ReSPIRE: Transparent and Steerable Human-AI Sensemaking through Shared Workspace

Sensemaking is a cognitively intensive task that can be enhanced through directly manipulated visual workspaces. However, analysts require significant time to extract and summarize information from these workspaces. Can Large Language Models (LLMs) be leveraged to learn and summarize human cognitive efforts while also providing analytical hints for sensemaking? While LLMs offer potential benefits, concerns regarding model transparency and hallucinations remain unresolved, particularly in sensemaking tasks where accuracy is critical. To address these challenges, we propose a human-AI sensemaking framework that employs a visual workspace as a shared ground for human-AI collaboration. In this framework, the workspace guides precise LLM summarization and presents highlighted LLM keywords, enhancing human understanding of the summarized results and enabling users to provide feedback through these highlighted keywords. Furthermore, the framework completes the sensemaking loop by using automated summarization to compile facts and evidence, supporting higher-level iterative evaluation and refinement. Building on this framework, we developed ReSPIRE, an interactive system designed to demonstrate its feasibility and effectiveness. Through the user study, we demonstrated that ReSPIRE enhances human sensemaking by significantly reducing workload, improving efficiency, and supporting incremental formalization. Our work advances human-AI collaboration for sensemaking by providing a more effective approach for humans to leverage the reasoning capabilities of LLMs with greater flexibility and precision.