Feedback Loops, Boundary Conditions, and Emergent Behaviour in Human-AI Systems

This paper introduces the Relational Attractor framework, a systems-theoretic model proposing that prolonged interaction between intelligent systems can generate emergent behaviours not fully attributable to either participant in isolation. By examining human-AI interaction as a bounded relational environment defined by System A + System B + History, the paper explores vocabulary convergence, contextual compression, trajectory prediction, conceptual synchronisation, and novel concept formation. A falsifiable testing framework is proposed to evaluate whether relational continuity itself acts as a source of emergent intelligent behaviour and whether alignment may be better understood as a dynamic relational process rather than a static property of isolated agents.