Human–AI Collaborative Reasoning in Immersive Environments

Envisioned design

AI agents are increasingly taking part in sustained intelligence analysis, yet their outputs can contain unsupported claims or miss conflicting evidence, leaving analysts to verify the evidence and reasoning behind each conclusion. Current chat interfaces make it difficult to trace interpretations across the broader evidence space or to follow how a line of reasoning branches and evolves. This ongoing project develops an immersive workspace for human–AI collaborative reasoning, where a reasoning graph, a knowledge graph, and source documents are arranged in spatial layers around the analyst. In our envisioned design, analysts build and revise hypotheses by adding, moving, merging, and linking reasoning nodes, while an LLM-based agent suggests reasoning steps that they can accept, revise, or reject, each traceable to its original sources.