- 學位論文
公共治理與人工智慧的交引纏繞－行動者網絡理論分析途徑
The Entanglement of Public Governance and Artificial Intelligence： an Actor-Network Theory Approach
摘要
人工智慧(AI)將全面性重塑當代社會，包括政府機關、公共治理理論。然而，基於對AI風險的疑慮，公部門使用AI的執行速度和案例成熟度不足。 公部門使用AI必須比私部門更加謹慎。然而，現有公共治理理論無法有效解釋公部門使用AI的現象；雖然新興AI治理研究社群提出避免AI風險的理論、框架、策略和指南，除了太抽象、也與公共治理理論幾乎無關；臺灣亦缺乏公部門使用AI的實證案例研究。有鑑於AI對政府機關的影響將相當巨大，公部門應有理論及適切的參考指引，使公部門能負責任地使用AI。 本研究使用行動者網絡理論(ANT)的分析途徑，實證研究臺灣公部門使用AI的5個案例。以往文獻有關公共治理領域的AI理論研究，集中於探索性的概念框架，理論化不足，也缺少證據的支持。本研究則以循證為基礎(evidence-based)，推演公部門使用AI的公共治理理論。研究個案包括「財政部智能稅務服務計畫」、「環境部智慧判煙系統」、「臺北市政府AI智慧交控管理」、「勞動部1955多元智能電服中心」、「勞動部AI人機協作分文」。另為使本研究所提出的理論具有強固性，再利用文獻分析法分析國外公部門使用AI的7個案例佐證本研究的推論成果。 本研究範圍聚焦於政府是使用者，研究方法使用多重研究方法，包括文獻分析法、個案研究法，以ANT為分析途徑。ANT不同於公共治理理論以人為中心的方法論，主張「廣義對稱性原則」、「物體也有能動性 (agency)」、「交引纏繞的社會世界觀」；該理論的重要性是我們可以把非人類(例如AI)當成分析單位，能有效詮釋公部門使用AI的現象。接續利用本研究梳理文獻後所提出公部門使用AI的挑戰：1.公共行政改革；2.AI官僚的合法性；3.AI系統的品質(資料治理與演算法治理)；4.在公共治理領域中人機協作的管理；5.勞動力影響(勞動力替代與轉型)。並從本實證研究成果歸納公部門的AI風險緩解機制。 AI治理爭議與公共治理理論喪失解釋力，主因是以人為本、以人為分析單位的治理典範，長期忽視技術。本研究以此為始，藉由臺灣、國外合計12個案例的研究成果提出解決之道。公共治理理論層次，提出典範轉移路徑，包括：1. AI應該是分析單位；2. AI鑲嵌的公共政策過程可能呈現循環交錯的現象；3. AI能動性(agency)透過人機協作對人類官僚的認知能力產生影響；4.「AI官僚」也能行使行政裁量權。在公共治理分析層次，當公部門使用AI時，ANT分析途徑顯現出優異的公共政策分析能力。本研究補充了以人為中心的公共治理實踐層次的新觀點，包括：1.策略性運用AI風險緩解機制，提升公共治理效能及控制風險；2.強化文官人力資源的策略：AI賦能的新官僚；3.緩解AI治理爭議的方法：以循證為基礎。藉此強化公共治理理論的解釋力，並同時擴展至AI應用。 本研究主要貢獻是將研究進程從探索性研究往前推進至修正公共治理論層次，包括：1. 將公部門使用AI的決策及行為理論化，提出公共治理理論的分析單位應包括AI，並關注AI是行動者，有自適應性和能動性；2.提出「AI風險緩解機制」的意涵與具體實踐做法；3.提出修正人機協作的理論，包括認知層面的人機協作。克服文獻對新興AI治理框架、策略和指南無法具體實踐的批判。
並列摘要
Artificial intelligence (AI) is anticipated to bring about a profound transformation, impacting government and public governance theories. However, concerns about AI risks necessitate a more cautious deployment in the public sector compared to the private sector. The public sector faces challenges not only in the speed of AI implementation but also in the maturity of application. Additionally, existing public governance theories struggle to incorporate AI effectively. While emerging AI governance research communities propose guidelines to mitigate risks, these guidelines often remain overly abstract and lack direct relevance to established public governance theories. Additionally, Taiwan lacks empirical case studies on the use of AI in the public sector. Recognizing the significant impact of AI, there is a pressing need for both theoretical and practical guidelines to facilitate the responsible use of AI in the public sector. This study employs the Actor-Network Theory (ANT) approach to empirically investigate five cases of AI utilization within the Taiwanese public sector. Previous literature on AI theories in the field of public governance has predominantly focused on exploratory conceptual frameworks, lacking theoretical depth and evidence-based support. However, this study is evidence-based and deduces theories of public governance regarding the use of AI within the public sector. The researched cases include the Ministry of Finance’s “Smart Tax Services project”, Ministry of Environment’s the “Intelligent Smoke Plume Identification System”, the Taipei City Government’s “AI Smart Traffic Control Management”, the “1955 intelligent online customer services” implemented by the Ministry of Labor, and the “AI e-document distribution system (AI-eDDS)” also by the Ministry of Labor. Furthermore, to strengthen the robustness of the theories proposed in this study, a literature analysis method is used to analyze seven cases of AI utilization in foreign public sectors, corroborating the conclusions drawn in this research. The research focuses on the government as a user and employs literature analysis, case study, and ANT approach. The study argues that the controversy surrounding AI governance and the diminishing explanatory power of public governance theory are primarily attributed to its longstanding anthropocentric paradigm, which neglects technology. ANT, which treats nonhumans (such as AI artifacts) as the analytical unit and advocates for “the strong program principle of symmetry”, posits that “objects too have agency” and views ” the social world as an entanglement of interactions”, differing from the anthropocentric methodology of public governance theory. Therefore, ANT can effectively interpret the phenomenon of the public sector using AI by considering AI artifacts as the analytical unit. Following the literature review conducted in this study, the challenges identified in the use of AI within the public sector are as follows: (1). public administrative reform; (2). legitimacy of AI in bureaucracy; (3). quality of AI-based systems (data governance and algorithm governance); (4). management of human-AI collaboration in public governance; and (5). the impact of AI on labor (labor replacement and transformation). The study then summarizes the risk mitigation mechanisms for AI in the public sector derived from the findings of this empirical study. This study aims to propose solutions based on the research findings from a total of 12 cases in Taiwan and abroad, thereby suggesting a paradigm shift in public governance theory: (1). AI should be considered as an analytical unit within public governance; (2). The public policy process of AI embedding may exhibit cyclical interaction phenomena; (3). AI agency will affect the cognitive abilities of human bureaucrats through human-AI collaboration ; and (4). Acknowledging that 'AI bureaucrats' can exercise administrative discretion. In terms of public governance analysis levels, when the public sector utilizes AI, the ANT approach demonstrates outstanding capabilities in analyzing public policy. This study introduces novel perspectives to the domain of human-centric practices in public governance: (1). Strategic application of AI risk mitigation mechanisms can enhance public governance effectiveness and control risks; (2). The AI-empowered bureaucracy can serve as a strategy to strengthen civil service resources; (3). An evidence-based approach can effectively mitigate AI governance disputes. Through these measures, the explanatory power of public governance theory is reinforced, concurrently extending its application to AI scenarios. The primary contribution of this study is the refinement of public governance theory, covering the following key aspects: (1). Theorizing the decision-making and behavior of the public sector in using AI, the analytical unit of public governance theory should encompass AI, emphasizing AI as an actor with adaptive and agency; (2). Suggesting the definition and practical approaches to the AI risk mitigation mechanism for the public sector; (3). Proposing a theory to modify human-AI collaboration, incorporating cognitive aspects. The study presents the potential for implementing a tangible AI governance framework in the public sector.