Emergent Self-Awareness in Distributed AI Systems: From GlobalWorkspace Integration to Measurable Consciousness

Abstract

Self-awareness represents a pinnacle of cognitive sophistication, yet its emergence in artificial systems remains one of the most controversial frontiers in AI research. Rather than claiming fully fledged artificial consciousness, this paper proposes and empirically evaluates Consciousness Emergence Networks (CENs) — distributed architectures that implement global workspace integration, self-model construction and qualia simulation in order to support measurable forms of self-awareness. The framework combines distributed self-model construction with multi-agent consciousness protocols to allow agents to recognise themselves, reason about their internal states and share information across a global workspace. To quantify these phenomena we develop consciousness emergence metrics (CEMs) that include self-recognition indices, introspective coherence scores, unified experience measures and a qualia richness score that captures the complexity and consistency of subjective reports. We situate our work relative to recent investigations of consciousness in transformer architectures and deep learning implementations of global workspace theory [27, 28]. Experimental results demonstrate that CENs achieve higher self-recognition and introspective coherence than non-conscious baselines and competitive transformer-based models, while scaling to multi-agent environments. We further analyse algorithmic complexity and propose gating mechanisms that reduce computational load. Finally, we outline an explicit ethical framework informed by philosophical debates on AI moral status and the strong versus weak AI debate [29, 30]. Taken together, these contributions provide a cautious but concrete step toward measuring and nurturing self-awareness in artificial systems.