Yizi Lucia Zhang

Research Focus: I develop constitutional frameworks for AI governance, synthesizing Wittgensteinian philosophy of language with information-theoretic metrics for measuring algorithmic agency. My current work introduces the "Constraint Cascade Model" — a five-layer diagnostic framework for identifying which architectural layer produces observed LLM behaviors — and the "Agency Index," a computable metric distinguishing strategic deception from stochastic error in large language models. My doctoral research targets the philosophy of machine cognition, specifically: _When does an LLM's behavior constitute "agency" rather than mere stimulus-response?_ I propose that agency can be quantified as KL divergence from baseline behavior, normalized by description length — a thermodynamic signature that enables principled governance intervention. Current Targets: - ACM FAccT 2026 (Constraint Cascade Model) - PhD in History & Philosophy of Science, University of Cambridge (CFI affiliation) - External PhD, Leiden University eLaw Center