INSTAN: A Framework for Measuring Joint Thinking in Human–AI Dialogue

Current AI evaluation frameworks optimize for correctness, safety, and post-hoc user satisfaction. Yet practitioners regularly observe a qualitatively distinct phenomenon: moments when a dialogue with an AI system suddenly becomes alive — thought clarifies, direction emerges, and continuation feels natural. This paper introduces INSTAN, a framework describing dialogue quality as the product of two dimensions: Energy (E) — engagement and conversational momentum — and Vector of Thought (V) — semantic coherence and directional development of ideas. INSTAN = E × V The paper proposes operational proxies for both variables, introduces the gradient ∇INSTAN(t) for tracking dialogue dynamics, and outlines the IADM (INSTAN Adaptive Dialogue Monitor) architecture for real-time adaptive response. Concept originated by Oleksandr Prokashev, Kyiv, Ukraine. Term INSTAN first registered in Ukrainian Wiktionary, February 22, 2026.