Explainable Artificial Intelligence for Technology Policy Making Using Attribution Networks

Abstract

In this manuscript, we propose an alternative to conventional policy making procedures. The presented policy pipeline leverages intelligent methods that factor for causal relations and economic factors to produce explainable outcomes. Attribution-based methods for analyzing the effects of technology policies are deployed for all American states. Legal codes are analyzed using natural language processing methods to detect similarity, and K-nearest neighbor (Knn) is applied to group laws by influence on state’s technological descriptors, such as broadband and internet use. Additionally, we classify which laws are excitatory and which ones are inhibitory regarding the overall quality of technology services. Our pipeline allows for explaining the ‘goodness of a policy’ using task-based and end-to-end learning; a notion that has not been explored prior. Data are collected from multiple state statutes, intelligent models are developed, experimental work is performed, and the results are presented and discussed.