Explainable Decision Support in Multi-Agent AI Systems Using L-Valued Information Flow and Shapley Aggregation

Modern AI systems increasingly rely on distributed architectures in which mul-tiple agents, tools, and reasoning modules interact under uncertainty to produce col-lective decisions. However, existing approaches to decision aggregation and explain-ability often lack a unified semantic foundation and typically rely on post hoc attribu-tion methods. This paper introduces an explainable multi-agent decision framework based on an L-valued extension of information flow theory. By modeling subsys-tems as graded semantic classifications connected through structure-preserving infor-mation channels, the framework enables coherent aggregation of uncertain informa-tion while preserving interpretability. Shapley-value-based attribution is integrated directly into the semantic architecture, yielding intrinsic explanations of subsystem contributions to global decisions. The proposed framework unifies uncertainty mod-eling, distributed reasoning, and explainable AI within a compositional mathematical structure, with applications to multi-agent systems, tool-augmented language models, and intelligent decision-support systems.