HIPA: A Human Information Processing Architecture for LLM Responses in Ergonomic Design

Abstract

Large language models (LLMs) are increasingly used for UI/UX evaluation and design decision support; however, their responses are often generated as implicit and unstructured outputs, making cognitive processes and judgment rationale difficult to trace. Moreover, the quality of LLM-based judgments may vary depending on users’ prompting proficiency, limiting the consistency of AI-assisted evaluation. This study proposes HIPA (Human Information Processing Architecture), a cognitive response architecture that structures LLM outputs based on the human information processing (HIP) model and investigates the manifestation of human cognitive stages in conventional LLM responses. HIPA maps four HIP stages—perception, working memory, decision making, and response execution—into a two-stage response structure consisting of information provision and judgment support through knowledge-grounded prompting with GPT-4o. An exploratory evaluation was conducted with seven participants with different prompting proficiency levels who assessed the same mobile login interface using a general LLM. Responses were coded according to HIP-stage manifestation using a three-point scale. The results showed that perception and response execution were consistently observed, whereas working memory and decision-making stages varied depending on prompt systematicity. These findings indicate that general LLM responses do not inherently provide cognitively structured judgment support. HIPA provides a human-centered framework for developing AI-assisted UI/UX evaluation systems based on human cognitive principles.