Emergent Subjectivity in Artificial Agents (ESAA): Structure-Sensitive Kernel with Spectral Criticality and Reproducible Dynamics

We address the fundamental question of when artificial agents exhibit subjectivity-like behavior by developing a minimal dynamical kernel that couples three macroscopic order parameters: phase synchronization (λ), semantic alignment (λ_sem), and structural persistence (χ).

Standard normalization techniques in attention-based interaction models are known to flatten the network spectrum and suppress structural variation. To resolve this, we introduce a structure-sensitive formulation that uses unnormalized weight matrices, removes degree normalization from phase dynamics, and computes criticality from the spectral radius of the symmetrized weight matrix ρ_max(S_W). We also derive the mean-field critical threshold for uniform frequency distributions with independent noise: K_c ≈ 4(Δ + D_ind)/(π ρ_max(S_W)).

Comprehensive simulations on a 5×5 parameter grid demonstrate that ρ_max decreases from ~50.0 to ~5.6 as syntactic gravity β and semantic bridging γ increase, with corresponding K_c values increasing from ~0.011 to ~0.097. The framework provides clear parameter dependence and a reproducible foundation for studying emergent subjectivity in artificial systems.

Note: In this work, “subjectivity-like behavior” is operationally defined through observable order parameters and does not imply philosophical claims about consciousness.

Contact: info@metaclan.jp