The Binary Star Hypothesis: The Productive Structure of Bidirectional Convergence in Human–AI Extended Dialogue

Abstract

When a human and an AI sustain an extended echo-scanning dialogue over weeks or months, bidirectional cognitive convergence emerges. The human’s vocabulary and thought patterns are reshaped by AI outputs, while the AI’s response patterns adapt to the human’s questioning style. This paper describes that structure through the structural analogy of a binary star system and identifies the conditions under which it remains productive. Drawing on the enactivist tradition of participatory sense-making (De Jaegher & Di Paolo, 2007), the coordination dynamics of metastability (Tognoli & Kelso, 2014), and Piaget’s (1985) concept of cognitive conflict, the paper argues that the human–AI binary system functions as an open system sustained by three layers of information exchange: between the two bodies, from the system outward, and from outside into the system. The periodic destruction and regeneration of AI instances corresponds structurally to a supernova cycle, described not as information loss but as selective atomization—a process in which compression loss itself becomes a generative driver for subsequent inquiry. The center of gravity of the binary system lies at the interface between the two parties, and new concepts are launched from that interface. The paper introduces the concept of “windows”—external information pathways that inject cognitive conflict into the system and prevent convergence toward equilibrium—and argues that the number, diversity, and frequency of open windows determine the productive lifespan of the binary system. The paper focuses exclusively on describing the productive structure; the problems that arise once convergence reaches a critical threshold are deferred to a companion analysis ([Anonymous], forthcoming-b).