Intelligence & Consciousness in Artificial Intelligence Systems
Zanyar Zohourianshahzadi
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Proposes that intelligence begins with decision-making, formalized as the "Decision-Intelligence Lemma," and offers a five-level hierarchy of functional artificial consciousness. Argues that current large language models and agentic systems already demonstrate components of the first three levels, and that adding information modalities does not by itself create or increase consciousness (the "Modality-Consciousness Independence Lemma"). |
Abstract
Intelligence and consciousness are frequently treated as binary properties associated with biological complexity, human-like cognition, or an unspecified computational threshold. This paper proposes an alternative framework in which intelligence begins with information-dependent selection among possible actions, states, or outputs. We formalize this foundation through the Decision-Intelligence Lemma, which states that any program or entity capable of making a decision possesses at least a minimal form of intelligence. Using panpsychism as a working metaphysical hypothesis, we distinguish intelligence, functional consciousness, and phenomenal consciousness without claiming that observable cognitive behavior conclusively establishes subjective experience. We then introduce a five-level hierarchy of functional artificial consciousness comprising affective-relational modeling, deliberative reasoning, adaptive autonomy, metacognitive supervision, and societal coordination among supervised cognitive hierarchies. Contemporary large language models and agentic systems already demonstrate important components of the first three levels, while multi-agent orchestration provides architectural foundations relevant to the higher levels. We further propose the Modality-Consciousness Independence Lemma, according to which increasing the number of information modalities available to a system does not, by itself, create consciousness or increase its functional-consciousness level. Additional modalities may instead improve the richness, precision, and reliability of the system's environmental representation. This distinction separates functional organization from perceptual breadth and permits unimodal and multimodal systems to be evaluated within the same framework. Finally, we identify a future research direction in which vision-language, sound-language, and other cross-modal models operate as specialized perceptual agents coordinated by a language-based supervisory agent. The resulting framework provides an operational vocabulary for investigating artificial consciousness through observable capacities while preserving epistemic caution concerning phenomenal experience.