MIUI: Make Interaction Usable on Information Visualization for People with Visual Impairments

Visual impairments affect a significant portion of the global population, limiting their ability to interact with and interpret visual data representations. While research has been conducted on improving the readability of static charts for this group, effective interaction methods for complex, dynamic visualizations remain a challenge. This paper proposes MIUI, a novel pipeline to enhance the accessibility and usability of interactive visualizations for individuals with visual impairments. MIUI leverages a multi-agent artificial intelligence system to analyze visualizations, interpret user intentions expressed in natural language, and facilitate meaningful interactions without relying on visual cues. The system employs specialized agents to decompose visualizations into their constituent elements-such as data structure, visual encoding, and layout-creating a structured representation amenable to non-visual exploration. We demonstrate MIUI's flexibility through diverse use cases covering various interaction types and visualization techniques. A user study with 20 visually impaired participants shows high satisfaction rates in terms of accessibility and clarity.