Proceedings of the Annual Meeting of the Cognitive Science Society

Large language models (LLMs) display linguistic fluency but lack human_like understanding. Philosophical theories of consciousness, rooted in subjective experience, and AI research, which evaluates observable performance, often diverge methodologically. We reconcile these views by treating conscious agency as an emergent property of adaptive dyadic communication, using Tyler's Ten Testable Properties of Consciousness as functional constraints rather than evidence of inner experience. An analysis of current LLM_based dialogue systems shows they can mimic isolated properties yet cannot satisfy all constraints simultaneously, due to weak persistent interaction states, limited interlocutor modeling, and absent temporally structured episodic experience. To move beyond technology_deterministic approaches, we propose a framework in which agency_like properties arise from three adaptive network layers: (i) a User_Model Network encoding goals and intentions, (ii) an Agent_State Network encoding semantic_pragmatic_experiential state, and (iii) a Communicative_Reasoning Layer monitoring coherence and alignment. We instantiate this design as the Large Communication Model (LCM).