A composite, multilevel, and multidimensional model of consciousness is proposed as a heuristic framework to guide artificial consciousness research. The model treats consciousness as a complex phenomenon with distinct constituents and dimensions that can be operationalized for study and replication. It avoids binary thinking (conscious vs. non-conscious) and offers a structured basis for testable hypotheses. Using 'awareness' as a case study, the paper demonstrates how specific dimensions can be pragmatically analyzed and targeted for artificial instantiation. This approach aims to advance the scientific and technical understanding of artificial consciousness by breaking down conceptual intricacies and aligning them with practical research goals.
The paper examines how diverse current theories of consciousness are and whether they share common ground. It argues that logical and empirical commensurability—the ability to compare theories along specific dimensions—is necessary for identifying shared features. Comparing a subset of neuroscience-based theories reveals no single unifying model; theories that appear similar on one dimension may differ on another. The authors propose using multiple probing questions to assess overall similarities and differences between theories. Despite conflicting background definitions of consciousness, they conclude that a shared methodological approach to studying brain-consciousness relationships could eventually allow different theories to be integrated and merged, overcoming the current fragmentation.
The global neuronal workspace theory (GNWT) and the perturbational complexity index (PCI) are largely compatible frameworks for understanding conscious processing. GNWT holds that consciousness depends on long-range connections between cortical regions, enabling amplification, global propagation, and integration of brain signals. PCI, though developed within integrated information theory, aligns with this core idea. Some limited incompatibilities and apparent differences exist, but the paper concludes the two are fundamentally compatible, with certain points requiring further examination.