identify institutional drift mechanisms

To address performance degradation and increased failure risk in industrial systems caused by concept drift in machine learning, this paper proposes the first causality-based drift explanation framework. Unlike conventional correlation-driven approaches, our method integrates causal modeling with model comparison techniques to identify causal features—rather than superficial statistical changes—that drive drift from time-series data. Its key contribution lies in elevating drift attribution to an interventionally meaningful causal level, thereby substantially improving interpretability, actionability, and domain-specific guidance. Evaluated across multiple industrial manufacturing and critical infrastructure use cases, the framework successfully pinpoints true causal sources of drift, enabling precise anomaly attribution and timely operational response. It thus provides both theoretical foundations and practical pathways for effective drift mitigation.

Linear regulatory frameworks systematically fail in higher education student retention and AI governance, as they presuppose stable causal relationships, predictable agent behavior, and well-defined system boundaries—thereby neglecting the defining features of Complex Adaptive Systems (CAS). Method: We establish, for the first time, structural isomorphism between educational interventions and AI governance, adapting the empirically validated CAPIRE framework to AI regulation. This yields five transferable principles underpinning the Complex Systems AI Governance (CSAIG) methodology, integrating longitudinal educational data analysis, causal inference, structural mapping, prototype clustering, and simulation-based policy design. Contribution: We identify the systemic failure mechanisms of linear regulation in CAS contexts and propose the first complexity-aware AI governance architecture. CSAIG markedly enhances intervention foresight and system resilience, offering a theoretically grounded, empirically informed alternative to reductionist regulatory paradigms.

In contexts characterized by coordination failures, institutional inertia, and path dependence, conventional marginal incentives often fail to disrupt inefficient status quo equilibria. This study proposes an intervention paradigm centered on restructuring the feasible action space—specifically by removing or substituting pivotal action options—thereby fundamentally altering the underlying game structure rather than relying on price mechanisms or merely expanding choice sets. Drawing on a game-theoretic framework that incorporates state-dependent equilibrium selection, and supported by both theoretical proofs and cross-domain case studies (including climate transition, platform regulation, and financial reform), the research demonstrates that such structural interventions can effectively overcome status quo inertia. The findings indicate these approaches substantially outperform traditional policy instruments in real-world settings, offering a novel pathway to dismantle institutionalized inefficient equilibria.

This study investigates how to reshape the internal political-economic structure of a target state—without directly intervening in its policy-making—so as to catalyze endogenous forces that organically generate policy pressures aligned with the preferences of the influencing actor. To this end, the work proposes an innovative framework based on switched dynamical systems, modeling political-economic evolution as a regime-dependent dynamic process. The state transition kernel is treated as a designable variable, and a structured switching vector decomposition mechanism governs transitions between “permissive” and “adversarial” regimes. Integrating combinatorial optimization, regime-contingent equilibrium analysis, and language model–assisted generation, the project derives comparative static relationships between mechanism credibility and identifiability, validates the approach through simulations in a reduced switching space, and demonstrates via small-scale blind experiments the efficacy of language models in generating viable mechanism prototypes.

This paper addresses structural risks arising from the deep integration of AI into socio-technical systems—emergent threats (e.g., eroded trust, power asymmetries, decisional authority degradation) that transcend technical failures and malicious misuse, stemming instead from systemic coupling and feedback-driven evolution. Methodologically, it first systematically classifies three root causes: pre-existing societal structural vulnerabilities, inherent AI system design flaws, and their interaction-induced vicious feedback loops; it then develops a dynamic risk analysis framework integrating scenario mapping, system dynamics simulation, and exploratory foresight. The study identifies cross-level structural vulnerability points and proposes policy pathways to strengthen institutional resilience and adaptive governance. Its contributions include a novel, theoretically grounded yet operationally viable paradigm for global AI governance—one that advances both conceptual rigor and practical applicability in addressing AI’s systemic societal impacts.

This work addresses the significant performance degradation of reinforcement learning (RL) agents under distribution shifts arising from mismatches or dynamics between training and deployment environments, a challenge exacerbated by the lack of a systematic understanding of their causal origins. By modeling agent–environment interaction through the lens of partially observable Markov decision processes (POMDPs), the study decomposes the RL framework into causal components—states, observations, policies, rewards, and transitions—and, incorporating temporal boundaries of shift occurrence, offers the first unified characterization of distribution shifts grounded in causal mechanisms. It distinguishes between internal (agent-driven) and external (environment-driven) sources and introduces a novel taxonomy encompassing explicit, implicit, and hybrid shifts. This framework establishes a structured classification and evaluation system for distribution shifts, enabling systematic analysis and targeted improvements of RL robustness.

Causal discovery is often hindered by violations of the independent and identically distributed (i.i.d.) assumption or the presence of unobserved variables, leading to confounding and selection bias that are difficult to accurately identify. This work proposes a novel approach that analyzes dependency patterns in the transfer of causal mechanisms across multiple environments to uncover the type of structural bias and the variables it affects. The key innovation lies in establishing a testable criterion based on mutual information, which for the first time enables principled differentiation between confounding and selection bias. Building on this criterion, the authors develop the StruBI algorithm for efficient inference. Experimental results demonstrate that the proposed method significantly outperforms state-of-the-art approaches on both synthetic and real-world datasets, accurately identifying the source of bias and its impacted variables.

This study addresses a critical limitation of conventional Cross-Impact Balance (CIB) analysis, which yields only static consistent scenarios and cannot quantify dynamic structural aspects such as transition efforts, key leverage points, timing of adjustments, or responses to external shocks. To overcome this, the authors introduce linear response theory into the CIB framework, exploiting the structural isomorphism between the CIB drift matrix and the Leontief input-output matrix. This enables the derivation of four analytical constructs—Type I cross-impact multipliers, perturbation budgets, impulse response functions, and unit impulse shock profiles—each admitting closed-form solutions that characterize indirect effects, transition resistance, dynamic adjustment pathways, and network sensitivity in socio-technical systems. Applied to an energy transition case, the approach successfully computes all dynamic indicators across five structural equilibria, offering a transferable quantitative toolkit for assessing system resilience, designing transition pathways, and informing policy interventions.

This work addresses a fundamental limitation in existing adaptive methods, which treat environmental non-stationarity—particularly drift—as mere noise or distributional shift, thereby overlooking the progressive loss of organizational coherence between system and environment over time. To overcome this, the paper introduces the principle of Egregious Drift Regulation (EDR), reframing drift as a regulatory signal of coherence mismatch. EDR enables long-term coherent adaptation by dynamically adjusting the system’s internal structure to maintain, reorganize, or transition its operational mechanisms. Departing from conventional error-minimization objectives, this approach shifts the adaptive goal toward coherence regulation, integrating adaptive control with embodied cognition theory. It realizes a mechanism-centered “emergent machine” architecture that unifies state mechanisms, attractor dynamics, coherence metrics, reconfiguration dynamics, and cross-mechanism memory. The resulting framework offers a principled solution for intelligent systems operating in persistently non-stationary environments, substantially enhancing their long-term functional coherence.

This study uncovers a paradox wherein intensified AI regulation weakens organizations’ substantive control over their AI systems, proposing the “governance inversion” hypothesis. Drawing on institutional theory, organizational governance, and AI accountability literature, it develops a conceptual model that identifies four interrelated mechanisms—fragmentation of authority, expansion of symbolic governance, externalization of control, and paralysis of authority—through which regulatory intensification fosters formalistic compliance at the expense of operational control. By introducing the novel concept of “governance inversion,” this work extends institutional decoupling theory and challenges the prevailing assumption that more regulation inherently yields stronger control. It offers critical theoretical insights for rethinking the design and implementation of AI governance frameworks in practice.