From Decoding to Encoding: Cognitive Reorganization of Human Coding under Generative AI

jongwon ryu, Junyeong Kim

Keywords: Cognitive Science, Cognitive Psychology, Human–Computer Interaction (HCI), Human–AI Interaction, Generative Models

TL;DR: Generative AI automates the decoding phase of coding, allowing humans to focus on what to build (encoding) rather than how to build it.

Abstract: The emergence of large generative models has reshaped human coding cognition. Traditional programming requires both encoding-conceptualizing a problem, and decoding-translating concepts into code. Generative models automate much of the decoding process, externalizing expressive cognition and allowing humans to focus on higher-level conceptual reasoning. This study examines how such automation reorganizes cognitive processes in programming. In a within subject pilot experiment , participants completed equivalent coding tasks with and without assistance from ChatGPT-5. Behavioral, verbal, and self-report data revealed greater conceptual focus and reduced low-level control effort under AI-assisted conditions. Overall, the findings suggest that generative AI redistributes cognitive load, shifting programming from an implementation task to a conceptual design activity.

Submission Number: 29