Treröstssgränssnittet : Att förhindra röstkollaps: Design av AI-stödda läsgränssnitt för epistemisk handlingskraft
Year: 2026
Type: dissertation
Abstract: Students increasingly use generative AI tools to support academic reading, drawn by the ability to get instant summaries and explanations of complex texts. While these tools reduce cognitive effort, empirical research shows they consistently promote passive consumption over active meaning-making, leaving students with hidden knowledge deficits despite higher immediate task performance. This study argues that unconstrained chat architecture acts as a major contributing condition to this passivity. Standard interfaces place AI at the centre of the reading process, structurally conflating the author’s voice, the AI’s output, and the reader’s emerging interpretation in a single workspace. This study terms this structural risk Voice Collapse: the simultaneous erosion of the author’s intentional craft and the reader’s developing understanding. To address this, we introduce the Three-Voice Interface, a reading environment that separates Author, AI, and Reader into distinct protected zones, provides a persistent AI-free notes space, and constrains the AI to source-grounded Socratic guidance. A formative study (N = 17) and a mixed-methods user study (N = 8) evaluated the interface through interaction logs and psychometric measures of psychological ownership, self-efficacy, and cognitive load. Three conditional findings emerged. Psychological ownership was supported only where readers produced reflective notes in their own voice. Epistemic agency was driven by active curation of AI-provided guidance, characterized by high-precision engagement with AI-recommended citations rather than unguided exploration. Constraining the AI to Socratic guidance increased source text engagement while reducing generative note-taking, indicating that AI interaction can substitute for the reader’s own articulation even when direct copying is structurally blocked. These findings offer preliminary evidence that architectural design can partially mitigate Voice Collapse. Protecting learner-owned reflection spaces from AI write-access is a necessary, if not sufficient, condition for preserving the reader’s interpretive voice in AI-assisted academic reading.
Author: Syed, Sanaa
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7223162130