The Significance of D.I. Dubrovsky’s Information Theory of Consciousness for Designing Artificial Intelligence Architectures

Abstract

The article explores the challenges of applying philosophical and neuroscientific theories of consciousness to artificial intelligence (AI) architectures, particularly large language models (LLMs). It argues that most of these theories were developed to explain biological consciousness and therefore cannot be seamlessly mapped onto artificial systems without risking oversimplified or unfounded analogies. This presents a conceptual challenge in the study of modern AI: while these systems demonstrate a capacity for complex “thinking,” there remains no compelling basis to attribute consciousness to them. To address this gap, the study adopts David Dubrovsky’s information theory of consciousness as its methodological foundation. Unlike classical philosophical approaches that postulate special entities to account for conscious perceptual experience, Dubrovsky’s framework offers a functional-informational account of subjective reality. Drawing on the principles of code dependence and substrate independence (the invariance of information relative to its physical carrier), the theory frames consciousness as the product of self-determination within a multilevel informational structure – an “ego-system.” This perspective provides a viable pathway for modeling consciousness on non-biological substrates. To operationalize these theoretical premises, the paper proposes two experimental architectures. The first aims to equip LLMs with an intrinsic or “self-context” – an autonomous, dynamic informational field that serves as a functional analogue to inner experience. Through interaction with the external environment and other agents, the model employs reinforcement learning to optimize this field, effectively emulating the ability to process “information about information.” The second architecture introduces a two-tiered cognitive system that separates continuous (sub-symbolic) computation from a discrete linguistic interface. By enabling controlled modifications, these designs allow researchers to formulate testable hypotheses and empirically investigate the mechanisms behind phenomena typically associated with consciousness, disentangling them from the machine’s baseline cognitive functions. The proposed approach does not purport to resolve the hard problem of consciousness. Rather, it highlights the substantial heuristic potential of Dubrovsky’s theory, whose principles of structural and functional organization of informational systems show a notable convergence with current trajectories in the development of advanced AI architectures.