A Systemic Network Approach to Identifying Core Failure Drivers and Propagation Characteristics in AI‐Driven Higher Education Transformation Processes: Integrating Improved K‐Shell and SEIR Models

ABSTRACT AI is rapidly reshaping higher education systems, introducing both transformative opportunities and complex systemic risks that challenge traditional governance approaches. This study investigates the mechanisms underlying documented AI‐related failure incidents in higher education transformation from a systemic and network‐oriented perspective. Higher education is conceptualized as a complex socio‐technical system, where failures emerge from cross‐layer interactions among model, human, organizational and institutional dimensions rather than isolated factors. To capture both structural interdependencies and dynamic propagation processes, a structure–dynamics coupling framework is developed by integrating an improved K‐shell decomposition method with a susceptible–exposed–infected–recovered (SEIR) model. Based on 63 publicly documented AI‐related failure incidents collected from established AI incident databases, a directed weighted network of failure drivers is constructed. The improved K‐shell method incorporates node heterogeneity and edge weights to identify structurally critical nodes, while the SEIR model simulates the temporal evolution of risk propagation through heterogeneous transmission and recovery mechanisms. The results reveal a pronounced core–periphery structure, where a small number of nodes dominate systemic risk diffusion. Model‐related and human‐related factors are identified as high‐impact ‘super‐spreaders’, whereas organizational and institutional factors primarily function as structural enablers shaping propagation pathways. The importance–propagability matrix further uncovers nonlinear threshold effects, indicating that small changes in key nodes can trigger large‐scale systemic risks. From a system‐of‐systems perspective, the findings demonstrate that AI‐driven failures are emergent outcomes of recursive cross‐layer feedback loops. Based on this, the study proposes model‐informed governance strategies, including targeted intervention on high‐risk nodes, enhancement of AI literacy, strengthening of organizational buffering mechanisms and threshold‐based early warning systems. This research advances a systemic analytical framework integrating complex network analysis and diffusion modelling, providing both theoretical insights and practical guidance for improving the resilience and governance of AI‐enabled higher education systems.