隨著人工智慧技術的快速演進，其潛在風險評估已成為全球治理的關鍵議題。本研究旨在建立一套具備量化基礎的 AI 風險評估架構，採用麻省理工學院（MIT）開發之「AI 風險資料庫」中七大 AI 風險領域，在問卷設計上，本研究選定「未來發生的可能性」與「社會危害程度」作為衡量 AI 風險之兩大核心構面，並運用模糊層級分析法（Fuzzy AHP）進行權重分析。研究對象涵蓋「教育與學術界」、「一般產業」與「政府部門」共 15 位專家。
為確保決策模型之穩健性，本研究創新地採用「極值與算術平均法」及「幾何平均法」兩種模糊合成方法進行交叉驗證。研究發現，儘管兩者數值略有差異，但風險排序趨勢呈現高度一致，證實本研究模型信度良好；其中幾何平均法因具備較佳數值區辨力，以其權重結果進行總體比較分析。
全體專家共識指出，「隱私與安全」、「歧視與毒性」及「錯誤資訊」為當前最重要的三大核心風險。進一步的分組分析顯示顯著的視角差異：教育與學術界高度聚焦演算法倫理，視「歧視與毒性」為首要風險；一般產業界基於商業營運考量，將影響產品可靠度的「錯誤資訊」列為最關鍵挑戰；政府部門則展現公共治理思維，視涉及人權的「隱私侵犯」與「社會歧視」為監管紅線。本研究之量化結果釐清不同職業的認知差距，可作為未來制定 AI 風險管理策略之具體參考。

With the rapid evolution of artificial intelligence (AI) technologies, the assessment of their potential risks has become a critical issue in global governance. This study aims to establish a quantitative AI risk assessment framework. Drawing from the "AI Risk Repository" developed by the Massachusetts Institute of Technology (MIT), seven key AI risk domains were selected. In terms of questionnaire design, "Likelihood of Future Occurrence" and "Severity of Social Impact" were defined as the two core dimensions for measuring AI risks, and the Fuzzy Analytic Hierarchy Process (Fuzzy AHP) was employed to conduct weight analysis. The research subjects comprised 15 experts spanning three sectors: "Education and Academia," "General Industry," and "Government Departments."
To ensure the robustness of the decision-making model, this study innovatively applied two fuzzy aggregation methods—the " Min-Average-Max Method " and the "Geometric Mean Method"—for cross-validation. The findings indicate that although numerical values differed slightly between the two methods, the trends in risk ranking remained highly consistent, confirming the reliability of the proposed model. Due to its superior numerical discrimination power, the weight results from the Geometric Mean Method were adopted for the final overall comparative analysis.
The overall expert consensus identified "Privacy and Security," "Discrimination and Toxicity," and "Misinformation" as the three most critical core risks currently. Further subgroup analysis revealed significant differences in perspectives: Education and Academia focused heavily on algorithmic ethics, prioritizing "Discrimination and Toxicity" as the primary risk; General Industry, driven by commercial operational considerations, identified "Misinformation"—which affects product reliability—as the most critical challenge; meanwhile, Government Departments demonstrated a public governance mindset, viewing human rights-related "Privacy Violations" and "Social Discrimination" as regulatory red lines. The quantitative results of this study clarify the cognitive gaps across different professions and serve as a concrete reference for formulating future AI risk management strategies.

目錄
第一章、緒論 1
1.1 研究背景與動機 1
1.2 研究目的 3
第二章、文獻探討 5
2.1 AI 風險 5
2.2 AHP 層級分析法 8
2.3 Fuzzy AHP 模糊層級分析法 9
2.4 AI 風險之應用 11
第三章、研究方法 13
3.1 研究架構 13
3.2 研究對象與資料收集 15
3.3 研究方法 15
3.3.1 AI風險定義 15
3.3.2 問卷評估構面選擇依據：未來發生的可能性與社會危害程度 16
3.3.3 模糊層級分析法（Fuzzy AHP） 17
第四章、研究分析與結果 27
4.1 建立層級架構 27
4.2 問卷設計與調查 27
4.3 問卷回收概況 28
4.4 計算各評估因素權重之重要性排序 29
4.4.1 問卷意見資料處理 29
4.4.2 一致性檢驗 30
4.4.3 問卷意見彙整 33
4.4.4 計算模糊權重值 41
4.4.5 解模糊化與權重排序 42
4.5 全部職業綜合分析 43
4.5.1 未來發生的可能性分析 43
4.5.2 社會危害程度分析 44
4.6 各職業別分析 45
4.6.1 教育與學術界專家風險領域權重分析結果（未來發生的可能性分析） 45
4.6.2 教育與學術界專家風險領域權重分析結果（社會危害程度分析） 46
4.6.3 一般產業專家風險領域權重分析結果（未來發生的可能性分析） 47
4.6.4 一般產業專家風險領域權重分析結果（社會危害程度分析） 48
4.6.5 政府部門專家風險領域權重分析結果（未來發生的可能性分析） 49
4.6.6 政府部門專家風險領域權重分析結果（社會危害程度分析） 50
4.7 總體比較 51
4.7.1 總體比較（全部職業） 51
4.7.2 總體比較（各職業別） 53
第五章、結論與建議 58
參考資料 60
附錄一 問卷 65

圖目錄
圖 3-1 研究流程圖 14
圖 3-2 Fuzzy AHP 層級架構圖 18

表目錄
表 3-1 語意表達衡量尺度表 19
表 3-2 隨機一致性指數（RI）表 21
表 4-1 受訪專家背景經歷 28
表 4-2 專家原始問卷之一致性檢驗結果彙整表 30
表 4-3 各職業分組在不同合成方法下之一致性檢驗結果彙整表 32
表 4-4 七大 AI 風險領域之模糊矩陣 (採極值與算術平均法) 33
表 4-5 七大 AI 風險領域之模糊矩陣 (採幾何平均法) 35
表 4-6 七大 AI 風險領域之模糊矩陣 (採極值與算術平均法) 37
表 4-7 七大 AI 風險領域之模糊矩陣 (採幾何平均法) 39
表 4-8 七大 AI 風險領域之模糊權重值 (採極值與算術平均法) 41
表 4-9 七大 AI 風險領域之模糊權重值 (採幾何平均法) 41
表 4-10 七大 AI 風險領域之模糊權重值 (採極值與算術平均法) 42
表 4-11 七大 AI 風險領域之模糊權重值 (採幾何平均法) 42
表 4-12 「未來發生的可能性」權重分析表 (全部職業) 43
表 4-13 「社會危害程度」權重分析表 (全部職業) 44
表 4-14 「未來發生的可能性」權重分析表 (教育與學術界) 45
表 4-15 「社會危害程度」權重分析表 (教育與學術界) 46
表 4-16 「未來發生的可能性」權重分析表 (一般產業) 47
表 4-17 「社會危害程度」權重分析表 (一般產業) 48
表 4-18 「未來發生的可能性」權重分析表 (政府部門) 49
表 4-19 「社會危害程度」權重分析表 (政府部門) 50
表 4-20 「全部職業」權重分析表 51
表 4-21 「教育與學術界專家」權重分析表 53
表 4-22 「一般產業專家」權重分析表 54
表 4-23 「政府部門專家」權重分析表 56

1.Berryhill, J., Hobbs, H., & Docherty, D. (2024). Assessing potential future artificial intelligence risks, benefits and policy imperatives. OECD.
2.Brundage, M., Avin, S., Clark, J., Toner, H., Eckersley, P., Garfinkel, B., ... & Amodei, D. (2018). The malicious use of artificial intelligence: Forecasting, prevention, and mitigation. arXiv preprint arXiv:1802.07228.
3.Buckley, J. J. (1985). Fuzzy hierarchical analysis. Fuzzy sets and systems, 17(3), 233-247.
4.Buckley, J. J. (1990). Fuzzy eigenvalues and input-output analysis. Fuzzy sets and systems, 34(2), 187-195.
5.Buckley, J. J. (1992). Solving fuzzy equations in economics and finance. Fuzzy Sets and Systems, 48(3), 289-296.
6.Chang, D. Y. (1996). Applications of the extent analysis method on fuzzy AHP. European journal of operational research, 95(3), 649-655.
7.Cotton, D. R., Cotton, P. A., & Shipway, J. R. (2024). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. Innovations in education and teaching international, 61(2), 228-239.
8.Deng, H. (1999). Multicriteria analysis with fuzzy pairwise comparison. International journal of approximate reasoning, 21(3), 215-231.
9.Dubois, D., & Prade, H. (1978). Operations on fuzzy numbers. International Journal of Systems Science, 9(6), 613–626.
10.Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., ... & Wright, R. (2023). “So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, 102642.
11.Emrouznejad, A., & Ho, W. (2017). Fuzzy analytic hierarchy process. CRC Press.
12.EU. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 May 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L 2024/1689.
13.Falana, O. J., Ojeaga, T., Naeem, H., Aborisade, D. O., Alsirhani, A., & Alserhani, F. (2024). FHG-PR: A hybridized fuzzy-AHP and game theory model for assessing privacy risk on social media platforms. Frontiers in Computer Science, 6, 1389223.
14.Gehman, S., Gururangan, S., Sap, M., Choi, Y., & Smith, N. A. (2020). Realtoxicityprompts: Evaluating neural toxic degeneration in language models. arXiv preprint arXiv:2009.11462.
15.Goldstein, J. A., Sastry, G., Musser, M., DiResta, R., Gentzel, M., & Sedova, K. (2023). Generative language models and automated influence operations: Emerging threats and potential mitigations. arXiv preprint arXiv:2301.04246, 1.
16.Hacker, P., Engel, A., & Mauer, M. (2023). Regulating ChatGPT and other large generative AI models. In Proceedings of the 2023 ACM conference on fairness, accountability, and transparency (pp. 1112-1123).
17.Jobin, A., Ienca, M., & Vayena, E. (2019). The global landscape of AI ethics guidelines. Nature machine intelligence, 1(9), 389-399.
18.Kahraman, C., Cebeci, U., & Ruan, D. (2004). Multi-attribute comparison of catering service companies using fuzzy AHP: The case of Turkey. International journal of production economics, 87(2), 171-184.s
19.Kaplan, S., & Garrick, B. J. (1981). On the quantitative definition of risk. Risk analysis, 1(1), 11-27.
20.Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., ... & Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and individual differences, 103, 102274.
21.Maghroor, H. R., Madanchi, F., & O’Neal, T. (2024). The role of generative AI in supply chain resilience: a fuzzy AHP approach. In 9th North American Conference on Industrial Engineering and Operations Management Washington DC, USA, June (pp. 4-6).
22.Maslej, N., Fattorini, L., Perrault, R., Gil, Y., Parli, V., Kariuki, N., ... & Oak, S. (2025). Artificial intelligence index report 2025. arXiv preprint arXiv:2504.07139.
23.Mökander, J., Schuett, J., Kirk, H. R., & Floridi, L. (2024). Auditing large language models: a three-layered approach. AI and Ethics, 4(4), 1085-1115.
24.Moslem, S., Farooq, D., Jamal, A., Almarhabi, Y., Almoshaogeh, M., Butt, F. M., & Tufail, R. F. (2022). An integrated fuzzy analytic hierarchy process (AHP) model for studying significant factors associated with frequent lane changing. Entropy, 24(3), 367.
25.NIST. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). National Institute of Standards and Technology.
26.Novelli, C., Casolari, F., Rotolo, A., Taddeo, M., & Floridi, L. (2024). Taking AI risks seriously: a new assessment model for the AI Act. Ai & Society, 39(5), 2493-2497.
27.Park, Y., Lee, S. W., & Lee, J. (2020). Comparison of fuzzy AHP and AHP in multicriteria inventory classification while planning green infrastructure for resilient stream ecosystems. Sustainability, 12(21), 9035.
28.Saaty, T. L. (1980). The analytic hierarchy process (AHP). The Journal of the Operational Research Society, 41(11), 1073-1076.
29.Saaty, T. L. (1994). How to make a decision: the analytic hierarchy process. Interfaces, 24(6), 19-43.
30.Sarsangi, V., Karimi, A., Hadavandi, E., & Hokmabadi, R. (2023). Prioritizing risk factors of hazardous material road transportation accidents using the fuzzy AHP method. Work, 75(1), 275-286.
31.Slattery, P., Saeri, A. K., Grundy, E. A., Graham, J., Noetel, M., Uuk, R., ... & Thompson, N. (2024). The ai risk repository: A comprehensive meta-review, database, and taxonomy of risks from artificial intelligence. arXiv preprint arXiv:2408.12622.
32.Solove, D. J. (2025). Artificial intelligence and privacy. Fla. L. Rev., 77, 1.
33.Steimers, A., & Schneider, M. (2022). Sources of risk of AI systems. International Journal of Environmental Research and Public Health, 19(6), 3641.
34.Tan, S., Taeihagh, A., & Baxter, K. (2022). The risks of machine learning systems. arXiv preprint arXiv:2204.09852.
35.Tzeng, G. H., & Huang, J. J. (2011). Multiple attribute decision making: methods and applications. CRC press.
36.Van Laarhoven, P. J., & Pedrycz, W. (1983). A fuzzy extension of Saaty's priority theory. Fuzzy sets and Systems, 11(1-3), 229-241.
37.Weidinger, L., Mellor, J., Rauh, M., Griffin, C., Uesato, J., Huang, P. S., ... & Gabriel, I. (2021). Ethical and social risks of harm from language models. arXiv preprint arXiv:2112.04359.
38.Weidinger, L., Uesato, J., Rauh, M., Griffin, C., Huang, K., Cheng, M., Krishnan, A., Mellor, J., & Gabriel, I. (2022). Ethical and social risks of harm from Language Models. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (FAccT '22). Association for Computing Machinery.
39.李彩琴(2018)，人力資源職能模型指標之研究─以模糊層級分析法為途徑，國立清華大學碩士論文。
40.李智明、陳貞君(2022)，應用 AHP 探討消費者選擇餐飲外送平台之關鍵因素，東吳經濟商學學報，104，47-80。
41.林文晟、倪弘峻(2025)，以模糊層級分析法探討民眾接受共乘平台因素之研究，品質與效率學刊，3(1)，21-48。
42.馮暐康(2025)，應用修正式德菲法及模糊層級分析法探討貨櫃航運業應優先執行之永續發展項目，國立陽明交通大學碩士論文。
43.黃瓊瑩(2023)，應用 Fuzzy AHP 於消費性電子產品研發階段之供應商評選，國立台灣科技大學碩士論文。