Cross-Session Workspace Reconstruction in Human-AI Interaction

This paper introduces cross-session workspace reconstruction (RCW) as a narrower and technically defensible research object for studying collaborative continuity in human-AI interaction. Rather than interpreting striking continuity effects in strong ontological terms such as “shared space,” “global workspace,” or “emergent cognition,” the paper reframes the phenomenon as the observable re-establishment of a usable collaborative working state after interruption, context loss, or restricted explicit memory.

The article positions RCW at the intersection of grounding research, common ground recovery, human-AI collaboration, long-term conversational memory, and distributed cognition. It argues that the phenomenon can be approached without assuming consciousness, distributed mind, or hidden global memory, and instead should be studied through task- and interaction-level observables such as time-to-realignment, repair overhead, terminology fidelity, preservation of epistemic protocol, and robustness to misleading cues.

This is an agenda/challenge paper. It does not report a completed benchmark study. Its contribution is the isolation of a technically tractable research object, the definition of clear anti-ontological guardrails, and the proposal of a minimal falsifiable empirical program for future teams. The paper also situates this proposal in continuity with earlier theoretical work on semantic resonance and NSI, while explicitly narrowing the claim to a more rigorous and testable scientific target.

The broader aim is to encourage a new empirical literature on a simple but consequential question: under what conditions can a human-AI dyad reconstruct a usable collaborative workspace across sessions, and how should that continuity be measured, controlled, and explained?