Computer Science Beyond the Instrumental Paradox

Abstract

Computer science confronts a strange moment: the technology unsettling its foundations is its own creation. Generative AI—built from the discipline's theory, algorithms, architectures, and data—now produces the very artifact around which the discipline organized itself: executable code. The dominant response is instrumental: adopt the technology as a faster way to program. This article argues that the instrumental response, though necessary, mistakes the nature of the change. Generative models organize much of their behavior in a learned, high-dimensional semantic space—the first practically decisive departure from the symbolic paradigm, in which every semantically significant distinction had to be stipulated by a person. That space supports context-dependent navigation among plausible continuations; when navigation becomes recurrent, coupled to action and feedback, and organized around a persistent goal, it constitutes operational agency, a property of an architecture rather than a claim about consciousness. An object organized this way cannot be exhaustively pre-specified, and the corresponding human competence—Navigational Thinking, complementing Computational Thinking—consists of orienting, constraining, stabilizing, and verifying a field of possible behavior rather than authoring a complete transition sequence. The article makes this change precise in four claims—on representation, on agency, on the engineered object, and on Navigational Thinking—and derives from them a reorganized computer science curriculum.Preprint. A version of this article is under consideration for journal publication.