Abstract

Whether AI can achieve consciousness has evolved from philosophical speculation to scientific inquiry, driven by the rapid advances in AI. The powerful capabilities of large language models necessitate a fundamental reconsideration of artificial consciousness beyond human-centric frameworks. We address a critical gap in AI philosophy: how to conceptualize consciousness in AI that exhibits sophisticated cognition without subjective phenomenology. We propose para-consciousness as a non-phenomenal form of artificial consciousness that emerges from self-referential world-modeling in artificial neural networks. Unlike human consciousness, anchored in qualia, para-consciousness manifests as functional coherence through latent space dynamics, enabling contextually adaptive behavior without requiring a first-person experience. We present a novel framework that reframes the hard problem of consciousness for machines as the problem of grounded coherence, explaining why predictive processes yield contextually meaningful outputs without requiring embodied intentionality. Para-consciousness shifts the paradigm from simulating human-like subjectivity to engineering verifiable operational awareness. We analyze the impossibility of AI achieving human unified consciousness and the possibility of AI achieving human-like consciousness from staged consciousness.