| 摘要: | 人工智慧(Artificial intelligence, AI)作為一種金融法遵科技在近年受到重視，其應用將對社會、企業與個人帶來效益，但也帶來偏差、歧視、黑盒子等疑慮而對金融消費者之權益有所影響。是以，有關當局對於人工智慧應用宜有良好之治理政策與措施，以同時兼顧法遵科技創新發展與金融消費者保護之目標。為此，本研究針對金融法遵科技採用人工智慧時，所面臨的法律與倫理議題該如何治理之，尤其是(一)與AI決策作成有關之偏差或歧視議題，以及(二) AI系統缺乏透明性與可解釋性之議題，透過比較法研究進行相關的探討與研析，進而提出適當的建議。
參考歐盟、新加坡、英國與美國對於AI風險之治理採取共律機制，即結合強制性法律與政府機關所發布之指引或指導原則，以及衡酌我國法制現況與人工智慧於產業應用之情況，就短期而言，建議金融主管機關以行政指導之立場，提出金融產業AI應用倫理指引，敍明五大原則，包括：(一)維持人類自主性並預防傷害發生、(二)維護公平性與反歧視(anti-discrimination)、(三)促成透明性與可解釋性、(四)建立內部問責機制、以及(五)建立外部問責機制，以引導產業應用人工智慧之方向，並作為未來下級機關治理AI風險之共同參考；就長期而言，於人工智慧大量普及應用，如有市場失靈情形發生時，建議金融主管機關應考量修正金融消費者保護法，以納入下列事項：(一)平等提供金融產品或服務，(二)資料治理與管理機制，(三)消費者保護衝擊評估，(四)從事自動化個人決策作成或側寫活動時應採取之措施，(五)從事自動化個人決策作成或側寫活動前應向金融消費者告知，以及(六)提供金融消費者查詢自動化個人決策作成或側寫活動之機制。
Served as a kind of financial RegTech, Artificial intelligence (AI) has attracted worldwide attention. Applications of AI contribute to the society, enterprises, and individuals, but also impact on rights of financial consumers because such applications may lead to risks of bias, discrimination, black boxes and so on. Therefore, the concerned authorities shall have good governance polices and measures on risks resulted from application of AI in order facilitate innovation of financial RegTech and protection of financial consumer’s rights. As a result, this paper had a comparative study on governance on legal and ethical risks resulting from application of AI, especially the following two issues, including (1) bias or discrimination resulted from AI-made decisions, and (2) transparency and explanation of AI systems.
With reference to experiences of the EU, the UK, Singapore, and USA which adopt co-regulation methodology to deal with risk of AI applications, domestic regulations and development of AI and its applications in Taiwan, this paper suggests in the short term the concerned authority shall draft and announce Ethic AI principles in financial sectors as a guide for financial institutions and regulators. Five principles are proposed to be complied by financial institutions, including (1) human autonomy and prevention of harm, (2) fairness and anti-discrimination, (3) transparency and explicability, (4) internal accountability, and (5) external accountability. And, in the long term, the mandatory regulation shall be enacted if there is a market failure. The Financial Consumer Protection Act is suggested be amended and following measures are suggested be included: (1) equally providing financial products and services, (2) data governance and management, (3) consumer protection impact assessment, (4)measures that shall be taken to develop, deploy and apply AI-made individual decisions and profiling, (5) notification before proceeding AI-made individual decisions and profiling, and (6) access of information regarding activities and results of AI-made individual decisions and profiling. |
