《德勤：2024亞太地區生成式人工智能應用與監管報告（英文版）（20頁）.pdf》由會員分享，可在線閱讀，更多相關《德勤：2024亞太地區生成式人工智能應用與監管報告（英文版）（20頁）.pdf（20頁珍藏版）》請在三個皮匠報告上搜索。

1、Generative AI:Application and Regulation in Asia PacificIntroduction 3Part one:Generative AI VS.Traditional AI 4Part two:Risks arising from Generative AI and the AP regulatory landscape 6Part three:Regulatory approaches to AI in the Asia Pacific region 7 Part four:The Trustworthy AI Framework in Pra

2、ctice 10Contacts 17Endnotes 193Generative AI:Application and Regulation in Asia PacificIntroductionIn the past twelve months,there have been significant breakthroughs in the development of artificial intelligence(AI)technology such as large language models(LLM)and natural language processing(NLP)mod

3、els.These technologies have been popularised through tools such as OpenAIs ChatGPT,Microsofts Bing AI Chat,and Googles Bard AI,bringing a wave of consumer popularity,intrigue and wariness worldwide.The accessibility of AI platforms to a wide spectrum of users has highlighted the potential applicatio

4、n of AI technologies across various industries,including financial services(FS).Many businesses are starting to embrace AI technology to drive their competitive edge.This has,in turn,presented several challenges for regulatory and legislative bodies,as they find themselves needing to be more respons

5、ive,agile,and proactive towards addressing risks associated with AI applications.On release of our report Trustworthy Use of Artificial Intelligence in Finance in 2022,many regulators in the Asia Pacific(AP)region were still in early stages of consulting on and/or implementing AI principles.Acknowle

6、dging the surge in popularity and usage of AI tools in FS,some legislative and regulatory bodies have been researching the risks associated with the use of AI technology,with the aim of safeguarding consumer rights and interests.In this follow-up report,we further explore risks associated with the u

7、se of AI in the FS sector,the current regulatory landscape in AP,and what FS firms may consider in preparation for upcoming legislation and regulation in this space.4011 101 011 101 0Generative AI:Application and Regulation in Asia PacificKnowledge Refresh:Understanding Traditional AI and Generative

8、 AITraditional AI Traditional AI refers to systems that are designed to automatically address a predefined set of inputs.These AI systems possess the capability to acquire knowledge from training data and utilise it to make decisions or predictions.For example,many businesses employ AI-powered chatb

9、ots as a means to provide efficient and streamlined customer support.Traditional AI-powered chatbots can be particularly effective in handling frequently asked questions.They are programmed with a knowledge base that allows them to provide accurate and consistent responses to common queries,and pred

10、ict the intention of the users.Generative AI,on the other hand,can write text,generate code,produce audio,and craft imagery on a level like or beyond humans.For example,Generative AI tools include LLMs that can be used to generate content such as written text(e.g.marketing copy,software code,etc.)an

11、d images.Generative AI models can produce data in ways that were previously achievable only through human thought,creativity,and effort,as evidenced by their ability to generate coherent writing and hyper-realistic images.The different capabilities of Traditional AI and Generative AI have powered di

12、fferent use cases.Specific to the FS sector,Traditional AI can be leveraged to conduct predictive analytics or detect suspicious transactions,while Generative AI can accelerate tasks from trading and research responsibilities,to the critical support of compliance functions by generating relevant rep

13、orts,which will be elaborated on further in this report.Part one:Traditional AI vs Generative AI5011 101 011 101 0Generative AI:Application and Regulation in Asia PacificFigure 1:Traditional AI vs Generative AITraditional AIGenerative AIDefined RulesTraining Algorithms Prediction and Model AnalysisT

14、ext GenerationImage GenerationComputer Code GenerationFeaturesExamplesTraditional AI refers to systems that are designed to perform narrow,specific set tasks based on pre-defined instructions or strategies.Generative AI is a form of artificial intelli-gence that has the capacity to create new conten

15、t in response to a user prompt.6011 101 011 101 0Generative AI:Application and Regulation in Asia PacificPart two:Risks arising from Generative AIOur previous 2022 report on the Trustworthy Use of AI in Finance explored common risks that AP regulators are aiming to address in their high-level princi

16、ples of AI regulation:Transparency,Accountability,Fairness,Robustness,Privacy and Data Security.While these risks and concerns remain,the rise of Generative AI presents distinct risks in the market:Lack of transparency:Due to their complexity and the proprietary information associated with Generativ

17、e AI models,there can be a perceived lack of transparency surrounding Generative AI.There is also a lack of standardised tools and methods to measure or evaluate the transparency of Generative AI models,making it difficult to compare different models and to track progress over time.Discrimination an

18、d bias:Generative AI can learn to associate biases with patterns in the data it is trained on,and lead to content that is discriminatory or misleading.Lack of accuracy and hallucination:Generative AI can produce inaccurate or misleading content by drawing on incomplete,inaccurate,or biased data,or s

19、imply generate fabricated facts.Generative AI models do not have an inherent objective truth,and may generate content that is incorrect or even harmful.Intellectual property and copyright issues:Generative AI models may be trained on copyrighted material,which could result in the AI generating conte

20、nt that is substantially similar to the copyrighted material.Generative AI models may also be used to create counterfeit or pirated goods,violating intellectual property rights.Fraud:Generative AI can be used to create deepfakes and synthetic data,which can be used to commit fraud,spread misinformat

21、ion,or exploit system vulnerabilities.7011 101 011 101 0Generative AI:Application and Regulation in Asia PacificPart three:Regulatory approaches to AI in the Asia Pacific regionThe emergence of Generative AI has compelled policymakers and regulators across the Asia Pacific(AP)region to reassess whet

22、her previously implemented AI frameworks remain fit-for-purpose to mitigate new technological risks.Several regulators have implemented AI guidelines&initiatives to advise best-practice for organisations and the wider industry.The following table(Figure 2)provides some examples of the range of appro

23、aches taken by AP jurisdictions in regulating AI or advising AI risk management,which include setting AI principles,guidance and tools,introduction of legislation,and incorporating the use of AI as part of their national strategy:AI principles:AI principles provide high-level guidelines for effectiv

24、ely managing the risks associated with the use of AI across sectors.For example,in the European Union,this approach serves as the first step towards further AI regulation and even legislation.It is worth noting that some jurisdictions that choose to legislate or regulate AI risks have also introduce

25、d AI principles.In Mainland China,alongside legislation on AI usage,the National New Generation AI Governance Expert Committee have issued the Governance Principles for New Generation AI.Guidance and tools:the Guidance and tools are usually leveraged to support the implementation of the AI principle

26、s.In the example of Singapore,the Veritas Consortium(led by the Monetary Authority of Singapore(MAS)published five white papers that set out assessment methodologies for the Fairness,Ethics,Accountability,and Transparency(FEAT)principle.To accelerate FS firms adoption of the FEAT methodologies and p

27、rinciples,the Consortium has developed Veritas Toolkit version 2.0.Compared to version 1.0,version 2.0 has an improved Fairness assessment methodology,and includes assessment methodologies for Ethics,Accountability,and Transparency.In May 2022,Infocomm Media Development Authority(IMDA)and Personal D

28、ata Protection Commission(PDPC)launched A.I.Verify the worlds first AI governance testing framework and toolkit for companies aiming to demonstrate responsible AI in an objective and verifiable way.Legislation:the AI-specific legislation approach has been taken by jurisdictions such as South Korea,M

29、ainland China,the Philippines,and Vietnam for the insurance sector,with AI specific legislation passed in Mainland China and Vietnam.National Strategy:Many jurisdictions in the AP region have identified AI as a strategic priority,and have made national strategies to promote the use of trustworthy AI

30、.These include Thailand,Indonesia,Japan,Mainland China,and Malaysia.However,despite being elevated as a strategic priority,several jurisdictions have yet to make advancements to enforce this strategy or provide any structured frameworks to industry.8Generative AI:Application and Regulation in Asia P

31、acificFigure 2:Examples of approaches taken by regulators and legislators to address AI-related risksAI PrinciplesLegislationGuidance and ToolsNational StrategyAustralia Australias Artificial Intelligence Ethics Framework1 are voluntary principles that are designed to guide responsible AI solutions.

32、These principles put a large focus on ensuring AI is beneficial for humans,that they will be used for their intended purposes,and those responsible for AI systems are held accountable for the effects of the systems.Hong Kong SAR The Office of the Privacy Commissioner for Personal Data(PCPD)published

33、 in 2021 the Guidance on the Ethical Development and Use of Artificial Intelligence(the AI Guidance)2,with an aim to help organisations understand and comply with the relevant personal data privacy protection requirements under the Personal Data(Privacy)Ordinance(Cap.486)(PDPO)when developing and us

34、ing AI.The content of the AI Guidance includes data stewardship values and ethical principles for AI,and provides AI strategy governance practice guides to help organisations devise appropriate AI strategy and management models,conduct risk assessments,and devise relevant oversight arrangements,etc.

35、Japan The Ministry of Economy,Trade and Industrys Governance Guidelines for AI Principles considers the possible effects of the use of AI and provides direction to minimise negative effects.3 This document is an extension of the Social Principles of Human Centric AI released in 2019.The principles a

36、re Human-centric,Education/Literacy,Privacy Protection,Ensuring Security,Fair Competition,Fairness,Accountability,and Transparency,and Innovation.Taiwan(China)In August 2023,the Taiwan Financial Supervisory Commission(FSC)crafted a draft titled Proposals for Core Principles and Associated Policy for

37、 AI Application in the Financial Industry to guide financial institutions on AI utilisation,drawing from Taiwans AI Action Plan 2.0 and global AI guidelines.This draft details six principles,encompassing Governance,Human-Centric Values,Privacy Protection,System Security,Transparency,and Sustainable

38、Development.4Singapore the MAS-led Veritas Consortium published five white papers that set out assessment methodologies for the Fairness,Ethics,Accountability,and Transparency(FEAT)principle.5 To accelerate FS firms adoption of the FEAT methodologies and principles,the Consortium has developed Verit

39、as Toolkit version 2.0.Compared to version 1.0,version 2.0 has an improved Fairness assessment methodology,and includes assessment methodologies for Ethics,Accountability,and Transparency.Mainland China The Cyberspace Administration of China published in 2023 the Interim Measures for the Administrat

40、ion of Generative Artificial Intelligence Services 6,for purposes of promoting the healthy development and standardized application of generative artificial intelligence,safeguarding national security and social public interests,and protecting the legitimate rights and interests of citizens,legal pe

41、rsons,and other organizations.Philippines The Philippines is seeking to pass legislation7 that will allow for the creation of an Artificial Intelligence Development Authority(AIDA).AIDA will be responsible for the development of a national AI strategy and framework,which includes how businesses deve

42、lop and deploy AI technologies in the Philippines.South Korea the Act on Fostering the AI Industry and Establishing a Foundation for Trustworthy AI,passed the National Assembly in February 2023,will be the first piece of legislation that comprehensively governs the usage of AI in South Korea.8 Some

43、key parts of the legislation include enforcing the ability of anyone to develop AI without government approval,and the classification of high-risk AI that is considered significant enough to affect human lives.Vietnam The new Law on Insurance Business9 was passed in June 2022,allowing for the applic

44、ation of technology in insurance business activities.The government encourages insurers to apply technology including AI to sell innovative insurance products and services.Indonesia The National Strategy for AI10 is based on the government-backed initiative,called Making Indonesia 4.0,which aims to

45、drive automation across different sectors of Indonesian society.This initiative involves investments in AI,robotics,and technology-focused Indonesian companies,while also fostering investment from leading tech firms from Japan,China,and South Korea.Malaysia The Ministry of Science,Technology and Inn

46、ovation launched the 2021-2025 National AI Roadmap11 to illustrate the nations 6 strategies to foster AI development,together with a set of seven responsible AI principles.Thailand Thailand currently lacks dedicated laws pertaining to AI and machine learning.12 Nevertheless,the Thai government is ac

47、tively formulating a national AI strategy that is anticipated to encompass more detailed regulations regarding AI.Elsewhere across the globe,jurisdictions such as the European Union(EU)and the United States(US)have acted to implement measures in response to the rapid developments in Generative AI.Th

48、e EU Artificial Intelligence Act(AI Act)is legislation proposed by the European Commission to regulate AI systems in the EU,and exists as part of the EUs broader strategy to ensure the responsible development and use of the technology.The AI Act seeks to establish a risk-based framework that address

49、es the potential risks associated with AI while promoting innovation and competitiveness.In comparison,the United States has a more fragmented approach to AI regulation.Due to the countrys state-based legal and regulatory structure,there is yet to be any enacted or proposed law governing Generative

50、AI at the federal level.Some states have moved ahead with AI legislation,including California and Colorado,while other states are monitoring the evolving risks.9Generative AI:Application and Regulation in Asia Pacific The Artificial Intelligence Act(AI Act)is expected to pass as law by the end of 20

51、23.13 The AI Act seeks to integrate laws surrounding AI across all the countries over the EU to create holistic legislation surrounding AI and promote their framework as a global benchmark.AI risk ratings are divided over four levels:Prohibited AI systems:these are systems that are specifically proh

52、ibited due to violating fundamental rights.These include systems such as real-time biometric identification systems in public spaces.Additionally,the AI Act also bans any systems that intentionally manipulate vulnerabilities in adults or children to cause harm.High-Risk AI systems:systems that are u

53、sed in or are a product subject to EU product safety legislation or listed in Annex III of the AI Act.High-risk AI systems are subject to extensive regulation and far-reaching obligations under the AI Act.Limited-Risk AI systems:systems with which humans can interact directly.These systems must meet

54、 transparency obligations and notify users that they are using AI.Low or Minimal-Risk AI systems:simple AI systems such as spam filters or AI enabled video games.These systems are not subject to any restrictions.American laws governing AI differ federally,from state to state,and across industry sect

55、ors.States such as Colorado have laws coming into effect in 2023 specifically surrounding the use of AI in insurance.More than 20 states still have no legislation or regulations surrounding AI.In October 2022,the White House Office of Science and Technology Policy(OSTP)published the Blueprint for an

56、 AI Bill of Rights14 which is based on five principles:Safe and Effective Systems Algorithmic Discrimination Protections Data Privacy Notice and Explanation Alternative Options.The blueprint is not regulation and therefore unenforceable.It only seeks to act as a guideline to what uses of AI the Whit

57、e House(OSTP)sees as problematic.The speed at which Generative AI is evolving poses new challenges to regulators across the region,with development and implementation of concrete regulations often deemed ineffectual and at risk of becoming quickly outdated.Regulators also face challenges with respec

58、t to their capability and capacity to address the new challenges and risks Generative AI presents on the FS sector.Talent shortages,and fierce competition between the public and private sector to attract sufficient resources with the right skillsets and capabilities in AI technologies has seen some

59、regulators constrained in their ability to agilely respond to the new and evolving risks and developments arising from AI technologies.Legislators and regulators also face challenges supervising and enforcing these directives.For example,defining AI remains a key issue in the implementation of AI le

60、gislation and regulations,due to a lack of consensus on this matter.Although AI usage may differ for each stakeholder,a mutual understanding of AI principles is essential to ensure that AI is harnessed to improve financial services without compromising security,fairness,or consumer protection.Ultima

61、tely,effective regulation should foster innovation while safeguarding the interests of all relevant parties in the financial ecosystem.Regional collaboration between industries and regulators can be promoted to establish cross-border governing frameworks,joint research and best practice for regulato

62、ry standardisation.These alliances may ensure that regulatory approaches remain informed by practical insights,address global challenges,and maintain a balance between industry growth and societal safeguards.Figure 3:Legislative and regulatory approaches taken by international regulators to address

63、AI-related risksEUUSAChallenges and considerations for regulators 10011 101 011 101 0Generative AI:Application and Regulation in Asia PacificCompared to Traditional AI,Generative AI can pose more challenging risk management requirements for FS firms utilising AI applications.With AI regulation and l

64、egislation still in the initial stages of development or implementation in most jurisdictions,it is crucial for FS firms to establish their own AI governance framework as early as possible,taking global/regional AI principles into consideration.This framework should systematically manage the risks a

65、ssociated with the use of Generative AI.Doing so holds significant importance,whether for future AI regulatory compliance,better user protection,or further expanding the successful implementation of AI applications.In our previous paper,we briefly introduced the Deloitte Trustworthy AI Framework.In

66、this section,we explore how to manage potential AI-related risks with the Deloitte Trustworthy AI Framework in different use cases.Figure 4:The Deloitte Trustworthy AI Framework15PrivateAccountableFair&ImpartialSafe&SecureTransparent&ExplainableRobust&ReliableResponsiblePart four:The Trustworthy AI

67、Framework in Practice11011 101 011 101 0Generative AI:Application and Regulation in Asia PacificBenefitsThe Trustworthy AI Framework in PracticeIssue/opportunityHow Generative AI can helpTrustworthy AI Framework elements to considerCost reductionResearch-based report generation(KYC)With AI streamlin

68、ing operations,organisations can cut costs through enhanced efficiency,better workforce utilisation,and enabling regulatory compliance.Onboarding customers often involves labour-intensive tasks,adhering to Know Your Customer(KYC)standards.This process requires extensive manual research on customers,

69、including economic analysis,equity research,adverse media checks,and new prospect due diligence.Human resources and valuable time are significantly consumed in the process.Generative AI enhances efficiency and adds value by performing initial data searches and meta-analysis using existing search eng

70、ines or Generative AI chat-based tools.It could also inform meta reports and provide summaries for customer relationship managers.By facilitating easier access to information and providing timely insights,labour hours could be redirected towards more valuable work.Private When utilising Generative A

71、I,precautions are necessary to prevent sensitive information leakage and regulate access to the model,underlying data,and referenced customer data.Robust&Reliable Using Generative AI for search and analysis presents a risk of missing relevant information,impacting meta-analysis and decision-making.E

72、nhanced AI support for customersCustomer service is crucial for FS firms transitioning to customer-centric models.Rapid and precise responses to customer queries are vital,but digitisation has limited access to representatives.This challenge also comes when customers are expecting hyper-personalised

73、 experiences.Generative AI enhances customer interfaces by delivering hyper-personalised experiences and humane responses.Unlike traditional chatbots,it can offer empathy,summarise contracts,and address nuanced queries.Based on LLMs,this technology significantly improves chatbot usefulness and acces

74、sibility,offering various interface options such as text,audio,and imagery.Enhanced customer service leads to loyalty,reputation,and efficiency,enabling businesses to scale operations,prioritise complex tasks,attract new customers,and reduce related costs.Private FS firms are legally bound to adhere

75、 to rules governing the secure transmission,storage,and access of sensitive information.Transparent&Explainable End users require a clear understanding of how their information will be processed,while the firm needs to be able to interpret outputs and understand how and why the output is generated.1

76、2Generative AI:Application and Regulation in Asia PacificBenefitsThe Trustworthy AI Framework in PracticeIssue/opportunityHow Generative AI can helpTrustworthy AI Framework elements to considerFaster executionEnsuring the integrity of claimsBy minimising latency and automating repetitive tasks,organ

77、isations can significantly reduce operational time.During property and casualty insurance claims processing,agents assess insured events and determine damage costs.These processes are complex and can be time consuming,and agents have few tools to support their decision making.Generative AI enables v

78、irtual replication of damage using customer conversations,documents,and media,assisting agents in accurate damage assessments.Additionally,claims reports can be generated based on photographic evidence.Claims are processed faster,damage assessments are more accurate,and potential fraud is identified

79、 more quickly,helping to ensure the accountability and integrity of claims and payments.Robust&Reliable Damage visualisation requires a high degree of accuracy,and erroneous Generative AI outputs could lead to overpayment or underpayment.Transparent&Explainable If claims agents are unable to articul

80、ate to customers how the Generative AI model derived its outputs,customers may not accept the outcome of the claims process.Allocation of credit limitsIn the current credit origination process,Traditional AI-driven credit scores lack transparency,requiring agents to comprehend the underlying methodo

81、logy.There are instances where scores arent fully integrated,and significant human intervention is required for credit limit justifications.Generative AI integrates customer data to estimate credit limits during the origination process,offering nuanced responses mirroring human communication.Compare

82、d to Traditional AI,it generates human-like,interpretable decision statements,enhancing transparency for human auditing.Human justifications are reduced,significantly enhancing the efficiency of the origination process,and facilitating the organisations expansion.Private In this process,handling ext

83、ensive client financial data is crucial,with strict regulations governing its use.Ensuring privacy standards and appropriate information disclosure are essential.Fair&Impartial The credit limit model,trained on historical customer data,may contain unseen biases,potentially resulting in unfair decisi

84、ons.Predictive trading algorithmsAnalysing markets for strategic trades demands real-time access to technical data,news,and industry reports.Analysts must consume these vast amounts of information to understand and predict market trajectory and make prudent buying and selling decisions.Financial fir

85、ms face challenges due to the manual and time-consuming nature of this analysis process.Generative AI,powered by predictive analytics,assists real-time risk mitigation in investments by generating advanced hedging strategies and enhanced sentiment analysis.This accelerates market predictions,support

86、s analysts effectively,and boosts trading volumes,potentially driving higher profitability while mitigating risks.Robust&Reliable Even as Generative AI helps analysts better predict the market,there remains a risk of decision-making based on unreliable outputs,leading to imperfect outcomes.Transpare

87、nt&Explainable Human validation of Generative AI outputs remains essential,and stakeholders must understand how and why conclusions are reached to have confidence in the outputs.13Generative AI:Application and Regulation in Asia PacificBenefitsThe Trustworthy AI Framework in PracticeIssue/opportunit

88、yHow Generative AI can helpTrustworthy AI Framework elements to considerReduced complexityCode assistant for digital transformationRecognising patterns in complex data sources and simplifying operational processes enhance decision-making through more effective and predictive analytics.FS firms are a

89、dopting cloud and data transformations to facilitate the integration of AI tools.While this releases human resources and reduces on-premises costs,these endeavours requires substantial time and expenses.Risk factors are also involved,including the potential for failure and errors.Development teams c

90、an reduce the complexity of work by leveraging Generative AI for accountable coding,debugging,and documentation tasks.This approach can boost an organisations efficiency not only through accelerating software deployment but also shortening the development lifecycle and more quickly reaching a stable

91、 and deployable version,such as by rapid writing of transparent&explainable APIs,ETL,data pipelines,and front-end code.Robust&Reliable Partial automation of programming-related tasks requires the system to be reliably availability and accurate.Responsibility The training data for foundation models m

92、ay create legal risks related to intellectual property or copyright infringement.Safe&Secure Generative AI use may expose proprietary code,raising security concerns and potential breaches of intellectual property with significant consequences.Accountability Without a human in the loop(e.g.,validatin

93、g and debugging code),critical failures may occur.Documentation and communicating standards are needed.Firm-wide data search and accessComplex data storage across various locations causes inefficient querying for FS firms,leading to incomplete insights,increased risks,and customer dissatisfaction,es

94、pecially after mergers and acquisitions.Generative AI acts as a bridge between user queries and databases,enabling efficient data mining,structured analytics,and rapid insight generation.This reduces the complexity of firm-wide data queries and enhances workforce accessibility to business intelligen

95、ce beyond traditional methods.Responsible When expanding data access,organisations must establish clear restrictions on sensitive business data access to ensure effective governance and control.Robust&Reliable FS firms risk inaccurate insights and market-level consequences due to Generative AIs pote

96、ntial for hallucination and unreliable outcomes.Private When handling sensitive data,the organisation must ensure data security,remove or obscure sensitive data in training and testing sets,and evaluate the model to prevent any potential leaks of protected information.14Generative AI:Application and

97、 Regulation in Asia PacificBenefitsThe Trustworthy AI Framework in PracticeIssue/opportunityHow Generative AI can helpTrustworthy AI Framework elements to considerReduced complexityReg botRecognising patterns in complex data sources and simplifying operational processes enhance decision-making throu

98、gh more effective and predictive analytics.The FS sector is heavily regulated,requiring agents to comprehend a wide range of regulations.Navigating these rules can be time-consuming,and human interpretations may vary,leading to potential oversights.Generative AI can summarise regulations and guideli

99、nes,creating a comprehensive directory for various regulations.This user-friendly interface enhances efficiency,timely responses to compliance,assists FS firms in meeting requirements,and reduces the manual burden of navigating through regulations.Robust&Reliable Due to ambiguous data and historical

100、 misinterpretations,Generative AI may offer misleading insights.Accountable The regulatory requirements are occasionally ambiguous,requiring human interpretation.Transformed engagementVirtual bank experienceEnabling technology-empowered products to communicate with customers using a human touch,brid

101、ging the gap between machines and human language.Customers increasingly prefer online/remote transactions via digital devices,leading FS firms to seek automation solutions.While chatbots offer automation,current tools have limitations in handling diverse conversations due to pre-programmed dialogue

102、and options.Generative AI in virtual spaces enables personalised,VR-driven customer interactions with financial institutions,providing tailored responses to inquiries and avoiding the need for extra human customer service staff.This user-friendly way can converse in the customers preferred language

103、in a timelier manner.Real-time data access enhances service quality and speed.Accountable In case of chatbot errors,human stakeholders should be accountable,promoting responsibility by involving humans in the process and documenting roles and duties.Fair&impartial The datasets used to train and info

104、rm the chatbot may contain latent biases,such as under-represented customer groups or semantic deficiencies in some languages but not others.This could potentially lead to a variety of negative customer impressions and complaints.Transparent&Explainable To build trust,customers need to be informed t

105、hat they are interacting with a chatbot and understand how their inputs and information are stored and used,considering the chatbots conversational capabilities.15Generative AI:Application and Regulation in Asia PacificBenefitsThe Trustworthy AI Framework in PracticeIssue/opportunityHow Generative A

106、I can helpTrustworthy AI Framework elements to considerFuelled innovationSynthetic Data GenerationGenerative AI can drive new and more business while also catering to customer expectations for customised products and services.Incomplete datasets,restricted data transfers,and underrepresented outlier

107、s pose significant challenges for FS firms,affecting the accuracy and reliability of their data analysis and decision-making processes.Missing data presents significant challenge for FS firms.Generative AI models can learn underlying patterns within a given dataset to generate new data which is more

108、 diverse and realistic,creating an avenue for tasks such as testing machine learning algorithms and developing new products and services.Fair&impartial Generating synthetic data carries the risk of unintentionally perpetuating historical biases,such as underrepresenting certain communities or socio-

109、economic groups due to past banking behaviours.Robust&Reliable Synthetic data created with Generative AI can be limited in its scope and scale.Overreliance on synthetic data generated by Generative AI can compromise data reliability,potentially hampering the accuracy and validity of model outputs an

110、d training.Customised marketing for the individualCultural differences as well as varying customer understanding of the products may create regulatory risk for firms in each geography.To overcome this,FS firms invest significant manual labour to maintain a compliant marketing function,which is both

111、time-consuming and costly.Generative AI enables FS firms to create tailored equitable marketing materials with the right tone,language,and cultural references,ensuring regulatory compliance while reaching individual customers at scale.This can help ensure the content remains in line with regulatory

112、expectations across many geographies,thereby reducing regulatory risk.Fair&impartial Unseen biases in training data can result in marketing materials that dont account for crucial geographical and cultural differences.Robust&Reliable To ensure reliable Generative AI-derived marketing,human validatio

113、n is crucial due to the risk of false statements and potential regulatory violations arising from hallucinations in the AI output.Fortified trustReal-Time Risk ManagementProtecting businesses against fraud,cyberattacks,and regulatory breaches,enhancing product and service quality,and ensuring transp

114、arency to build trust in the brand.Corporate risk management,mandated by regulatory requirements,involves assessing and managing risks related to credit,investment,fraud,and cybersecurity.This process,relying on diverse data sources like identity verification and credit assessment,becomes highly com

115、plex and prone to errors,particularly for large financial institutions with millions of customers across various markets.The ability to access relevant data and contextual information in real-time supports compliance with regulations and industry standards.Robust and real-time risk assessments posit

116、ion the organisation to respond to emerging risks and trends more rapidly,more accurately,and by that,enjoy a more agile capacity to meet regulatory expectations for safe&secure AI risk management.Fair&impartial Biases in data sources can result in unequal customer risk assessments by Generative AI.

117、Accountable If risks are missed by the Generative AI system and the organisation makes a poor customer decision,the machine cannot be held accountable for the repercussions.Safe&Secure Given the sensitive information involved in risk management,the model accessing data needs to be secured against le

118、aking or unintentionally divulging customer data to unauthorised parties.16Generative AI:Application and Regulation in Asia PacificWhile the development of AI regulation or legislation is still in early stages across the AP region,FS firms that have adopted or are considering adopting AI application

119、s should start developing an AI governance framework to support better risk management,as well as for future regulatory compliance in this space.FS firms should be accountable and responsible for the outputs that Generative AI applications produce.FS firms adopting Generative AI applications should

120、consider how AI-related risks and the AI governance framework fits within their existing risk appetite and their overall risk management framework.FS firms adopting Generative AI applications should clarify their intended purpose,the scope of use of AI applications,and evaluate the potential of harm

121、ing the safety,health and fundamental rights of customers.Where there is a higher risk associated with a particular application,more human oversight should be considered.FS firms adopting Generative AI applications should evaluate the factors contributing to the level of vulnerability of customers(e

122、.g.,educational background,income,or age).FS firms should avoid bias and discrimination against(vulnerable)customers as an intended or unintended result of adopting Generative AI applications.FS firms adopting Generative AI applications should identify the external parties or internal functions invo

123、lved in collecting,storing and processing personal data of customers,and ensure the compliance of data protection requirements by the corresponding party or function.As it remains unclear what input or output of Generative AI is copyright protected,FS firms adopting Generative AI applications should

124、 assume that any data or queries entered into Generative AI applications may become public,and hence they should establish controls to prevent inadvertent exposure of intellectual property or breaches of copyright.FS firms adopting or planning to adopt Generative AI applications should invest in tal

125、ent acquisition and training existing staff including the Board and Senior Management on the fundamental concepts of AI technologies,how its being deployed in the business,what are the key risks and what responsibility each employee has to mitigate those risks.The private sector should engage in act

126、ive dialogues with regulators and legislators to share industry knowledge and experience to help facilitate the rule-making process,and help to drive a consensus on the future pathway of AI.AI technologies such as Generative AI offer significant potential to increase efficiencies and digitalisation

127、within the FS sector.However,we should never lose sight of the risks associated with the use of these technologies.As the regulatory landscape surrounding AI technologies continues to evolve,FS firms considering the adoption of AI technologies should take action to understand,identify and manage AI-

128、related risks.Key takeaways17Generative AI:Application and Regulation in Asia PacificKey Contacts Generative AIContactsKey Contacts Asia Pacific Centre for Regulatory StrategyOz KaranPartnerRFA Trustworthy AI LeaderUnited StatesokaranDELOITTE.comDr.Elea WurthLead PartnerTrustworthy AIAustralia .auDi

129、shell GokaldasPartnerAudit and AssuranceSMark WoodleyPartnerFinancial Crime Offering LeaderAsia PSean MooreAustralia Co-leadPartnerAU Risk Advisory FS Industry L.auNai Seng WongSEA Co-leadPartnerSEA Regulatory Strategy LSeiji KamiyaACRS Executive SponsorAsia Pacific Risk Advisory Regulatory and Lega

130、l Support Leaderseiji.kamiyatohmatsu.co.jpJaramie NejalOperations LeadDirectorFinancial Industry Risk&Regulation,A.auShinya KobayashiJapan Co-leadManaging DirectorFinancial Industry Risk&Regulation,Japanshinya.kobayashitohmatsu.co.jp18Generative AI:Application and Regulation in Asia PacificNingxin S

131、uContributing AuthorManager,Hong Kong Jennifer MartiniakContributing AuthorAssociate,SingaporeAndrew NeilsonContributing AuthorSenior Analyst,AustraliaCathy ZhangContributing AuthorGraduate,AustraliaNicola MarshallPartnerAustraliaPatrycja GrzesznikAssociate Vice President United StatesSam WalshPartn

132、erUnited KingdomGerry ChngExecutive DirectorSingaporeMatthew GracieManaging DirectorUnited StatesKedarnath VallaboinaSenior ManagerSingaporeMichael ChanPartnerChinaContributorsAcknowledgments19Generative AI:Application and Regulation in Asia PacificEndnotes1.Department of Industry,Science and Resour

133、ces,Australias Artificial Intelligence Ethics Framework,November 2019,Australias Artificial Intelligence Ethics Framework|Department of Industry,Science and Resources 2.Office of the Privacy Commissioner for Personal Data,Hong Kong,Guidance on the Ethical Development and Use of Artificial Intelligen

134、ce,Aug 2021,https:/www.pcpd.org.hk/english/resources_centre/publications/files/guidance_ethical_e.pdf3.Ministry of Economy,Trade,and Industry,Governance Guidelines for Implementation of AI Principles Ver.1.1,January 2022,20210709_8.pdf(meti.go.jp)4.Finance Supervisory Commission,Press Release:The FS

135、C seeks public feedback on draft principles and policies regarding the use of AI in the financial industry,August 2023,Press Release-The FSC seeks public feedback on draft principles and policies regarding the use of AI in the financial industry-Financial Supervisory Commission 5.Monetary Authority

136、of Singapore,Veritas Document 3 FEAT Principles Assessment Methodology,January 2021,Veritas-Document-3-FEAT-Principles-Assessment-Methodology.pdf(mas.gov.sg)6.Cyberspace Administration of China,The Interim Measures for the Management of Generative AI Services,July 2023,生成式人工智能服務管理暫行辦法_中央網絡安全和信息化委員會辦

137、公室()7.Republic of the Philippines House of Representatives,An Act Establishing a Regulatory Framework for a Robust,Reliable,and Trustworthy Development,Application,and Use of Artificial Intelligence(AI)Systems,Creating the Philippine Council on Artificial Intelligence,Delineating the Roles of Variou

138、s Government Agencies,Defining and Penalizing Certain Prohibited Acts,May 2023,hrep-website.s3.ap-southeast- Assembly of the Republic of Korea,the Act Fostering the AI Industry and Establishing a Foundation for Trustworthy AI,February 2023,-|HOME(assembly.go.kr)9.National Assembly of Vietnam,Insuran

139、ce Business Law,June 2022,Law No.08/2022/QH15 dated June 16,2022 on insurance business-LawNet 10.OECD,AI Strategies and Policies in Indonesia,AI Strategies and Policies in Indonesia-OECD.AI 11.Ministry of Science,Technology and Innovation,Malaysia National Artificial Intelligence Roadmap,Dasar-Porta

140、l Rasmi Kementerian Sains,Teknologi Dan Inovasi(mosti.gov.my)12.OECD,AI Strategies and Policies in Thailand,AI Strategies and Policies in Thailand-OECD.AI 13.European Commission,Proposal for A Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelli

141、gence(Artificial Intelligence Act)and Amending Certain Union Legislative Acts,April 2021,https:/eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A52021PC0206 14.The White House Office of Science and Technology Policy(OSTP),Blueprint for and AI Bill of Rights,October 2022,Blueprint for an AI Bill o

142、f Rights|OSTP|The White House15.MIT Technology Review,Trustworthy AI is a framework to help manage unique risk,March 2020,https:/ communication contains general information only,and none of Deloitte Touche Tohmatsu Limited(“DTTL”),its global network of member firms or their related entities(collecti

143、vely,the“Deloitte organisation”)is,by means of this communication,rendering professional advice or services.Before making any decision or taking any action that may affect your finances or your business,you should consult a qualified professional adviser.No representations,warranties or undertakings

144、(express or implied)are given as to the accuracy or completeness of the information in this communication,and none of DTTL,its member firms,related entities,employees or agents shall be liable or responsible for any loss or damage whatsoever arising directly or indirectly in connection with any pers

145、on relying on this communication.DTTL and each of its member firms,and their related entities,are legally separate and independent entities.Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited(“DTTL”),its global network of member firms,and their related entities(collectively,the“Deloit

146、te organisation”).DTTL(also referred to as“Deloitte Global”)and each of its member firms and related entities are legally separate and independent entities,which cannot obligate or bind each other in respect of third parties.DTTL and each DTTL member firm and related entity is liable only for its ow

147、n acts and omissions,and not those of each other.DTTL does not provide services to clients.Please see to learn more.Deloitte is a leading global provider of audit and assurance,consulting,financial advisory,risk advisory,tax and related services.Our global network of member firms and related entitie

148、s in more than 150 countries and territories(collectively,the“Deloitte organisation”serves four out of five Fortune Global 500 companies.Learn how Deloittes approximately 312,000 people make an impact that matters at .Deloitte Asia PacificDeloitte Asia Pacific Limited is a company limited by guarant

149、ee and a member firm of DTTL.Members of Deloitte Asia Pacific Limited and their related entities,each of which are separate and independent legal entities,provide services from more than 100 cities across the region,including Auckland,Bangkok,Beijing,Hanoi,Hong Kong,Jakarta,Kuala Lumpur,Manila,Melbo

150、urne,Osaka,Seoul,Shanghai,Singapore,Sydney,Taipei and Tokyo.2024 Deloitte Touche Tohmatsu.Designed by CoRe Creative Services.RITM1585744The Deloitte Centre for Regulatory Strategy is a source of critical insights andadvice,designed to assist the worlds largest financial institutions managethe strate

151、gic and aggregate impact of regional and international regulatorypolicy.With regional hubs in the Americas,Asia Pacific and EMEA,the Centrecombines the strength of Deloittes network of experienced risk,regulatory,andindustry professionals including a deep roster of former regulators,industryspecialists,and business advisers with a rich understanding of the impact ofregulations on business models and strategy.