quantum computation, should be the preferred type of information processing assumed and used in cognitive science.Recall that computation is formalized within the space of functions from the natural numbers N = {0, 1, 2, . ..} (or pairs, triples, quadruples, . . .thereof) to the natural numbers; that is, withinThis is a rather large set.A very small (but infinite) proper subset of it, T (hence T ⊂ F), is composed of functions that Turing machines and their equivalents (Register machines, programs written