Unveiling intrinsic interactions of science and technology in artificial intelligence using a network portrait divergence approach

Abstract

Artificial Intelligence (AI) is experiencing unprecedented innovation and transformation, potentially attributed to intimate interactions between science and technology (S&T) within the field. To identify S&T linkages and detect intrinsic interactions within AI, this paper introduces a network portrait divergence approach, where S&T knowledge networks are prototyped as two-dimensional network portraits based on graph-invariant probability distributions, and comparing them by coupling network portrait divergence with knowledge content. Specifically, S&T knowledge of AI is first extracted and unified through KeyBERT and word-alignment algorithms. Subsequently, temporal S&T knowledge networks are constructed and visualized as two network portraits: node portraits and edge-weight portraits. Network portrait divergence, an information-theoretic, graph-like measure for comparing networks, is applied to calculate varying S&T portrait divergences. Finally, internal knowledge flows within S&T and dynamic interactions between them are unearthed based on multiscale backbone analysis. Empirical experiments on both synthetic networks (random graph ensembles) and real-world AI datasets underscore the feasibility and reliability of the network portrait divergence approach.