Structure, Meaning, and the Answerability of Automated Systems

Abstract

This paper presents an ongoing PhD research project that reframes transparency in AI systems from a property of models to a participatory relation between a system and the people interacting with it. Rather than asking whether a model is interpretable, the project asks what a system and its deployment context must be like for the demands of those interacting with it --- their questions, their need to act, their accountability and governance requirements --- to be genuinely satisfiable. This alignment between system structure and observer demand is formalized as semantic transparency, using refineable graph structures grounded in logic and inquisitive semantics. The framework has been validated in sociotechnical decision processes through close collaboration with public sector organizations on EU AI Act compliance and recruitment AI, producing formal results and practical tools for accountability and fairness analysis. Ongoing work extends the framework to the causal structure of intentional agency and knowledge formation as mediated through AI systems.