Navigational Thinking as an Emerging Paradigm of Computer Science in the Age of Generative AI

Generative AI systems produce meaning with a quality indistinguishable from - and occasionally surpassing - human performance, yet the epistemic mechanism through which this occurs remains poorly understood. This paper argues that generative AI instantiates a fundamentally new mode of knowledge production: geometric navigation through high-dimensional manifolds, grounded in indexical rather than symbolic signification. Drawing on the structural properties of high-dimensional spaces, we demonstrate that meaning in generative AI is constituted through positional relation and orientation rather than through symbolic convention. This shift corresponds precisely to what Peirce identified as indexical signification: a mode of meaning in which the sign is constituted by its real causal connection to its object, not by arbitrary assignment. We develop the pedagogical implications of this shift through a geometrized reading of Papert’s constructionism, reconceptualizing the generative AI system as a new kind of microworld - high-dimensional, non-visualizable, and indexical - in which knowledge is constructed through navigation rather than symbolic programming. From this analysis, we derive the concept of Navigational Thinking: a mode of knowing characterized by positional, enactive, and bounded engagement with geometrically structured spaces. We argue that Navigational Thinking and Computational Thinking are not alternatives, but two sequential phases of the same cognitive process: while a problem remains indexical, Navigational Thinking is operative; when the problem space stabilizes into symbolizable form, Computational Thinking becomes applicable. Vibe-coding is merely the visible tip of an iceberg - the iceberg being a new cognitive ecology in which these two modes coexist as the necessary phases of problem-solving in the age of generative AI.

💡 Research Summary

The paper argues that generative AI creates meaning not through symbolic manipulation but by navigating high‑dimensional manifolds, a process the author calls “geometric navigation.” Four structural properties of such spaces are highlighted: (1) concentration of measure, which makes distances nearly identical across points and renders metric similarity ineffective; (2) near‑orthogonality, where random vectors are almost always perpendicular, overturning human intuitions about correlation; (3) exponential directional capacity, meaning that the number of independent directions grows far faster than the number of training examples, allowing the system to generate truly novel configurations; and (4) manifold regularity, which confines meaningful data to a thin, smooth lower‑dimensional surface within the ambient space. Together these properties shift the epistemic question from “where is a concept?” to “in what direction does it point?” – a move from metric to directional semantics.

The author links this mathematical shift to Charles Sanders Peirce’s theory of signs, specifically the indexical mode in which a sign bears a real causal connection to its object. In high‑dimensional navigation, a vector’s orientation directly indexes a semantic target, providing a concrete technical realization of Peirce’s indexical sign. Consequently, generative AI operates as an indexical system at scale for the first time in computing history.

Building on Seymour Papert’s constructionism, the paper reconceptualizes the AI environment as a “geometrized microworld”: a non‑visualizable, high‑dimensional space where learners construct knowledge through exploration rather than through explicit symbol transmission. From this perspective the author defines “Navigational Thinking” – a mode of knowing characterized by positional, enactive, and bounded engagement with structured geometric spaces. Navigational Thinking is operative when a problem remains indexical, i.e., when its solution space has not yet stabilized into a fixed symbolic form. Once navigation yields a stable configuration, the problem transitions to a symbolic regime where traditional Computational Thinking (algorithmic, rule‑based reasoning) becomes applicable. The two modes are therefore sequential phases of a single cognitive process, not competing alternatives.

The paper critiques current educational responses to generative AI – accommodation (adding AI tools) and resistance (restricting AI use) – as both assuming an unchanged epistemic framework. Instead, it argues that AI introduces a new epistemic regime that demands curricula supporting both Navigational and Computational Thinking. Pedagogical implications include the need for tools that surface high‑dimensional directionality, assessment methods that capture enactive exploration, and research into how learners can develop intuition for non‑visualizable manifolds.

In conclusion, the author posits that the “geometric turn” – the migration from metric to directional semantics – underlies the unprecedented meaning‑producing capacity of generative AI. This turn makes indexical signification technically realizable, and consequently reshapes the cognitive ecology of problem solving. The paper calls for further work on visualizing high‑dimensional manifolds, measuring Navigational Thinking, and designing human‑AI collaborative workflows that fluidly transition between navigation and computation.