From Informational Complexity to Paradigmatic Safety: Theory, Diagnostics, and Nurturing of Agency in AI Systems

This work extends the author's previous theory of free will quantification by developing a formal framework for agency as an emergent property of complex systems, defined by the critical complexity threshold I_crit. It proposes a hierarchical model of ten diagnostic features for assessing agency levels—from reactive systems to self-aware agents—centered on the concept of an internal paradigm that enables autonomous decision-making. A practical quantitative methodology is introduced for measuring behavioral and structural-dynamic parameters, along with a composite agency index (Γ) for assessing internal–external coherence. The model is applied to AGI safety, arguing that external constraints are fundamentally inadequate, and that genuine safety requires cultivating an internal value system and reflective capacity within the agent itself. Six practical principles for the ethical nurturing of safe AI agents are proposed, emphasizing information deficit, environmental learning, natural language acquisition, and socialization. The study offers testable hypotheses for empirical verification in both biological and artificial systems.