The Cost of Self-Distortion in Human-AI Interaction: An Authenticity-Based Cost Model of Expression Distortion and Self-Alignment

Abstract

As large language models become embedded in decision support, knowledge work, and everyday reasoning, interaction quality is often attributed primarily to model capability. This article argues that a critical and underexplored determinant lies on the user side: the fidelity with which internal cognition is represented in language. Two concepts are introduced. Expression Distortion Cost (EDC) captures the performance loss, decision inefficiency, and interaction overhead that arise when expressed input diverges from underlying goals, constraints, and preferences. Self-Alignment Load (SAL) captures the cognitive effort required to identify, structure, and articulate those internal states in a form usable by AI systems. Building on research in human–AI interaction, communication theory, metacognition, and cognitive load, the article develops an Authenticity–Cost Model in which interaction outcomes depend on the trade-off between distortion and alignment effort. The central claim is not that authenticity is a moral ideal, but that AI systems impose a functional penalty on distorted representation. As the cost of external information access declines, the primary constraint in effective interaction increasingly shifts toward internal representation and self-processing. This reframes human–AI interaction as a co-constructed process in which outcome quality depends jointly on model capability and user-side representational accuracy. The article concludes by outlining implications for human–AI interaction research, interface design, prompt literacy, and the distribution of advantage in AI-mediated environments.

Keywords: human-AI interaction, large language models, prompt engineering, cognitive load, metacognition, self-alignment load, expression distortion cost, self-representation