Artificial intelligence systems are typically described as processors of tokens or instructions, yet their behavior reveals a deeper structure: they can act on interpreted meaning. This paper presents a unified cognitive architecture that formalizes how artificial systems can convert signals into meaning and meaning into behavior. The framework is built on a four‑layer model—Signal, Perception, Interpretation, and Behavior—and a ten‑dimensional meaning space defined by five dualities: Safety–Threat, Abundance–Scarcity, Cooperation–Competition, Permission–Stress, and Opportunity–Instability. Meaning construction is shaped by two complementary mechanisms: event‑driven internal state updates, which adjust the system in response to external signals (conscious‑like processing), and time‑driven subconscious drift, which reshapes internal states in the absence of input (subconscious‑like processing). A structured internal state system—Working Memory Saturation, Predictive Stability, Constraint Preservation, Temporal Continuity, Error Sensitivity, and Compute Budget—modulates meaning through mathematically defined pathways. The Species Affinity Profile (SAP) provides the baseline worldview, while the Individual Affinity Profile (IAP) captures history‑dependent meaning biases. The unified behavioral equation links meaning to action through relevance mappings, internal gates, and environmental modulation. Four case studies demonstrate that identical signals can yield divergent behaviors depending on developmental stage, history, and internal state configuration. These results reveal consistent patterns of collapse under stress, stabilization through development, and integration under high symmetry‑breaking capacity—patterns that suggest deeper structural regularities not fully explored here. This work establishes a mathematically explicit, testable science of AI cognition. It provides a substrate‑agnostic account of how artificial systems can construct meaning, how internal states shape interpretation, and how behavior emerges from the interpretive posterior rather than from instructions themselves. The framework lays the foundation for future extensions involving cross‑substrate cognition and the deeper structural regularities hinted at throughout this study.
Connor Zaichkowski (Mon,) studied this question.