《埃森哲：2024從合規到自信：企業AI成熟度報告-擁抱新思維 提升負責任AI成熟度（英文版）（33頁）.pdf》由會員分享，可在線閱讀，更多相關《埃森哲：2024從合規到自信：企業AI成熟度報告-擁抱新思維 提升負責任AI成熟度（英文版）（33頁）.pdf（33頁珍藏版）》請在三個皮匠報告上搜索。

1、From compliance to confidenceEmbracing a new mindset to advance responsible AI maturityPage 03Executive summary:rethinking responsible AIPage 12Becoming reinvention ready:the milestones of responsible AI maturityPage 05 Risky business:the three challenges of the current risk landscapePage 15Responsi

2、bility reality check:how ready are companies for responsible AI?Page 10 From compliance to value:companies are acknowledging the impact of responsible AIPage 22Ready,set,grow:five priorities for responsible AIContentsResponsible AI:From compliance to confidence2Executive summary:rethinking responsib

3、le AIAs generative AI becomes pervasive in business and society,so too do the risks that come with using it.Consider the chatbot that gave incorrect advice to a customer,creating liability for the company that launched the chatbot.Or the employee who accidentally exposed proprietary company data aft

4、er using ChatGPT.Or the algorithm that wrongly flagged thousands of individuals for fraud.In this new environment,the responsible development and use of data and AI becomes a critical enabler for organizations to both minimize the risks that come with using these technologies and to unlock its many

5、benefitsfrom helping individuals perform tasks better and faster to allowing companies to reinvent themselves to create more value and gain a competitive edge through innovation.This is,of course,easier said than done,for the challenges of scaling responsible AI across an organization are daunting.A

6、I-related risks continue to accumulate as generative AI creates and accelerates both data and AI risks.New AI-focused laws and regulations are germinating everywhere.And AI value chains1 are growing in complexity,especially as more companies become both developers and buyers of AI models.Responsible

7、 AI 101What is responsible AI?Responsible AI implies taking intentional actions to design,deploy and use AI to create value and build trust by protecting against the potential risks of AI.What is mature responsible AI?Having fully operationalized responsible AI efforts as a platform to take a more s

8、ystemic,future-orientated approach that unlocks the true value of AI.The North Star for responsible AI maturity is to become a pioneer.No companies are yet at this stage.How do you build a responsible AI program?Our research and work advising clients has shown that all companies can benefit from foc

9、using on these five priorities to improve their maturity and begin to reap the benefits of AI.01Establish AI governance and principles02Conduct AI risk assessments03Systemic enablement for responsible AI testing04Ongoing monitoring and compliance 05Workforce impact,sustainability,privacy,securityRes

10、ponsible AI:From compliance to confidence3Todays era of generative AI creates new requirements when it comes to responsible AI.To be a leader in responsible AI,companies must pursue an anticipatory mindset,commit to continuous improvement and extend their focus beyond their organization to their ent

11、ire value chain and wider AI ecosystem.We call this new level of maturity“becoming a responsible AI pioneer”,and no companies are yet at this stage.To better understand companies attitudes toward AI-related risks,as well as their approach to responsible AI,we collaborated with Stanford University to

12、 Our survey respondents estimate that when a company becomes a pioneer in responsible AI,its AI-related revenue will increase by18%on average.survey C-suite executives across 1,000 companies,spanning 19 industries and 22 countries.2 We assessed companies maturity in responsible AI by developing a fo

13、ur-stage frameworkthe higher the stage,the greater the progress.We then applied that framework to analyze the organizational and operational maturity of the 1,000 companies we surveyed.To their credit,most business leaders recognize the importance of responsible AI in unlocking business value.Respon

14、sible data is an integral part of the responsible AI journeyIts important to note that when we talk about responsible AI,the responsible use of data is a critical part of the equation.The importance of data in the age of generative AI cannot be underestimated.Most large organizations are still grapp

15、ling with longstanding issues around data quality,availability and governance.In a separate global survey by Accenture of 2,000 business leaders in 2024,48%said their organizations lacked enough high-quality data to operationalize their generative AI initiatives.3 Companies with high data readiness

16、have the right data,with the right quality and quantity.They have a scaled data management and governance practice in place which allows use of data seamlessly across their processes,ensuring responsible adoption and enabling the economic value of data.Click here to learn more about key actions comp

17、anies can take to improve their data readiness in the era of generative AI.They estimate that when a company becomes a pioneer in responsible AI,its AI-related revenue will increase by 18%on average.However,our research found that a great majority of companies we surveyed are not as prepared for res

18、ponsible AI as they would like.Years of client experience show us that companies can take action to systemically operationalize responsible AI.In this report,we will explore the actions that will provide the organizational infrastructure to become a responsible AI pioneer and unlock the true value o

19、f AI.Responsible AI:From compliance to confidence4010203An increasing range and frequency of riskA continuously evolving regulatory landscapeAn expanding scope of risk management across the value chainAI-related risks are evolving quickly,especially as generative AI spreads far and wide.In the last

20、12 months alone,we have seen a new wave of risks emerge(such as those connected to hallucinations,intellectual property(IP),cybersecurity and environmental impact)while more established risks like data privacy,reliability and transparency take on renewed prominence.This creates an increasingly compl

21、ex risk landscape.We found that the risk landscape will continue to expand and evolve across three main areas.Risky business:the three challenges of the current risk landscapeResponsible AI:From compliance to confidence501 An increasing range and frequency of riskGenerative AIs impact on the evolvin

22、g AI risk landscape is also visible in many of the risks that worry companies most.These touch on everything from transparency and the challenge of“black box”models(cited by 44%of respondents),to the environment and the energy demands of data centers(a concern for 30%of respondents),to accountabilit

23、y and concerns about infringing on copyright and other intellectual property(cited by 24%of respondents).To learn more about the risk dimensions we assessed,click here to read the paper we produced in collaboration with Stanford University.As the number and type of AI risks have increased sharply in

24、 recent years,AI-driven incidents(bias,deepfakes,hallucinations,privacy breaches etc.)have become much more common,increasing 32.3%in 2023,according to the AI Incident Database,a website that monitors such occurrences4.This trend should raise red flags for companies.The complexity of AI risk continu

25、es to expand in terms of range and frequency,accelerated by generative AI advances.As a result,responsible AI mitigation strategies must also evolve and change,so that companies can continue to adopt AI at speed.In this dynamic environment,companies can no longer react to risks,they must learn to an

26、ticipate them.The most-cited risk,as Figure 1 shows,involved concerns about privacy and data governance(51%of respondents cited this as a risk for their company).Security(cited by 47%of respondents)and reliability risks,such as output errors,hallucinations and model failure(a concern for 45%of respo

27、ndents)were second and third,respectively.Figure 1:Top AI-related risks for companies Source:Accenture Stanford Executive Survey,N=1,000.See“About the research”for details.Privacy&data governanceSecurity Reliability Transparency Human interaction Client/customer Societal Environmental Diversity&non-

28、discrimination Compliance and lawfulness Brand/reputational Accountability Organizational/business 51%47%45%44%35%34%33%30%29%29%26%24%12%Responsible AI:From compliance to confidence6Its also true that companies that fall behind on responsible AI will expose themselves to a growing risk of non-compl

29、iance,as more governments begin to regulate artificial intelligence.The European Unions AI Act started the engines,with other AI regulation currently under consideration across more than 37 countries.As governments inevitably choose to regulate AI in different ways,compliance will grow increasingly

30、difficult for multinational firms.This reality is already impacting most of the companies we surveyed,with 77%either facing AI regulation already or expecting to be subject to it over the next five years.As governments tend to regulate AI in different ways,complexity and confusion will likely grow f

31、or multinational companies.Whats more,its not simply AI regulation organizations have to contend with.The emergence of new generative AI risks is forcing governments to take action by passing and amending existing laws,02 A continuously evolving regulatory landscape77%of companies we surveyed either

32、 already face AI regulation or expect to be subject to it over the next five years.90%of companies surveyed expect to be subject to AI-adjacent legal obligations such as cybersecurity,AI-related laws,data and consumer protection over the next five years.which only adds to the complexity.In fact,near

33、ly all(90%)of the companies surveyed expect to be subject to AI-adjacent legal obligations such as cybersecurity,AI-related laws,data and consumer protection over the next five years.The fact that such legislation is happening at national and sub-national levels is creating further compliance challe

34、nges for companies.For example,new IP or copyright laws have been proposed or introduced in China,Singapore,Brazil and Saudi Arabia.In 2023,South Korea introduced an AI liability law,which precedes a similar law planned in the EU,while the state of California amended the California Privacy Right Act

35、(CPRA),giving consumers the right to opt out of practices such as the selling and sharing of personal data.Deepfake laws have been proposed or introduced in France,UK,Australia and South Korea,with a number of countries investigating new AI cybersecurity laws.Responsible AI:From compliance to confid

36、ence7Global AI regulatory initiatives(as of publication)Listed initiatives are non-exhaustiveCanada Bill C-27:digital charter implementation act to include artificial intelligence and data act(AIDA part C)2022 Artificial Intelligence and Data Act(AIDA)2023$2.4B Federal investment 2024 Canada AI Safe

37、ty Institute 2024United States White House voluntary commitments from 15 leading tech companies on certain security and transparency requirements 2023 Federal Election Commission-proposed regulations on the use of deceptive AI in campaign ads 2023 National Telecommunications and Information Administ

38、ration(NTIA)RFI on AI accountability 2023 Executive Order on the Safe,Secure and Trustworthy Development and Use of Artificial Intelligence 2023 NIST Draft GenAI Companion to RMF 2024 AI Safety Institute 2024 National AI Research Resource(NAIRR)2024UK Artificial Intelligence Regulation White Paper 2

39、023 Generative Artificial Intelligence in Education 2023 Frontier AI Taskforce 2023 UK AI Safety Institute 2023 UK AI Safety Summit and Bletchley Declaration 2023 US-UK Partnership on Science of AI Safety 2024 UK-Canada Partnership on Science of AI Safety 2024 Regulating AI:the ICOs Strategic Approa

40、ch 2024EU Artificial intelligence liability directive 2022 Data Governance Act 2022 Digital Markets Act 2023 EU AI Act 2024 Digital Services Act 2024China Personal Information Protection Law 2022 Internet Information Service Algorithm 2022 Recommendation Management Regulations 2022 Chinas Deep Synth

41、esis Provisions 2023 Draft Measures for the Management of Generative AI Services 2024France Conseil detat approach on AI governance 2022 France AI Commission Report 2024 France genAI plan 2024Australia Safe and Responsible AI(interim response)(2024)National Framework for the Assurance of Artificial

42、Intelligence in Government(2024)Policy for the responsible use of AI in government(2024)Mandatory guardrails for AI in high-risk settings(proposed)(2024)Saudi Arabia Saudi Data&AI Authority AI Ethics Principles(proposed)2023 Saudi Data&AI Authority-Generative Artificial Intelligence Guidelines for G

43、overnment January 10,2024 Saudi Data&AI Authority Generative Artificial Intelligence Guidelines for the Public January 11,2024Brazil Brazilian ai strategy 2021 AI bill proposal 2021 AI Bill No.2238 2023(proposed)National Digital Government Strategy 2024Singapore AI governance approach+implementation

44、 self-assessment guide 2020 AI governance testing framework minimum viable product(mvp)2021 A.I.Verify 2022 Singapore National AI Strategy 2.0(NAIS 2.0)2023 AI safety labels(proposed)2024 AI funding initiative to power Singapores economic growth(2024)Model AI Governance Framework for Generative AI(2

45、024)Safety Guidelines for Model Developers and App Developers(2024)MAS principles to promote FEAT in the use of AIGermany Fundamentals on security of AI systems 2022 AI Action Plan 2023 Testing and Certification Center for AI-based Robots 2024 Guidelines for the use of AI in administrative and profe

46、ssional tasks 2024India Operationalizing principles for responsible AI 2021 Digital Personal Data Protection Act 2023 Approval of AI tools before public release 2024 India AI report 2023 India AI Mission Framework 2024 Recommendations on Encouraging Innovative Technologies Through Regulatory Sandbox

47、.2024Japan Social principles of human-centric AI 2019 AI utilisation guidelines 2019 AI governance guidelines 2022 AI strategy 2022 AI Strategy Council 2023 Hiroshima Process 2023 Japan AI Safety Institute 2024 Legislation to regulate generative AI(proposed)international framework for regulating and

48、 using generative AI 2024Responsible AI:From compliance to confidence803 An expanding scope of risk management across the value chainIts not just the types of risks and regulations that are changing.As firms begin buying and developing AI models,they also need to prepare for the risks that come from

49、 doing both.A developer,for instance,may be especially worried about being sued when training a model in a way that violates IP laws.A buyer,on the other hand,may be more worried about whether a newly acquired AI model will perform as advertized.When a company becomes both a buyer and developer of m

50、odels,the types of risks that it faces can get exponentially complex.This double-sided risk represents another big shift in the AI landscape.Until recently,few companies both bought and developed AI models.Today,however,almost one-third(28%)of the companies we surveyed take on the role of both buyer

51、 and developer(33%of firms just develop and 38%just buy).Our research also suggests that many companies are insufficiently prepared for the evolving complexity of their AI value chains.For example,just 43%of surveyed firms that are acquiring AI models have robust procurement measures in place,such a

52、s regulatory checks and third-party audits.Organizations must work now to assess whether third-party AI products or services meet the organizations AI standards and are monitored to enable ongoing compliance and risk management.As third-party risks continue to evolve,companies must think beyond thei

53、r own responsible AI strategy.The reality is that companies must do their due diligence and make sure all legal and regulatory responsibilities are agreed on and met by all parties along the value chain.For high-risk AI use cases,it will not be sufficient to rationalize outcomes based on an“unintend

54、ed consequence”.Companies should expect to be held accountable by their customers and regulators in their oversight of high-risk use cases.Only43%of companies acquiring AI models have procurement measures in place,such as regulatory checks and third-party audits.Responsible AI:From compliance to con

55、fidence9From compliance to value:companies are acknowledging the impact of responsible AIIn the past,businesses often made the mistake of treating responsible AI as a mere compliance issue,rather than as an essential contributor to value creation.Fortunately,nearly half of the surveyed firms do not

56、hold this point of view today.For instance,about half(49%)of the companies we surveyed said they view responsible AI as a key contributor to AI-related revenue growth for their firm,while only 29%of companies said that responsible AI is mainly a regulatory and compliance issue(Figure 2).Likewise,43%

57、of surveyed companies said that responsible AI is an important contributor to protecting their brands value,while just 24%of firms view responsible AI as simply a“cost of doing business”.49%of the companies we surveyed said they view responsible AI as a key contributor to AI-related revenue growth f

58、or their firm.Figure 2:C-suite perceptions of responsible AI 49%46%43%43%37%36%29%25%24%23%13%13%36%Grows AI-related revenueIndustrializes AI processesImproves brand reputation and AI trustworthinessEnsures safety and securityGains competitive advantageShapes the core AI strategy Avoids financial lo

59、ss/brand damageHelps meet regulatory complianceDemonstrates social responsibilityIs a cost of doing business Attracts/retains talentSlows down innovationIs not critical to using AISource:Accenture Stanford Executive Survey,N=1,000.See“About the research”for details.Responsible AI:From compliance to

60、confidence10To get a better sense of just how much financial value responsible AI can unlock,we asked executives to provide their own estimates.Their answers suggest that most companies today view responsible AI as hugely important.For instance,our survey respondents predicted that when a company be

61、comes a pioneer in responsible AI,its AI-related revenue will increase by 18%,on average.(AI-related revenue is the total revenue generated by AI-enabled products and services.)The opposite is also true.When responsible AI is not well-established in a company,brand value can be quickly destroyed wit

62、h its misuse.Our survey respondents,for their part,estimated that a single major,AI-related incident would,on average,erase 24%of the value of their firms market capitalization.These sentiments further prove why companies are revving up their investments in responsible AI.Of the companies we surveye

63、d,42%already devote more than 10%of their overall AI budget to responsible AI initiatives;over the next two years,79%of companies plan to hit this robust spending target.A growing number of companies are,in short,prioritizing responsible AI and are spending accordingly.So,how close are they to achie

64、ving their responsible AI goals?Responsible AI:From compliance to confidence11Becoming reinvention ready:the milestones of responsible AI maturityIn todays business landscape,continuous change is the new reality.Being set up for continuous change means you need reinvention-readiness in every functio

65、n and every component of your business.Responsible AI is no different.Reinvention-ready companies have the ability to be agile,they absorb the shifts that are constantly happening in-market and can proactively respond to those shifts at speed,exploiting new opportunities and mitigating unintended co

66、nsequences.Our research has shown that there are four groups of companies that are at a distinct milestone of their responsible AI evolution,where the goal is to be reinvention ready.What do we mean by being reinvention ready when it comes to responsible AI?Interestingly,no company has yet reached t

67、hat milestone,but those who get there first will be responsible AI pioneers.They will have fully operationalized responsible AI efforts as a platform to take a more systemic,future-oriented approach that unlocks the true value of AI.Weve defined four milestones of responsible AI maturity,which ultim

68、ately lead to being reinvention ready.Here,maturity is something that will continue to evolve,not a finite state or destination.What is mature today will likely be different in the months to come.Redefining maturity To measure companies responsible AI maturity,we developed a four-stage framework in

69、collaboration with Stanford University.The higher the stage,the greater the progress.Based on our analysis of the responses of the 1,000 executives we surveyed,we then placed organizations at their respective stage,awarding a score for organizational maturity and a separate score for operational mat

70、urity.Note:companies with no responsible AI initiatives were excluded from our analysis.Responsible AI:From compliance to confidence12The company has some foundational capabilities to develop AI systems,but its responsible AI efforts are ad hoc:Has defined a set of ethical AI principles and guidelin

71、es,including policies and rules for responsible and secure data access and usage Has no established processes for governing data quality,data privacy,data security and AI model risk management Occasionally conducts risk assessment reviews of data and AI projects Has deployed AI project workflows wit

72、hout systemic integration of responsible AI assessments across the data pipeline,model pipeline and AI applicationsSetting responsible AI principlesEstablishing a responsible AI programPutting responsible AI into practiceBecoming a responsible AI pioneerResponsible AI maturity milestonesStage 1Stage

73、 2Stage 4Stage 3Following a responsible AI assessment,the organization has put in place the following steps:Established a responsible AI strategy for the organization with a well-defined operating model and data and AI governance measures for translating vision into action Defined a robust approach

74、and process for AI risk assessment across the data pipeline,model pipeline and AI applications Established processes for creating transparency and auditability of training data,model inputs and outputswith appropriate decision-making Designed a framework for monitoring and controls across the data a

75、nd AI pipeline that can be executed during project workflows Implemented processes that are still at an early stage and without a more systemic enablement with tools and technologyThe company has systematically implemented the following measures across the organization to help meet the relevant regu

76、latory and legal obligations:Operationalized responsible AI strategy with implementation of principles,guidelines and processes through to enablement across the business Implemented risk assessment across the data pipeline,AI model and AI applications to enable traceability and transparency across t

77、he entire model lifecycle and adherence and compliance through regular audits Implemented controls for data sourcing with privacy filtering,anonymization and validation to remove sensitive information and mitigate data bias risks embedded into self-service tooling Enabled systemic AI testing with mo

78、del interpretability tools to ensure explainability,AI model performance testing for bias,accuracy,etc.and help guard that models operate within the required legal,ethical and operational boundaries Established a responsible AI control plane with human controls for continuous monitoring across the d

79、ata and AI value chain to alert and remediate any unintended risks or breaches Rolled out a responsible AI academy for employee training and enablement to drive responsible AI adoptionThe company has fully operationalized responsible AI efforts as a platform to take a more systemic,future-orientated

80、 approach that unlocks the true value of AI:Fully operationalized end-to-end systemic responsible AI effort powered by tech platforms,redesigned processes with the right talent and culture established Adopted an anticipatory approach to their responsible AI efforts,deploying dedicated resources,proc

81、esses,etc.to continuously assess current and future risks Proactively adapts their data and AI risk management and control processes as the external technology and regulatory environment expands and evolves Continuously refines and advances data governance and management practices,employing predicti

82、ve analytics and real-time data monitoring to dynamically understand and manage the impact of data on AI systems Shapes new standards,methods and approaches for safe development and use of AI including data privacy preservation,model explainability and bias measurement,adversarial testing and red te

83、aming Recognized as a leader in shaping responsible AI practices and actively engages with external stakeholdersincluding value chain partners,regulatory bodies and affected communitiesto ensure participation and inclusive feedback and support forward-looking regulatory compliance Proactively engage

84、s with third-party AI vendors to foster improvements,building trusted relationships and new collaborative opportunities to manage third-party AI risks effectivelyResponsible AI:From compliance to confidence13Setting responsible AI principlesEstablishing a responsible AI programPutting responsible AI

85、 into practiceStage 1Stage 2Stage 3An Asian manufacturing company wanted to protect against the risks of unchecked AI use.It established an internal AI policy and checklist to ensure safety,security,fairness,transparency and accountabilitybut needed help getting implementation right,including traini

86、ng employees.This company developed a playbook to support the AI ethics review process that cut in half the time required for field operators to run through the checklist.Now,this company can quickly identify AI ethical risks and integrate AI ethical governance into the core of its businessand carry

87、 out its vision responsibly.A multinational consumer healthcare company wanted to define a clear policy and vision to scale responsible AI across the enterprise and standardize processes and ways of building and deploying AI.The company did not have an inventory of high-risk AI applications and stru

88、ggled with the absence of dedicated responsible AI roles and decision-making accountabilities.They worked with Accenture to conduct a global benchmarking and assessment against the regulatory landscape and draft AI principles and policies,as well as a proposed responsible AI operating model.Risk scr

89、eening was conducted across the companys AI applications.A risk assessment for higher-risk cases ensures their applications align with principles and regulatory requirements.They also worked with third-party legal counsel to provide a framework for legislation monitoring.The company is now confident

90、 in its AI usage,including its risk management and accountability,to the extent that it plans to publish an external position paper on responsible AI.Over the past few years,Accenture has undergone its own efforts to increase responsible AI maturity,building on its existing AI principles.The company

91、 developed a rigorous responsible AI program and put it into practice.The key elements of the robust program are:Leadership oversight:Accenture appointed a Chief Responsible AI Officer to oversee the internal responsible AI compliance program,with the Accenture CEO and General Counsel as sponsors.Th

92、e companys Chief Technology Officer,Chief Responsible AI Officer6,General Counsel,and Data and AI Lead oversee the related steering committee.There is also regular board reporting of progress to an audit committee.Principles and governance:Accentures approach to developing and deploying AI solutions

93、 is founded on a set of principles7 that are applied to its own operations as well as its collaborations with clients,partners and suppliers.The company appointed a cross-functional team(including Legal,Security,CIO,Procurement,HR and responsible AI experts,among others)to design and lead the new re

94、sponsible AI compliance program with the responsible AI principles acting as the North Star and anchor for the design.A governance framework that implements key principles,policies and standards supports cross-use case supervision.Risk assessment and mitigation:Accenture takes a risk-based approach

95、in accordance with the EU AI Act and other key frameworks.This meant spending significant time up-front defining the higher-risk use cases.Over the last year,Accenture screened thousands of AI use cases across the company and completed detailed risk assessment and mitigations.Testing and enablement:

96、The company also spent time on systemic enablementdesigning its AI policies,standards and controls and embedding them into technologies,processes and systems.This required building responsible AI technical capabilities,expertise,tools and techniques and developing benchmark testing tools.Talent:Acce

97、nture is also growing its responsible AI skills and talent,with mandatory ethics and compliance training for those Accenture people who are most directly involved with AI,as well as new ethics and AI training through Accenture Technology Quotient(TQ)courses for its 774,000 people.Responsible AI matu

98、rity milestones:examples*The companies in these examples have achieved major elements of the respective stage.The North Star for responsible AI maturity is to become a pioneer.No companies are yet at this stage.Responsible AI:From compliance to confidence14Responsibility reality check:how ready are

99、companies for responsible AI?Responsible AI:From compliance to confidence15While good progress is being made,companies are likely experiencing a perception gap between their intention and execution,as even those who perceive themselves to be mature have a long way to go.Our findings indicate that co

100、mpanies may still be underestimating the number of risks they are exposed to,the quantity of measures required and the completeness of how they are implemented.Without a robust set of risk mitigation measures,organizations are not just exposed to existing risks,they will also struggle to adapt to ne

101、w regulations,anticipate new emerging risks and scale AI opportunities confidently.Responsible AI:From compliance to confidence16Operational versus organizational maturityTo advance responsible AI maturity,companies must translate organizational maturity into a comprehensive set of mitigation measur

102、es,or operational maturity.We observed that while organizational maturity has continued to grow over the last two years,there is a significant disconnect with operational maturity.Operational maturityOperational maturity is currently a big weakness for most companies.Operational maturity measures th

103、e extent to which a company has adopted responsible AI measures to mitigate specific AI-related risks,including those related to privacy and data governance,diversity and discrimination,reliability,security,human interaction,accountability and environmental impact.A high score on operational maturit

104、y indicates that a company excels at implementing AI responsibly across all risk areas that apply to their business.Alas,only a small minority of companies appear to be implementing responsible AI with success.We found that 6%of the companies we surveyed have reached the practice milestone of operat

105、ional maturity and less than 1%of companies are at the pioneer stage(Figure 3).Responsible generative AIWhen it comes to generative AI specifically,we analyzed firms ability to mitigate risks for both developers(some companies)and users of the technology(nearly all companies).For users of generative

106、 AI,we assessed mitigation measures applied at each stage of the AI lifecycle i.e.provider selection,evaluation,infrastructure,application,end-user support and monitoring,control and observability.For developers,we focused on AI infrastructure,model development and evaluation,application development

107、,end-user measures and post-deployment monitoring,control and observability.We found that operational maturity for generative AI and for AI in general were similarly low(Figure 3)just 13%of companies are at either the practice stage(10%of companies)or the pioneer stage(3%of companies).Responsible AI

108、:From compliance to confidence17Organizational maturityOrganizational maturity reflects the extent and effectiveness of an organizations responsible AI processes and practicesas measured by C-suite sponsorship,governance,data and AI risk identification and management,model development,procurement,mo

109、nitoring and control,cybersecurity and training.A high score on organizational maturity indicates that a company is well-prepared to use AI responsibly.For many of the companies we surveyed,organizational maturity is a relative strength(note:scores reflect companies self-reported capabilities).As Fi

110、gure 3 shows,72%are either at the practice or pioneer stage(63%and 9%respectively).Only 3%,meanwhile,are at the principles stage and 25%are at the program stage.Compared to two years ago,C-suite sponsorship of responsible AI has increased from 50%to 79%,while companies with a fully-operationalized r

111、isk management framework are up from 48%to 69%.This shows that companies are starting to take responsible AI seriously and are adopting a top-down,organization-wide approach,which is a foundational step in reaching maturity milestones.Prioritizing responsible AI initiatives has a direct impact on ma

112、turityOur analysis shows us that investing in responsible AI positively correlates to higher maturity in all three dimensions:operational,generative and organizational responsible AI maturity.Companies that spend more than 10%of their AI budget on responsible AI are two times more likely to be at th

113、e practice and pioneer stages of operational maturity and three times more likely to be at the practice and pioneer stages of responsible generative AI maturity.Responsible AI:From compliance to confidence18Overall responsible AI maturityTo fully understand the current responsible AI maturity landsc

114、ape,we designed a framework that uses a composite of individual scores across organizational maturity,operational maturity and generative AI maturity.Our research reveals that a vast majority(78%)of companies have established a responsible AI program*.A smaller portion,14%,have put responsible AI in

115、to practice,while 8%are just beginning their journey by setting responsible AI principles.Notably,none of the companies have become a pioneer.However,this aggregate view obscures the substantial differences that exist between organizational and operational maturity(Figure 3).For example,although at

116、an organizational and operational level pioneers exist,when you combine scoring,you are left with no pioneers.Organizations need to make sure they prioritize organizational maturity and operational maturity.Organizational maturity is the natural place to start and therefore companies are often more

117、advanced on that front.But they must remember that taking a systemic approach to operational maturity is also critically important(as thats when youre really putting the measures into practice in your organization).Otherwise,a perception gap can emerge which leaves companies exposed and unable to ad

118、apt to the changing risk landscape.Our analysis found that there are currently no significant differences in responsible AI maturity across different industries,suggesting a consistent approach to responsible AI across sectors.However,we observed that companies in Asia,for example in India and Singa

119、pore,are leading in responsible AI maturity,demonstrating more advanced practices compared to their global peers.Among European countries,Germany stands out,with its companies showing higher levels of maturity in responsible AI compared to other nations in the region.In many cases these are countrie

120、s that have established broader AI policy,investment and regulation,which likely helps to drive local adoption.0.8%3%3%6%40%53%10%51%37%9%63%25%PrinciplesPioneerPracticeProgramFigure 3:Percentage distribution of companies across four levels of responsible AI maturity by operational,generative AI and

121、 organizational areasGenerative AI operational maturityOrganizational maturityOperational maturitySource:Accenture Stanford Executive Survey,N=1,000.See“About the research”for details.*Companies with no responsible AI initiatives were excluded from our analysis.Responsible AI:From compliance to conf

122、idence19What good looks like:the mark of a responsible AI pioneerWhen it comes to responsible AI maturity,reaching the pioneer stage should be the new goal.As noted,a small minority(9%)of the companies we surveyed are responsible AI pioneers for organizational maturity;and even fewer(less than 1%)fo

123、r operational maturity.When combining operational and organizational maturity,no company has become a pioneer.From a geographic perspective,Asia leads globally with the highest number of companies at the pioneer stage,driven by strong performances in Singapore,Japan and India.Asia makes up 34%of org

124、anizational pioneers and 37%of operational pioneers.In Europe,Germany stands out as the regional leader,especially when it comes to organizational maturity.This is consistent with what wed expect,given these countries either have or are exploring regulation at a government level.In North America,whe

125、re regulation is emerging ad hoc at the individual state,city or industry level,we see they lag behind Asia and even some emerging regions in operational maturity.From an industry perspective,Communications,Media and Technology(particularly in High Tech and Software&Platforms)is most advanced in ter

126、ms of maturity,with 27%of companies advancing in operational maturity and 20%in organizational maturity.Within the Financial Services sector,Insurance and Banking are the most mature.These findings arent surprising,when you consider that the Financial Services industry is highly regulated,and Softwa

127、re&Platform players by and large see self-regulation as critical to their future.Theyve already had to put processes in place to monitor and mitigate the AI theyre using.Finally,the Retail and Industrial Equipment industries show strong maturity while Consumer Goods&Services lags.9%of companies have

128、 reached the pioneer milestone for organizational maturity,and this drops to less than 1%for operational maturity.When combining organizational and operational maturity,no company is a pioneer yet.Responsible AI:From compliance to confidence20AnticipatorsTo be skilled anticipators,pioneers continuou

129、sly adapt their risk monitoring and control processes as the external technology and regulatory environment evolves.They systematically align their AI activities with the organizations overall strategic objectives and tolerance for risk.And they have dedicated teams and built-in mechanisms to contin

130、uously assess and manage current and future risks,creating a platform for rapid adoption of new technology advances.When other companies hesitate to deploy the latest advances in AIsuch as“agentic”AI(AI systems that exhibit increasing levels of autonomy and the ability to take actions independently)

131、for fear of the risks,pioneers can move ahead with confidence,knowing that their responsible AI efforts are ready for the challenge.Responsible by designPioneers are“responsible by design”,placing responsible AI at the center of their overarching AI strategy.They understand that responsible AI is cr

132、ucial to day-to-day operations,providing a robust cross-functional,organizational framework that facilitates decision-making in real-time,supporting the adoption and scaling of new technologies confidently.Pioneers are considered leaders in AI innovation,investing in future planning and horizon scan

133、ning to maintain their competitive edge by perpetually refining and improving their responsible AI governance structure,principles,policies and standards.Proactive partnersPioneers also collaborate closely with their partners to manage risks,support multi-stakeholder participation,feedback and inclu

134、sive decision-making and support forward-looking regulatory compliance.For example,pioneers organize and lead candid discussions between stakeholders in their broader AI ecosystem.The goal of such gatherings is to build bonds of trust that facilitate the sharing of best practices around responsible

135、AI,while also providing a channel for offering technical support to partners that need it.By collaborating with partners on responsible AI,pioneers establish mutual trust,strengthening their own maturity and supporting new business opportunities.What sets responsible AI pioneers apart?As we move int

136、o the generative AI era,organizations that want to leverage responsible AI as a value lever must build on their organizational and operational maturity.Adopting a future-looking mindset to continuously improve and evolve in line with new technology advances and regulatory changes will be imperative.

137、Responsible AI pioneers are:Responsible AI:From compliance to confidence21As our data shows,organizations have reached very different milestones on their responsible AI maturity journey.A company at the principles stage of responsible AI maturity will inevitably have different priorities as it seeks

138、 to improve its responsible AI capabilities,than a company at the program or practice stages.Nevertheless,our research and work advising clients has shown that all companies can benefit from focusing on these five priorities to improve their maturity and begin to reap the benefits of AI.Ready,set,gr

139、ow:five priorities for responsible AI010203Establish AI governance and principlesConduct AI risk assessmentsSystemic enablement for responsible AI testing0405Ongoing monitoring and complianceWorkforce impact,sustainability,privacy,securityResponsible AI:From compliance to confidence22Developing a re

140、sponsible AI strategy and roadmap that includes clear policies,guidelines and controls is critical for the successful implementation,operationalization and governance of responsible AI practices.A foundational step in this process is to define,adopt and enforce a set of responsible AI principles(spe

141、cific to your company priorities,ethics and values)to ensure the ethical design,deployment and usage of AI.Our research shows that 70%of companies have established responsible AI principles,with 54%having translated these into policies.Implementing a robust AI governance operating model on which res

142、ponsible capabilities and controls can be built is another important step.The good news is that 76%of companies have fully operationalized their governance modela dramatic increase from just 31%two years ago.This is reflective of companies taking a top-down,cross-functional approach to responsible A

143、I,rather than attempting to manage risks in an ad-hoc manner.Companies can take this a step further and reinforce this operating model by providing employee training and change management initiatives to promote and sustain responsible AI across the organization.Priority 1 Establish AI governance and

144、 principlesResponsible AI:From compliance to confidence23Understanding risk exposure from an organizations use of AI is a key component of operationalizing responsible AI.If you dont know where youre exposed,how can you protect yourself?Most companies we surveyed appear to be underestimating the num

145、ber of AI-related risks they face.For example,when we presented 13 AI-related risks to our survey respondents and asked which of those risks their company was worried about(they could select as many as they wanted),companies selected just 4.4 risks,on average.This underestimation of AI risk is visib

146、le in another finding:over 50%of companies we surveyed do not have a systematic risk-identification process in place.Adopting a systematic approach to the screening and categorization of risks from AI use cases is important when conducting risk assessments.Quantitative(scenario analysis,stress testi

147、ng and key risk indicators)and qualitative(failure mode-and-effects analysis,root-cause analysis,expert human judgement)tools and assessments can help highlight the risks of an organizations AI use,including fairness,explainability,transparency,accuracy,safety and human impact.Scaling these and othe

148、r tools across an AIs lifecycle and value chain will help companies to better identify and respond to the many AI risks they face.Priority 2 Conduct AI risk assessmentsResponsible AI:From compliance to confidence24When firms comprehensively test and scale responsible AI,they deploy a broad range of

149、risk mitigation measures (we evaluated companies maturity on 44 measures in our survey)across both the AI lifecycle and value chain.Yet just 19%of surveyed companies had scaled more than half of the risk testing and mitigation measures that we asked them about.To comprehensively test and scale respo

150、nsible AI across the organization,companies need to first develop a reference architecture to seamlessly integrate client and other third-party tools and services to evaluate risks across the full AI lifecycle(data,model,application).With this in place,companies can test,fine-tune,integrate and depl

151、oy that architecture across the organization.In addition,they need to provide role-based training(such as for customer-service professionals,HR professionals and managerial roles)to enhance employee skills on the latest responsible AI processes and tools.Priority 3 Systemic enablement for responsibl

152、e AI testingResponsible AI:From compliance to confidence25Establishing a dedicated AI monitoring and compliance function is crucial for ensuring the compliant,ethical and sustainable performance of AI models within an organization.This step is particularly important when it comes to generative AI ap

153、plications,where there is currently less data and model transparency and a far higher frequency in incidents like hallucinations,bias and IP or copyright breaches occurring.As unintended and extended consequences continue to emerge,and companies deploy models or agents that can take actions and make

154、 decisions autonomously,the ability to monitor becomes an imperative.Despite this,43%of companies have yet to fully operationalize their monitoring and control processes,making it the weakest element of organizational maturity.Furthermore,52%of generative AI users do not yet have any monitoring,cont

155、rol and observability measures in place.Getting responsible AI right also requires having employees who are focused full-time on the cause.To do this,companies should start by opening an office tasked with planning,implementing and managing the organizations responsible AI initiatives.That office sh

156、ould be composed of a cross-functional team from a variety of disciplines,with well-defined roles for personnel and a clearly-delineated hierarchy of accountability.Office personnel in the“control tower”,for instance,might be accountable for monitoring risks in real time,while other personnel might

157、be charged with monitoring risks on the horizon.A dedicated AI monitoring and compliance office also needs to be equipped with advanced tools and methodologies to perform at the highest level.For instance,certain tools today can monitor the AI models themselves,track performance metrics and detect a

158、nomalies.Methodologies like drift detection can review data to ensure an AI model remains accurate over time;user feedback integration can make models more responsive to the needs of the people who use them.Priority 4 Ongoing monitoring and complianceResponsible AI:From compliance to confidence26For

159、 a successful responsible AI compliance program,cross-functional engagement must address the impact on the workforce,sustainability,and privacy and security programs across the enterprise.Workforce Talent is a top priority for leaders globally and is among the largest inhibitors of transformation su

160、ccess.Responsible AI is no different.Companies must ensure that their people have the right skills,in the right places,at the right times and know what the expectations are from a responsible AI perspective.A large majority(92%)of the companies we surveyed acknowledged that employees(and end users o

161、f AI)have important roles to play in mitigating risk.However,those same companies reported that developing human interaction capabilities is the single biggest challenge they face to improve their operational maturity,highlighting the need for reskilling in the age of generative AI.As Accentures rec

162、ent work with MIT demonstrates8,employees are the first line of defense on risk mitigation,but to be an effective defense,they need support to overcome potential hurdles(such as low awareness of or complacency toward risk).A multi-disciplinary approach that combines ongoing,high-quality training and

163、 reskilling with behavioral science,responsible design and technical interventions will become critical in ensuring optimized risk mitigation and overall performance.Priority 5 Workforce impact,sustainability,privacy,securityResponsible AI:From compliance to confidence27SustainabilityAwareness aroun

164、d AI carbon emissions continues to grow as researchers look beyond large language model training to inference and the wider AI lifecycle to understand where potential issues lie and solutions that can be applied.For example,training BLOOM,a single multilingual language model,produced 24.7 tons9 of e

165、missionsthe equivalent of 15 round-trip flights between New York and London.10 As a result,Amazon,Google and Microsoft have all announced plans to adopt atomic energy to support growing energy needs.Our survey results showed that only 36%of companies have set up carbon reduction strategies at the or

166、ganizational level.As new measurement and mitigation tools and techniques emerge over the next 12-24 months,companies must make sure they are set up to adopt these to manage AI carbon emissions and reduce cloud compute costs.CybersecurityTo help mitigate cybersecurity risk,the need for a dedicated A

167、I monitoring and compliance office is particularly urgent.For example,71%of surveyed companies said they had an AI-focused cybersecurity response plan in place.But far fewer(40%of firms)had designated a specific team to implement their response plan in the event of an incident.Although this space is

168、 still nascent,companies cannot afford to take a reactive approach.Responsible AI:From compliance to confidence28Conclusion:Turn AI risk into business valueIf the pursuit of responsible AI were ever merely a compliance afterthought,those days are long gone.Companies today know that to maximize their

169、 investments in generative AI and other AI technologies,they need to put responsible AI front and center.This is,of course,easier said than done,for the challenges of scaling responsible AI across the organization are daunting.AI-related risks are exploding.New AI-focused laws and regulations are ge

170、rminating everywhere.And AI value chains are growing in complexity,especially as more companies become both developers and buyers of AI models.Companies must embrace the five priorities above and become responsible AI pioneers if they want to stay competitive.As part of these efforts,companies must

171、pursue an anticipatory mindset,commit to continuous improvement and extend their focus beyond their organization to their entire value chain and wider AI ecosystem.The reward for becoming a responsible AI pioneer will be considerable:consistently turning AI risk into tremendous business value.Respon

172、sible AI:From compliance to confidence29About the researchFrom January to March 2024,Accenture surveyed 1,000 companies across 19 industries,headquartered in 22 countries.The sample includes a diverse group of business leaders,including CEOs,C-suite executives,Board members and directors.The questio

173、nnaire for the survey was co-developed with Stanford University,enabling a robust and comprehensive approach.The methodology of the responsible AI maturity index is based on three core pillars:organizational maturity,operational maturity and generative AI maturity.Each of these pillars is assigned e

174、qual weight and their combined scores create the overall responsible AI maturity index,offering a holistic view of companies readiness in the AI landscape.RegionSampleNorth America27%Asia22%Europe30%Central and South America9%Rest of the world12%TOTAL100%Industry(grouped)SampleAerospace,Automotive&T

175、ransport16%Communications,Media&Technology16%Financial Services15%Healthcare&Life Sciences10%Products16%Public Services6%Resources21%TOTAL100%About the Accenture-Stanford responsible AI maturity scoring modelScore:0 100Score:0 100Score:0 100ApplicationInfrastructureEvaluation Organizational maturity

176、Operational maturityGen AI maturity2550751000PrinciplesProgramPracticePioneerProvider selectionReliabilityRisk mgmt.FairnessGovernanceHuman interactionModel dev.TransparencyRisk ident.DataSponsorshipCybersecurityCybersec.End-userTrainingAccountabilityMon.&ctrl.Env.footprintProcurementModelMon.,ctrl.

177、&observabilityPost-deployment monitoringOverall responsible AI maturityResponsible AI:From compliance to confidence30Authors Acknowledgements Arnab ChakrabortyChief Responsible AI Officer,AccentureKarthik NarainGroup Chief ExecutiveTechnology and Chief Technology Officer,AccentureSenthil RamaniGloba

178、l Lead Data&AI,AccentureResearch directorPatrick Connolly Research teamPhilippe RoussierePraveen TanguturiJakub WiatrakShekhar TewariDevraj PatilDikshita VenkateshDavid KimbleMarketing team Sophie BurgessErika MarshallDries CuypersMicaela Soto AcebalMark KlingeReferences1.An“AI value chain”incorpora

179、tes not only the internal processes of a company,but also those of its partners and customers in creating,deploying and maintaining an AI system.2.The 19 industries were:aerospace/defense,automotive,banking,capital markets,chemicals,telecommunications/media/entertainment,consumer goods and services,

180、energy,healthcare,high tech,industrial equipment,insurance,life sciences,natural resources,public services,retail,software/platforms,travel/transport and utilities.The 22 countries were:Argentina,Australia,Brazil,Canada,China,Denmark,Germany,Finland,France,India,Italy,Japan,Mexico,Norway,Saudi Arabi

181、a,Singapore,South Africa,Spain,Sweden,United Arab Emirates,United Kingdom and United States.3.https:/ AccentureAbout Accenture ResearchAccenture is a leading global professional services company that helps the worlds leading organizations build their digital core,optimize their operations,accelerate

182、 revenue growth and enhance servicescreating tangible value at speed and scale.We are a talent and innovation-led company with 774,000 people serving clients in more than 120 countries.Technology is at the core of change today,and we are one of the worlds leaders in helping drive that change,with st

183、rong ecosystem relationships.We combine our strength in technology and leadership in cloud,data and AI with unmatched industry experience,functional expertise and global delivery capability.Our broad range of services,solutions and assets across Strategy&Consulting,Technology,Operations,Industry X a

184、nd Accenture Song,together with our culture of shared success and commitment to creating 360 value,enable us to help our clients succeed and build trusted,lasting relationships.We measure our success by the 360 value we create for our clients,each other,our shareholders,partners and communities.Visi

185、t us at .Accenture Research creates thought leadership about the most pressing business issues organizations face.Combining innovative research techniques,such as data-science-led analysis,with a deep understanding of industry and technology,our team of 300 researchers in 20 countries publish hundre

186、ds of reports,articles and points of view every year.Our thought-provoking research developed with world leading organizations helps our clients embrace change,create value and deliver on the power of technology and human ingenuity.For more information,visit Accenture Research on material in this do

187、cument reflects information available at the point in time at which this document was prepared as indicated by the date in the document properties,however the global situation is rapidly evolving and the position may change.This content is provided for general information purposes only,does not take

188、 into account the readers specific circumstances,and is not intended to be used in place of consultation with our professional advisors.Accenture disclaims,to the fullest extent permitted by applicable law,any and all liability for the accuracy and completeness of the information in this document an

189、d for any acts or omissions made based on such information.Accenture does not provide legal,regulatory,audit or tax advice.Readers are responsible for obtaining such advice from their own legal counsel or other licensed professionals.This document refers to marks owned by third parties.All such third-party marks are the property of their respective owners.No sponsorship,endorsement or approval of this content by the owners of such marks is intended,expressed or implied.Copyright 2024 Accenture.All rights reserved.Accenture and its logo are registered trademarks of Accenture.