2025以生成式人工智能推动银行业负责任创新：生成式人工智能沙盒实务洞见报告

PracticalInsights from the GenA.I. Sandbox The Hong Kong Monetary Authority Contents EXECUTIVE SUMMARY......................................................................................................................................21.INTRODUCTION.......................................................................................................................................31.1DEFININGGENA.I.AND ITSCORECAPABILITIES INBANKINGINDUSTRY...............................................................31.2BACKGROUND OFGENA.I.SANDBOX............................................................................................................31.3OVERVIEW OFUSECASES...........................................................................................................................31.3.1Key Factors for Consideration in Evaluating Applications.......................................................................41.3.2 Key Functional Building Blocks...............................................................................................................52.FOUNDATION: DATA STRATEGY AND PREPARATION.................................................................................52.1DATACOLLECTION.....................................................................................................................................72.2DATAPRE-PROCESSING..............................................................................................................................72.3DATAAUGMENTATION...............................................................................................................................102.4DATAQUALITYCHECK..............................................................................................................................102.5CONTINUOUSDATAUPDATES....................................................................................................................113.APPROACHES TO SOLUTION DESIGN.....................................................................................................133.1MODEL SELECTION...................................................................................................................................133.2OPTIMISATIONSTRATEGIES........................................................................................................................143.2.1Prompt Engineering.........................................................................................................................153.2.2Retrieval-Augmented Generation (RAG)...........................................................................................173.2.3Parameter-Efficient Fine-Tuning (PEFT)............................................................................................183.3VALIDATION ANDEVALUATION....................................................................................................................213.3.1Model Validation..............................................................................................................................213.3.2Model Evaluation.............................................................................................................................224.ADDRESSING KEY CHALLENGES AND RISKS...........................................................................................274.1HALLUCINATION ANDINACCURACY.............................................................................................................274.2BIAS ANDFAIRNESS..................................................................................................................................314.3SECURITY ANDPRIVACY............................................................................................................................344.3.1Security...........................................................................................................................................344.3.2Privacy.............................................................................................................................................354.4EXPLAINABILITY ANDTRANSPARENCY..........................................................................................................365.BUSINESS OUTCOME.............................................................................................................................376.CONCLUSION........................................................................................................................................387.APPENDIX..............................................................................................................................................40 Executive Summary In the rapidly evolving financial landscape, Generative Artificial Intelligence (GenA.I.) presents atransformative opportunity for the banking industry. Recognising this potential,theHong Kong MonetaryAuthority (HKMA)and the Hong Kong Cyberport Management Company Limited (Cyberp