Abstract: In an era characterized by intensifying globalization, rapid technological iteration, and geopolitical volatility, traditional organizational decision-making paradigms grounded in the theory of “bounded rationality” are facing unprecedented challenges. Decision-makers are increasingly constrained by inherent cognitive limits when processing real-time, high-dimensional information, leading to satisficing behaviors that are frequently insufficient for today’s high-risk, complex, and feedback-delayed environments. While Artificial Intelligence (AI) has emerged as a transformative force capable of reshaping the epistemological foundations of strategy and risk management, the theoretical mechanisms and boundary conditions governing “human-AI symbiosis” remain underexplored. Existing literature often treats AI as a monolithic variable or categorizes collaboration solely based on automation levels, failing to elucidate the underlying power dynamics—specifically, the allocation of decision rights and the locus of agency—that define the collaborative interface. Consequently, there is a critical need for a systematic framework that deconstructs how AI reconfigures the structural and procedural dimensions of organizational decision-making beyond mere technical functionalism. To address this significant theoretical gap, this study conducts a systematic literature review of 85 core articles published in UTD24 premier management journals between 2020 and 2025. Utilizing bibliometric analysis and systematic coding, the research maps the trajectory of AI application in organizational decision-making, identifying key thematic clusters such as information processing, machine learning, and trust dynamics. The central contribution of this work is the construction of the novel “Interaction-Support-Influence” classification framework, anchored in the power relations between humans and machines. This typology transcends prior functional descriptions to reveal distinct logics of agency and collaboration. The “Interaction Mode” represents a form of shared agency characterized by a recursive loop of output, feedback, and modification; here, decision power is dynamically shared, and AI acts as a reflexive partner that challenges human assumptions and co-creates solutions through bidirectional engagement. The “Support Mode” characterizes scenarios where AI functions as an augmentative tool for information processing or option generation, yet the final fiat remains exclusively with the human agent, focusing on enhancing efficiency and accuracy while retaining human cognitive sovereignty. The “Influence Mode” delineates scenarios where AI operates as an environmental variable or architect, shaping the decision-making context through information filtering or algorithmic choice architecture without explicit participation in the final choice, thereby structurally constraining human agency through latent algorithmic antecedents. The synthesis reveals that the efficacy of these modes is contingent upon a complex interplay of internal and external factors. Internally, individual traits such as algorithm aversion or appreciation, alongside organizational capabilities like absorptive capacity and strategic inertia, dictate adoption success. Externally, macro-institutional factors, including regulatory frameworks and cultural contexts, significantly bound the scope and legitimacy of collaboration. Regarding outcomes, the study distinguishes between performance metrics and behavioral evolution, highlighting a “double-edged sword” effect: while AI integration generally enhances financial efficiency and decision quality, it simultaneously introduces risks such as cognitive atrophy, algorithmic bias, and the erosion of unique human tacit knowledge. Despite these theoretical advances, the review identifies significant lacunae in current scholarship, particularly regarding the lack of integration between AI implementation and core organizational strategy. Existing research remains fixated on a “human-in-the-loop” paradigm that underestimates AI’s potential agency and neglects the profound impact of macro-environmental variables. To advance the field, this paper proposes a comprehensive future research agenda structured around two critical dimensions: the deep integration of AI with organizational strategy and the expansion of novel AI strategic horizons. First, regarding the deep integration of AI and strategy, the study calls for moving beyond static adoption to investigate dynamic human-machine authority allocation, the impact of macro-institutional variables—such as geopolitical shifts and cultural contexts—and the profound reconfiguration of organizational structures and relational networks. Second, regarding the expansion of new AI horizons, the paper emphasizes the urgent need to examine AI’s disruptive role across four emerging frontiers: facilitating scalable innovation and application, redefining power dynamics within platform ecosystem governance, managing systemic risks and compliance, and addressing the paradoxes of AI-driven ESG sustainability. Ultimately, this research extends the theoretical boundaries of sociotechnical systems in management, offering practitioners a diagnostic logic for designing differentiated human-AI mechanisms that align with specific strategic imperatives, thereby facilitating a shift from simple tool adoption to deep structural integration and sustainable competitive advantage.

摘要： 在全球化与数字化加剧的复杂不确定环境下，传统基于“有限理性”的决策模式难以应对实时、高维信息处理需求，人工智能（AI）正深刻重塑组织决策范式。组织需要引入AI提升决策效能，但其作用机制与适用边界尚缺乏系统性理论支撑。基于2020-2025年间发表于UTD24管理学期刊的85篇核心文献，文章系统厘清了AI在组织决策中的应用路径与研究进展，并依据人机权力关系，创新性地构建了“交互—支持—影响”人机决策模式分类框架。该框架突破了以往研究单纯依赖自动化程度划分的局限，揭示了不同模式下决策影响因素与作用结果的系统性差异。目前研究对AI主导模式、宏观情境及关系影响的探讨尚存不足。未来研究应聚焦AI与组织战略的深度整合及AI新视野的拓展，以深入揭示AI如何重构组织决策的底层范式与组织结构基础。此研究结论在理论上扩展了协同决策边界，在实践上为企业设计差异化人机协同机制提供参考，并为后续人机决策提供新的研究思路。