Thesis Proposal
Abstract
Intelligent interfaces like voice assistants and interactive reading tools increasingly use fine-grained augmentations, like elaborating on phrases or parts of images, to enhance information-dense documents. While recent advances in generative AI support their development at unprecedented sophistication and scale, designing and evaluating fine-grained augmentations in realistic scenarios remains challenging. This dissertation takes an in-depth, user-centered approach to show that fine-grained augmentations can help users understand complex documents. First, we conducted a needs-finding study by observing users cook with a commercially available voice assistant. Based on our findings, we extended an online reading interface with AI-generated fine-grained augmentations. We then observed users reading a research paper with our interface to analyze its strengths and weaknesses. Planned work will assess the effectiveness of our interface compared to a baseline, offering further insights on designing user-centered fine-grained augmentations.