Attention-Gated Virtual Sensorium: A Bandwidth-Limited Architecture for Consciousness-like Organization in Artificial Embodiment

Abstract

This paper proposes the Attention-Gated Virtual Sensorium (AGVS), a theoretical architecture for artificial consciousness-like organization in simulated or virtual embodiment. The central claim is that artificial consciousness-like states should not be modeled as complete access to world-state information. Instead, they should be modeled as finite-bandwidth, body-mediated, attention-gated integration of sensory and interoceptive information. AGVS introduces a virtual sensorium consisting of vision, audition, touch, olfaction, gustation, and interoception. These sensory channels are not processed equally or exhaustively. They are dynamically allocated according to precision-weighted prediction error, goal relevance, memory cues, nonlinear bodily need, emergency urgency, and processing cost. The theory further introduces dynamic conscious-like bandwidth B_max(t), dynamic softmax temperature tau(t), nonlinear bodily need N_i*(t), metacognitive monitoring, selective episodic memory encoding, and mood-like long-term modulation. These components allow the agent to exhibit functional signatures such as attention capture, fatigue-induced narrowing, pain-like priority shifts, uncertainty about its own perception, selective autobiographical memory, and recursive self-state updating. The paper does not claim to prove phenomenal consciousness. Rather, it defines architectural and computational conditions under which an artificial system may become organized in a consciousness-like way. The main conclusion is that subject-like artificial agents should not be designed as omniscient world-state readers, but as bounded virtual bodies that experience partially, predict imperfectly, remember selectively, and update themselves through limited embodied perception. Interdisciplinary fields: Artificial Intelligence, Artificial Consciousness, Embodied AI, Cognitive Science, Predictive Processing, Active Inference, Computational Neuroscience, Affective Computing, Philosophy of Mind, AI Safety, Human-Computer Interaction