1、Adopting AI Responsibly:Guidelines for Procurement of AI Solutions by the Private SectorI N S I G H T R E P O R TJ U N E 2 0 2 3In collaboration with GEPImages:Getty Images 2023 World Economic Forum.All rights reserved.No part of this publication may be reproduced or transmitted in any form or by an

2、y means,including photocopying and recording,or by any information storage and retrieval system.Disclaimer This document is published by the World Economic Forum as a contribution to a project,insight area or interaction.The findings,interpretations and conclusions expressed herein are a result of a

3、 collaborative process facilitated and endorsed by the World Economic Forum but whose results do not necessarily represent the views of the World Economic Forum,nor the entirety of its Members,Partners or other stakeholders.ContentsForeword 3Executive summary 4Introduction 5The AI acquisition framew

4、ork 111 Business strategy 122 Commercial strategy 153 Data strategy 184 Ethics and sustainability 205 Governance,risk and compliance 246 Conclusion 28Appendix 29Contributors 30Endnotes 32Adopting AI Responsibly:Guidelines for Procurement2ForewordAs the use of artificial intelligence(AI)continues to

5、grow across industries,it has become imperative for commercial enterprises to ensure that the AI solutions they procure are responsible and ethical.The growth of the global AI market is staggering valued at$136 billion in 2022 and is estimated to expand at a compound annual growth rate(CAGR)of 37.3%

6、from 2023 to 2030.The growing use and prevalence of generative AI technology and the emergence of advanced GPT models with their tremendous potential and unique ethical challenges underscores an even greater need to establish standards for responsible AI practices and procurement.Using AI responsibl

7、y is important not only because organizations have a social responsibility and a reputation to maintain but also for the long-term viability and well-being of business and society.That includes protecting businesses from potential risks down the road,particularly related to areas such as intellectua

8、l property,data and privacy.While AI systems are rapidly evolving,in-house AI expertise often stays limited,and there are no standard benchmarks and assessment criteria to aid the end-to-end procurement process.To address this gap,the World Economic Forum,in partnership with GEP,has released a compr

9、ehensive guide for commercial organizations,Adopting AI Responsibly:Guidelines for Procurement of AI Solutions by the Private Sector.In this report,we present a structured framework for evaluating the implications of acquiring AI solutions,emphasizing the importance of transparency,accountability an

10、d human-centred design from development to implementation.The framework offers guidance on evaluating potential solutions based on bias,privacy and security.It shares recommendations on integrating ethical principles into the procurement criteria and processes,often involving multiple stakeholders.F

11、inally,it offers insights on how to overlay all these steps with strong governance criteria,ensuring that these principles are applied.In short,this framework is an essential resource for business and procurement leaders who want to guarantee that their organizations are at the forefront of socially

12、 responsible AI adoption.Mudit Kumar Vice-President,Consulting,GEPCathy Li Head,AI,Data and Metaverse;Member of the Executive Committee,World Economic ForumAdopting AI Responsibly:Guidelines for ProcurementJune 2023Adopting AI Responsibly:Guidelines for Procurement3Executive summaryThe demand for ar

13、tificial intelligence(AI)solutions in enterprises has grown by leaps and bounds in the past few years,driven by improved data availability,advanced algorithms and increased processing power.While the use of AI tools delivers significant value,it is necessary to approach it carefully and avoid its po

14、tentially negative,and even dangerous,consequences.The World Economic Forum,in partnership with GEP,has released this comprehensive guide for commercial organizations across industries to facilitate the process of identifying,selecting and implementing AI solutions responsibly and ethically.This rep

15、ort is a practical toolkit that will help navigate the challenges of AI procurement through a structured framework.It is directional,not prescriptive and industry agnostic,not problem specific.It discusses the need for AI solutions to be closely aligned with business goals,ethical standards and regu

16、latory requirements,as well as the importance of stakeholder collaboration and an enterprise-wide evaluation process.It highlights five key considerations business strategy,commercial strategy,data strategy,ethics and sustainability,and governance,risk and compliance against which AI/machine learnin

17、g(ML)solutions can be assessed for responsible AI acquisition,with procurement as the orchestrator driving the implementation of this framework.Broadly,the report provides practical advice on:Assessing AI solutions ethical standards and regulatory compliance Aligning solutions with business and comm

18、ercial goals Evaluating their potential impact.From automating repetitive tasks,reducing errors or risks and optimizing pricing to identifying new opportunities,making accurate forecasts and enabling investment decisions,AI solutions can help enterprises perform a wide range of tactical and strategi

19、c activities to improve efficiency and support growth.With careful and responsible procurement,organizations can harness the power of AI to improve their productivity and gain a competitive edge.The customizable framework in this AI procurement guide aims to open doors for organizations looking to e

20、thically exploit AIs disruptive power.Offering practical guidance for responsible procurement and implementation of AI solutions in the private sector.Adopting AI Responsibly:Guidelines for Procurement4IntroductionArtificial intelligence(AI)in an enterprise refers to the use of AI technology and app

21、lications in private sector organizations to improve operations,increase efficiency and drive growth.AI adoption in enterprises has grown exponentially over the past decade because of the greater availability of data,the development of more sophisticated algorithms and the increased processing power

22、 of computers.Advancements in cloud computing and the emergence of edge computing have also made implementing AI-powered solutions easier and more cost-effective.However,private sector enterprises find navigating the procurement and deployment of AI technologies a tremendous challenge because of a v

23、ariety of reasons,from lack of requisite skills to the definition of strategy,data quality and established playbooks to follow.This report aims to facilitate the responsible and ethical procurement of AI/machine learning(ML)solutions in commercial organizations by developing comprehensive guidelines

24、 and a practical toolkit that are directional rather than prescriptive.It will help enterprises evaluate AI/ML solutions through a robust procurement framework and build a holistic approach for acquiring and deploying AI solutions to achieve organizational goals.Why enterprises need AIThe history of

25、 AI in enterprises can be traced back to the early days of computing when businesses first began experimenting with rule-based and expert systems,which were designed to perform specific tasks,such as analysing data and making predictions.These systems were limited in their capabilities and required

26、significant human input to function effectively.In the 1990s,ML algorithms began to be developed,allowing AI systems to learn from data and improve performance over time.This marked a turning point in the history of enterprise AI,as ML-powered systems could be used to analyse large amounts of data a

27、nd make predictions more efficiently and accurately.In the 2000s,advancements in computing power and the availability of large volumes of data began to drive the adoption of AI in enterprises.Businesses,realizing the potential of AI to improve operations and drive growth,started making significant i

28、nvestments in AI-powered systems.One of the key drivers of AI is automation.AI-powered automation can help businesses automate repetitive and time-consuming tasks such as data entry,customer service and inventory management and improve speed and accuracy as well as reduce costs.It can free up human

29、workers to focus on more complex and strategic tasks.AI-powered decision-making is another common use case in enterprises,where AI is used to analyse data and provide recommendations to decision-makers.AI can be used to optimize pricing,identify new business opportunities or even make investment dec

30、isions.While AI is becoming an increasingly important tool for organizations,to fully realize its benefits,they need to invest in the necessary infrastructure,talent and technologies.Additionally,businesses should address the challenges associated with acquiring and implementing AI solutions,such as

31、 lack of expertise,data privacy and security concerns.Why do we need responsible AI procurement guidelines?AI/ML adoption is skyrocketing,especially post-COVID-19,with far-reaching impact on businessesOrganizations worldwide are increasingly adopting AI and ML to support business growth,improve effi

32、ciencies and gain a competitive edge.The global AI market is anticipated to hit the$1,500 billion mark by 2030,1 driven by the substantial investments made by tech giants in R&D to advance technology.In many cases,AI solutions are deeply embedded into the organizations enterprise applications,making

33、 recommendations and predictions and influencing critical decision-making.Thus,its essential to ensure that AI delivers ethical,responsible and reliable solutions.Organizations must perform critical evaluations of AI solutions at the time of selection and establish an ongoing review process to make

34、sure that the solution remains aligned with business intent and organizational values.AI/ML solutions should address strategy alignment,data integration,ethical compliance and risk assessment.Adopting AI Responsibly:Guidelines for Procurement5Limited guidance on industry standard practices and ways

35、to minimize organizational risksWhile nearly all C-suite executives view AI as critical,most acknowledge the struggle to navigate the procurement and deployment of AI technologies.Enterprises must consider some key factors while adopting AI,including:Business strategy alignment:Does the AI strategy

36、align with larger corporate strategy?Does the AI solution enable optimized business outcomes,delivering high value to the organization?How does it align with long-term organizational vision and strategy?Business case:What are the non-monetary gains?Is the investment justified?Are commercial terms op

37、timized for long-term return on investment(ROI)and total cost optimization?Technology and data integration:Does the solution meet the companys and regulatory data protection requirements?How do you regulate the outsourcing of data management and hosting?What data is needed and how do you collect and

38、 cleanse it?Ethics alignment:Are the AI system and solution provider aligned with organizational values and complying with ethical standards?Risk assessment:What are the risks?How do you conduct an initial impact assessment to ensure ethical,equitable and sustainable deployment?What are the challeng

39、es and successes the industry has experienced?Agile and collaborative AI system integration:Is a scalable AI system already in place?How can AI applications be integrated into processes so they work effectively for delivery teams?Scope of the guidelinesThe scope of this AI procurement guidebook is f

40、ocused wholly on procuring AI applications.Understanding AI:tools,techniques,enablersFIGURE 1Core Al techniques and models are used to process,solve and learn based on intended application within the organization.Often,core Al techniques are combined with other technologies to drive the desired outc

41、omes.Core Al techniques are mathematical and statistical models and frameworks that are deployed to process vast amounts of data,make decisions,learn about outcomes,store results and use them as additional data point to improve future decisions.Al enablers and support infrastructure are the core fou

42、ndation technologies that an Al ecosystem needs to be successful.Advancements in these technologies improve the overall effectiveness and efficiency of Al solutions.AI applicationsAI applications-Augmented research(marketing,medicine,science etc.)-Autonomous vehicles and drones-Biometric,facial and

43、gesture recognition-Extended reality-Generative product design-Intelligent agents-Predictive systems(maintenance,retail,trading etc.)-Recommendation systems-Robotic process automation-Test,speech,image and video recognitionCore AI techniquesCore AI techniques-Machine learning-Deep learning-Neural ne

44、tworks-Reinforcement learning-Supervised learning-Unsupervised learning-Embodiment-Autonomous robotics-Human augmentationCore AI techniques-Cybersecurity-Data analytics-Ethics and responsibility-Structured and unstructured dataAI support infrastructure-Cloud-High-speed internet and processors-Sensor

45、s,internet of things-Mobile phones and devices-Robotics-Rules-based reasoning-Decision-making-Expert systems-Knwoledge representation-Planning and scheduling-Seach and optimization-Perception-Augmentation and virtual reality-Machine vision-Natural language processingAI enablersAI supportinfrastructu

46、reSource:World Economic Forum,Empowering AI Leadership,2022.Adopting AI Responsibly:Guidelines for Procurement6How the AI guidebook will helpThis report is a detailed,step-by-step guide that will help enterprises interested in acquiring AI solutions address each of the key considerations listed earl

47、ier.Each section comprises questions that can be used to assess and understand suppliers during the procurement process and will help find the best fit for your organizations requirements.While the toolkit focuses primarily on AI software,it can be tailored and adapted for solutions with AI capabili

48、ties.These guidelines will help:Provide a holistic assessment framework by evaluating everything from risk to ethics and biases to enable responsible acquisition of AI Serve as a general,industry-agnostic guidebook for AI acquisition,covering areas such as supplier landscaping,critical stakeholder i

49、dentification,supplier evaluation framework and supplier selection Enable decision-making while balancing business needs with social and ethical responsibility by developing an objective supplier questionnaire-based evaluation Provide a framework for collaboration bringing key stakeholders together.

50、The guidelines are not intended to:Be a lift-and-shift solution for all industries,but an easily customizable guidebook Outline specific regulations or laws to consider,but direct users towards evaluating AI solution providers for appropriate regulations and standards Define acceptable levels for ea

51、ch of the criteria outlined but capture the relevant criteria for responsible AI acquisition Provide a mutually exclusive,completely exhaustive(MECE)supplier questionnaire,but a directionally comprehensive questionnaire to aid decision-making on the procurement of AI solutions Identify suitable phys

52、ical robotic tools and AI hardware systems,but for AI software solutions/applications.How the guidelines were developedThis guidebook was designed by the World Economic Forums Centre for the Fourth Industrial Revolution in association with GEP and in active collaboration with multiple Forum communit

53、y partners representing both buy and sell sides from industries such as oil and gas,energy,construction and consulting,among others.It explores and incorporates key considerations and guidelines for responsible AI procurement in commercial organizations,addressing real-world business needs and chall

54、enges.Over one year,the core team organized several workshops with over 25 community partners representatives from industries and AI solution providers to coalesce collective knowledge of AI and procurement and design standardized guidelines that can be adopted easily across organizations and indust

55、ries.Adopting AI Responsibly:Guidelines for Procurement7How to use the guidelinesBefore an enterprise invests in AI solutions,it should consider the key principles outlined in this report.These guidelines are meant to be followed only after determining that AI is the best solution for the specific p

56、roblem.The guidelines shared here are not meant to be a one-size-fits-all solution for any and all challenges that private sector enterprises may face,but by shaping the way new AI solutions are procured,they can improve the overall fit of the AI solution in the organization.Handling each AI acquisi

57、tion separately can lead to a lack of alignment with the organizations overall strategy.The evaluation of AI solutions requires inputs from multiple internal stakeholders,each best suited to evaluate the solution on specific parameters.Its crucial to have a unified,collaborative approach in evaluati

58、ng and selecting the AI solution.Internal separation and lack of communication within an organization can obstruct the effective functioning of artificial intelligence systems.The procurement team(or,at an enterprise level,the procurement leadership team)for whom these guidelines are primarily desig

59、ned can serve as the central point to ensure that all stakeholders are involved in a timely and appropriate manner,leading to a cohesive effort in acquiring the solution.The guidelines in this report are designed to assist stakeholders in advancing their AI goals and to provide direction for the pro

60、curement of AI-driven solutions.They will help:Business teams to accelerate the achievement of their targets Procurement teams to create requests for proposals(RFPs)for AI products and manage the procurement process efficiently Data teams and practitioners to safeguard organizational interests and i

61、dentify and manage potential risks AI-solution providers to better understand the core expectations in enterprise AI projects and align their proposals with the emerging standards for AI procurement.Procurement and AI strategyFIGURE 2Data gathering and cleansingStakeholder needs across different bus

62、iness unitsSystem implementationand trackingRisk factorsIntegration withexisting IT infrastructureProject strategy,feasibility and scalability Procurement is a key enabler in AI adoptionProcurement is best positioned within an organization to help the C-suite define and create a holisticblueprint fo

63、r the organizations Al strategyProcurementAdopting AI Responsibly:Guidelines for Procurement8The following teams,led by procurement,should collaborate to ensure a holistic evaluation of the AI solution under consideration.Within these teams would be data scientists,data architects,data engineers,leg

64、al and risk management professionals,product owners and business analysts,among others.Procurement and the need for cross-functional collaborationTABLE 1TeamBusiness/end userIT and cybersecurityAI centre of excellence/data teamData managementProcurementRole Assessment and prioritization of use cases

65、 Process definitions Workflow design Defining business requirements Defining expected outcomes Exception management Change management Architecture and integration design Infrastructure assessment and design Security assessment and design Application management and support Platform deployment and sca

66、lability assessment Software and licence management AI and cognitive capability assessment Tool adaptability and configuration of needs Process modelling Service support capabilities Testing and debugging Governance,risk and compliance User interface(UI)/user experience(UX)management Change manageme

67、nt AI ethics and sustainability Data needs and consumption requirements Data governance framework Data integrity requirements Align with the business team on requirements Collaborate with IT and cybersecurity,AI centre of excellence and data management teams to define the scope of the AI project Ass

68、ess the market to identify and shortlist suitable supplier(s)Coordinate with all teams to enable the implementation of the AI solution Support the business team to monitor and measure relevant key performance indicators(KPIs)The guidelines include five key high-level recommendations,each comprising

69、multiple considerations and text explaining the reasoning and substance behind it.It is important to consider the proportionality of the AI solution being procured not all guidelines may be relevant to every procurement decision,and different use cases will determine the importance of each of these

70、considerations.However,following these guidelines will help identify key areas needing further investigation.Embedding guidelines in the sourcing processThe sourcing processFIGURE 3Project scoping and due diligenceMarketintelligenceSourcingexecutionEvaluation and negotiationsContracting andimplement

71、ationAdopting AI Responsibly:Guidelines for Procurement9Sourcing processEmbedding responsible AI procurement guidelinesProject scoping and due diligence Prioritize business use case for AI deployment Define business outcome criteria and objectives Document as-is processes and current gaps that the A

72、I solution will help solve Establish business governance prerequisites Establish risk management requirementsMarket intelligence Perform high-level market analysis on available AI solutions and solution providers Understand complexity of data as well as solution specifications and identify various s

73、upplier types that could help achieve the business goal Qualify suppliers that would best fit requirementsSourcing execution*RFx is a term used to describe a range of procurement processes that businesses use to solicit bids from suppliers.RFx can refer to request for information(RFI),request for pr

74、oposal(RFP),request for quotation(RFQ)or other types of requests for vendor bids.Draft scope requirements for the identified AI application Involve internal functions to identify must-have vs good-to-have functionalities and features Identify what would separate one solution from the other in each o

75、f the RFx*questionnaires Refer sample questions in the guidebook to ensure all focus areas and details are covered Execute RFx for responses from suppliers Prepare supplier evaluation matrix based on criteria across areas of considerationSupplier evaluation Perform objective evaluations through supp

76、lier scorecard-based questionnaire review Do total cost of ownership(TCO)analysis Conduct supplier risk assessmentContracting and Implementation Select suppliers Contract negotiations:ensure all requirements are met through various terms and conditions Consider outcome-based commercial model Draw up

77、 a detailed project implementation plan Create a plan for continuous improvement Plan for continuous retraining of algorithms Develop a framework for business governanceHow procurement teams can use AI procurement guidelines across typical sourcing processesTABLE 2Procurement is well-positioned to w

78、eave these facets together and help the C-suite define and create a holistic blueprint for the organizations AI strategy.Adopting AI Responsibly:Guidelines for Procurement10The AI acquisition frameworkAssess and benchmark solutions to ensure your organizations AI investment is responsible,ethical an

79、d aligned to larger business goals.The foundation of responsible AI acquisition lies in a holistic framework with ethics and sustainability at the core,driving business goals,commercial objectives and data strategy,all strongly supported by an ongoing governance,compliance and risk strategy.There ar

80、e underlying interconnections among these elements,nuances that should be factored in during implementation.The purpose of the developed framework is to ensure that the acquisition of AI systems by commercial enterprises is grounded in ethical principles with strong governance to ensure those princi

81、ples are applied,thereby mitigating bias and driving resilience.The AI procurement framework provides comprehensive directional guidelines rather than prescriptive rules,as the degree to which these factors apply to an AI solution or an enterprise can vary significantly.While it is sector agnostic,t

82、here could be sector-specific concerns that this framework is not designed to address.In the subsequent sections,this report discusses in detail the impact of each of the five pillars of responsible AI acquisition strategy,highlighting the key considerations and questions that should be raised and a

83、ddressed internally(within an enterprise)as well as externally(with the supplier partner/solution provider)through the sourcing life cycle and supplier relationship management(SRM).The report also shares a set of preliminary questions distilled from the comprehensive assessment questionnaire(see App

84、endix)as a first step for organizations considering an AI solution for the first time.While procurement can be the orchestrator driving the implementation of this framework,a strong partnership among the concerned functions will determine its success.Key considerations for responsible AI acquisition

85、FIGURE 4Governance,risk and complianceSustainabilityBusiness strategyCommercial strategyData strategyEthicsAdopting AI Responsibly:Guidelines for Procurement11Business strategy1Evaluate ways AI can create a competitive advantage through enhanced business decision-making,leading to positive changes i

86、n business outcomes.Adopting AI Responsibly:Guidelines for Procurement12As the benefits of AI,from boosting speed and productivity to improving forecasting and decision-making,become more and more visible,its use cases are multiplying.However,prioritization,strategic planning and evaluation will det

87、ermine the long-term winners in the AI game.One of the key considerations in procuring an AI solution is to evaluate and determine the strategic fit and capabilities of the solution in terms of the enterprises current and perceived future needs.Impressed by the inherent capabilities of AI solutions

88、in the market,organizations often try to force-fit the solution to meet their requirements.That is one of the primary reasons AI solutions fail to achieve desired business outcomes.In order to find the most suitable AI solutions,enterprises must define their desired business goals and link them with

89、 the capabilities they want to build.Its just as important to align the AI solution with the organizations technology and data strategy as AI will have a broader impact beyond the immediate application area in the long run.AI centre of excellence/data teams must be involved right from the start whil

90、e determining the scope,timeline and potential organizational impact.Solution adaptability should be another key consideration.Over time,an organizations business goals could evolve,or there could be changes in the underlying technology landscape,meaning the solution must adapt to new requirements a

91、nd continue to achieve the desired outcomes.Additionally,the AI solution should be able to detect and handle gradual and abrupt concept drifts and incorporate the required algorithm changes.The solution should also be evaluated for capabilities to scale on demand with minimal complexities.The soluti

92、on providers thought leadership and vision in the AI application domain and the underlying core AI techniques can help ascertain the ability to deliver on the organizations desired business goals.Achieving business goals:questions to askTABLE 3SpecificationsKey considerations1.1 How will the AI solu

93、tion deliver on expected business outcomes?Understanding how the solution will help enterprises meet their business goals.Is the supplier able to understand your business goals and explain how their capabilities will help achieve them?Are the solutions capabilities outlined clearly,and are there dem

94、onstrated use cases they can reference?How much custom development will be required for the AI solution to meet your requirements?What level of guarantee does the solution provider give for the process and business outcomes?Could the solution scale up to meet increased demand/productivity?If your bu

95、siness slows down,is the solution flexible enough to adjust to the changed needs?1.2 Is there transparency about what the AI solution can and cannot achieve?Often,businesses are not fully aware of the internal working mechanism of the AI model being used.It is the responsibility of the solution prov

96、ider to educate potential buyers and be transparent about the capabilities and limitations of their offering.Does the supplier explain the techniques applied in the AI system,including the use of algorithms and associated software libraries for the algorithms?Is the supplier able to articulate the w

97、orkings of the solution in an easy-to-understand manner?Does the supplier dedicate the required resource(s)to train/educate your team about the AI solution?Does the solution offer explainable results and transparency in the decision-making process?Does the supplier recognize and describe any limitat

98、ions of the AI system for the problem you want addressed?1.3 Can a non-AI solution deliver the same outcomes?Like any other technology,AI is not a magic bullet for all problems.It is essential that businesses explore alternative technology/solutions before deciding on an AI solution.Can you justify

99、why the use of AI/ML is the optimal approach to meeting the specified business goals?Can any other cost-effective,non-AI technology or process be used to achieve the expected business outcomes?Adopting AI Responsibly:Guidelines for Procurement13SpecificationsKey considerations1.4 Is there an inheren

100、t feedback loop in the system to adjust to changes in the external environment?An AI solution can become a self-learning system,provided a sufficient and appropriate feedback loop is embedded into it.Does the solution have an automatic feedback/retraining loop,or is there a human in the loop?How doe

101、s the solution measure business outcomes and user satisfaction?How do those insights get delivered and/or feed into any product changes or updates?1.5 How effectively can the solution be updated to accommodate change in requirements?AI is a highly dynamic system,so the supplier should partner with t

102、he buyer to redesign the solution in case business requirements change.This would be applicable especially if data complexity is high.*A steady state is when the system is acting in the intended manner,and all high-severity incidents are resolved How easy is it to adapt or customize the solution dur

103、ing the initial and steady state*?Does the solution update its behaviour based on newly ingested data?How will the supplier help in course correction if the AI model or prediction delivers unexpected results?How is the service tested and monitored for model or performance drift over time?Does the su

104、pplier provide KPIs for monitoring any performance drifts that may prompt retraining of the model(if there are unexpected changes)?Does the supplier provide documentation detailing how the AI system can be reconfigured or adapted if the results are not delivering the goals?1.6 Is the supplier a thou

105、ght leader in this domain?All new technologies require partnerships with suppliers that are thought leaders in this domain as their vision will steer the solution to meet future requirements.How much does the supplier invest in R&D in the domain of their solution and company requirement?How does the

106、 supplier product development stay in line with market trends?Does the supplier publish reports offering industry best practices and actionable insights in terms of optimizing AI outcomes?Does the supplier organize in-person/virtual networking events for its clients to increase awareness of existing

107、 AI solutions?How does the supplier solution differentiate itself from its competitors?Is the supplier investing in upskilling and training its talent pool to stay relevant with state-of-the-art technology?Achieving business goals:questions to ask continuedTABLE 3Adopting AI Responsibly:Guidelines f

108、or Procurement14Commercial strategy2How does the AI solution align with the broader commercial strategy?Adopting AI Responsibly:Guidelines for Procurement15Business strategy defines how organizations want to achieve their business objectives and is complemented by an accompanying commercial strategy

109、 that defines how much an organization would be willing to spend to achieve those goals.Generating commercial benefits for the enterprise is one of procurements key goals.In this case,it means identifying the most suitable AI solution with the most favourable commercial terms.These terms include cos

110、t,duration,payment terms,invoicing plan,discounts,rebates and investment risk,among others.However,commercial strategy should not be viewed solely through the lens of monetary returns;rather,it should focus on the value delivered for the money invested.As for all investments,procurement should ensur

111、e that the purchased AI solution can deliver on the organizations ROI expectations.ROI should include the overall value and potential benefits generated for the organization.Choosing and defining the most suitable KPIs while challenging for new technologies with many unknowns is crucial for continua

112、lly monitoring whether the investment is yielding value at all stages.The commercial considerations for an AI solution should be approached with slight variations compared to investing in other technology solutions.The timeframe for TCO is usually three to seven years for a non-AI solution.However,A

113、I is typically deployed for a longer term,so the cost considerations should include a much longer timeframe.Also,the ongoing cost of retraining and managing concept drifts will need to be considered while calculating the TCO.When it comes to new technologies,it is important to manage uncertainties o

114、f business outcomes.An effective AI system will deliver invaluable outputs to any business.However,as with all new technology,the learning curve and the unknowns carry unforeseen risks.This makes it imperative to have strong governance in place to protect investment.Supplier risk assessment,performa

115、nce management and appropriate risk mitigation plans with linked financial compensations are a few factors to consider.Organizations should look to include outcome-based pricing models to drive joint accountability and mitigate their investment risk.An effective AI system will deliver invaluable out

116、puts to any business.However,as with all new technology,the learning curve and the unknowns carry unforeseen risks.Adopting AI Responsibly:Guidelines for Procurement16SpecificationsKey considerations2.1 What is the expected value of the solution to be delivered?Commitment to business case requiremen

117、ts will be a key ingredient in a supplier partnership.In what ways is the supplier committed to achieving the business case objectives?Are there any dependencies/assumptions made by the supplier on achieving these objectives?Are there ways in which the value delivered is optimized during the steady

118、state*?Will the solution require any co-development/co-innovation?How will you benefit from the potential gains of the supplier in case of co-innovation?2.2 Do you understand all the costs involved in the purchase and maintenance of the AI system?Determining the viability of an AI solution for a lon

119、ger or shorter term can be done by calculating the TCO,which includes all direct and indirect costs throughout the lifetime of the solution from acquiring,building,running to retiring.*Much of the cost of AI projects is internal;the process of getting data ready to enable an AI solution is time-cons

120、uming and expensive.Can the supplier provide a breakdown of one-time costs based on project milestones/measurable deliverables?Has the supplier provided a resource-wise effort estimation required for each of these milestones?Are the recurring cost components clearly defined?Is there clarity on the f

121、requency of the recurring costs?What would be the cost of changes required in the existing systems/infrastructure?*Has the supplier provided the hourly/daily rates of consulting/technical resources to be involved in the project?What are the relevant skills/years of experience in the AI domain of the

122、 resources the supplier plans to deploy?Does the supplier offer different pricing options to accommodate the gradual expansion of the AI system deployment in the client environment?What are the discounts/rebates offered by the supplier?2.3 How can organizations mitigate the investment risks of the s

123、olution?There are multiple unknowns when organizations invest in new technologies and the team should evaluate possibilities to mitigate investment risks.Is the supplier able to delink the discovery and execution phases to restrict upfront investment while the business case is being tested?Can the s

124、olution/implementation be in phases and gradually widened to avoid large upfront costs?For e.g.implementation of a low-cost model or proof of concept before full-scale implementation?Is the supplier able to commit to an outcome-based pricing model?What would be the KPIs that can be linked to outcome

125、s?Is the supplier financially solvent?Is the supplier currently going through or has planned for an acquisition/merger?Who will own the AI model in the event of insolvency or ownership transfer?(e.g.code escrow,data escrow,model escrow)Meeting commercial targets:questions to askTABLE 4Adopting AI Re

126、sponsibly:Guidelines for Procurement17Data strategy3Evaluate the effectiveness of the AI solution in terms of analysing valuable data in support of enterprise objectives.Adopting AI Responsibly:Guidelines for Procurement18Analysing data and learning from it is the fundamental job of any AI solution,

127、so providing relevant data is an important first step.A robust data strategy serves as the foundation for any successful AI deployment.Organizations will have to prepare the input dataset for the AI solution,and the costs associated with this will be different.For instance,the cost will be higher fo

128、r those with data swamps rather than data lakes.2 Next comes the question of data relevance.Even if the data is available,is it the right data?Will it be sufficient to generate an exhaustive list of use cases for the purpose and scope of the AI solution?How can this data be accessed?The project busi

129、ness and central data management teams will have to work jointly to arrive at these answers.If the internal data available is insufficient for training,and the data acquisition costs are high,then generating synthetic data from the available internal dataset or going to the market for synthetic data

130、 sources may be the next best approach.3In parallel,organizations should start thinking about their future data strategy as they will have to manage the evolving requirements that arise from AI implementation.They should explore potentially suitable data platform(s),build a scalable data pipeline,es

131、tablish a centralized data governance mechanism to act as a support function and enable smaller teams to execute a top-down data strategy,devising policies and procedures for effective data management and adherence to legal and regulatory requirements.Organizations can also consider making use of bu

132、siness opportunities from the data they collect.How can additional benefits be derived from data being traded back to the market/ecosystem in terms of negotiating better contract terms with AI suppliers or generating additional revenue streams?Driving data for success:questions to askTABLE 5Specific

133、ationsKey considerations3.1 Do you have a clear and defined as-is and to-be data strategy?To maximize business value from implementation of the AI solution,organizations need to assess,identify gaps and adopt a long-term data strategy.What type of data platform does your organization currently have

134、in place?How will the new AI implementation impact your existing data platform?What is the enterprise data structure strategy if you are planning to migrate to the cloud and use cloud-based AI capabilities?What data will go on the cloud?Are you looking at single or multi-cloud providers?Do you want

135、to prioritize modularity or flexibility?Is there an existing data culture across the organization?Is senior leadership involved in establishing and communicating data strategies?How are different parts of the organization incentivized to share and reuse data?Are employees being trained in data quali

136、ty management?Does the data-based feedback need to be looped back into business decisions and how?Is there a designated team for data collection,validation,storage,governance,security and accountability structures across the entire data supply chain?Who is responsible for ethical data use in the ent

137、erprise?3.2 What are the different sources of data to be considered?For organization-specific contextual learning process,internally available data is ideal.Otherwise,external sources of data(synthetic data)can be considered.What are the different kinds of data expected to be ingested by the AI solu

138、tion?Can the AI solution meet existing and new enterprise objectives based on internally available data?Do you know where and how to collect the data internally?Is the available internal data ready to be consumed is it accurate,complete,consistent and up to date?Will external sources of data be need

139、ed?What is the cost of acquiring/generating synthetic data?Can the external data provider ensure complete,relevant,unbiased and timely data?Who is responsible for ensuring the quality,usability and reliability of third-party data?Should there be different contractual provisions for the exchange of d

140、ifferent categories of internal data and related data models outside your organization?For example,you may want to limit/encrypt sharing of data/data models built on personal identifiable information(PII)-related data.Is the collection of additional data necessary in the future for optimal AI perfor

141、mance?3.3 Does the supplier have satisfactory data management practices?The data for the AI model should meet data quality standards defined by the governance team.Furthermore,data ownership and management,right from storage to extraction,need to be efficient,secure and adhere to regulatory requirem

142、ents.Does the supplier have adequate data quality assurance processes and frameworks around storage,management and transfer of data?Who will have the ownership and be accountable for the data and derivative models?Does the supplier have the standard data privacy/security frameworks for its industry?

143、How will the supplier follow data privacy practices for sensitive data that falls outside of the General Data Protection Regulation(GDPR)?Where is the data collected for the AI model stored?What are the security measures to prevent a data breach?Who has access to the stored data?Elaborate on the lia

144、bility in case of data breach.Adopting AI Responsibly:Guidelines for Procurement19Ethics and sustainability4Evaluate if the AI solution creates a net positive impact from an ethical and sustainability perspective.Adopting AI Responsibly:Guidelines for Procurement20A responsible AI system upholds eth

145、ical standards of fairness,transparency,inclusivity and accountability while being environmentally sustainable and balancing profit with social responsibility.The best way to ensure that your organization invests in a responsible AI solution is by having dedicated internal stakeholders take responsi

146、bility and accountability for various facets of the AI implementation process.For instance,the chief analytics officer or chief data officer could oversee data governance,providing AI-focused training to staff and collaborating closely with third-party providers of data,AI services and software.A de

147、dicated team or function could oversee project direction and change management;another team could oversee the execution,solution adoption and performance tracking.However,certain pitfalls(such as bias and coding of unethical standards into the system)can corrupt the AI solution.Designers could unkno

148、wingly introduce bias into the model,or biases may enter the system through a training data set or during training interaction with end users.Hence,eliminating bias should be one of the top priorities during the selection and deployment of an AI system.The solution provider should have sufficient ch

149、ecks and balances within the system to eliminate or minimize potential biases and ethical issues from the outcomes.Its also important for the suppliers development team to be aware and consciously adopt an ethical mindset while designing and building the system.An ideal partner will have a mitigatio

150、n plan in case the AI solution starts producing biased outcomes.Conversely,these biased outcomes can be pre-empted if the supplier prioritizes diversity and inclusivity within the team from the initial stages.Seven core values for ethical AITABLE 6 Common good:The purpose of AI solutions should be t

151、o accelerate the development of society and improve quality of life of all people.Accountability:All stakeholders involved in the development and deployment of the AI solution should be compliant with standards and legislations,and accountable for the outcomes and impacts of the solution.Inclusivene

152、ss:Ensuring that the members of AI design teams themselves are diverse and inclusive will help proactively address biases as well as increase the trust external stakeholders place in the solution provider.Non-maleficence:All stakeholders should ensure that AI does not put humans at risk of harm,eith

153、er in the intended or unintended outcomes of its use,and that it is not used for nefarious purposes.Accessibility:AI technology and the knowledge on its development,deployment and use should be made widely accessible so that individuals can develop the ability to use AI technology and understand the

154、 potential it brings.Transparency:The decision-making process of the AI algorithm should be easily explainable,and developers should be transparent about its purpose and intentions to promote fairness and equity.Equitability:AI solution designers should keep in mind equitable outcomes for all from v

155、ulnerable communities to people with disabilities.Sources:World Economic Forum,Quantum Computing Governance Principles,2022;World Economic Forum,A Blueprint for Equity and Inclusion in Artificial Intelligence,2022.Emerging ethical risks such as misinformation(e.g.deepfakes),over-reliance and loss of

156、 skills,which already exist in some sectors today,could now affect many services in the supply chain because of the extensive use of AI.For example,AI-powered chatbots in customer service can lead to over-reliance on AI,no human involvement and a lack of empathy in the redressal of complaints,which

157、can reduce customer satisfaction.Mitigation measures for such risks need to be addressed with suppliers during due diligence and should be embedded in commercial clauses.Sustainability metrics are just as important when evaluating an AI solution.In its current form,AI is not a sustainable solution.S

158、ustainability in AI would mean developing AI systems that are compatible with sustaining environmental resources for the present and future generations.A responsible AI provider should have a clear road map to mitigate the environmental impact(such as a high level of energy consumption and resource

159、depletion)caused by the technology.Enterprises the buyers of AI solutions must also have processes in place to encourage the adoption of a responsible AI solution.Employees within the buyer organizations must have the right incentives and be recognized for doing the right(sustainable)thing.Companies

160、 should have a vetting process as part of their AI systems pre-implementation review to ensure that ethical and sustainable considerations have been addressed.Also,an organizational framework should be developed that maps the roles and responsibilities of each team involved and an escalation procedu

161、re in case of questionable actions.Companies should have a vetting process as part of their AI systems pre-implementation review to ensure ethical and sustainable considerations have been addressed.Adopting AI Responsibly:Guidelines for Procurement21Building an ethical and sustainable future:questio

162、ns to askTABLE 7SpecificationsKey considerations4.1 Describe the approach to eliminate or minimize bias and ethical prejudices from the AI solution.Understanding how the supplier and the solution will mitigate any potential bias in the outcomes.*Equalized odds and disparate impact are metrics used t

163、o measure fairness in ML algorithms*Zip code or postal code is a common proxy variable for socioeconomic status,which can introduce biases in the prediction of certain outcomes*Consequences scanning is a process that deliberately inserts friction into the product development process with the goal of

164、 identifying and mitigating negative or unintended consequences What methods are used to train the AI model?Do the training methods introduce fairness(eliminating biases like gender,age,ethnicity,region),interpretability,privacy and security into the model?What data is the supplier using to train th

165、e AI solution?Where is the data being sourced from and are there multiple data sources?Has the supplier trained the model on representative and complete data sets?How can you test for bias in training data?Does the model come with a set of model cards clearly describing the nature of the model and i

166、ts training data?Does the supplier use any open-source ethical tooling to assess the bias factor?Can the supplier identify possible sources of bias or unfairness and where they come from whether its data or techniques implemented or other sources?Is a conscious effort being made to implement checks

167、and balances in the AI to prevent biases from creeping in?Does the supplier use any metrics(e.g.equalized odds or disparate impact*)to evaluate bias in AI?Has the supplier monitored for any proxy variables that could lead to biased predictions?(e.g.postal code data*)Has the supplier published any wh

168、ite papers or technical reviews to evidence the technical robustness and unbiased outcomes of its models predictions?Does the supplier use any design ethics techniques?(e.g.consequences scanning*)Would the supplier be willing to participate in these exercises/procedures hosted by you?Does your solut

169、ion address new ethical risks such as misinformation,over-reliance and loss of skills?4.2 Describe the process for ensuring that the solution development team adopts an ethical mindset.The development team must be aware and responsible while designing ethical standards for their AI solution.Does the

170、 AI and ML team take responsibility for how their work will be used?What are the ethical standards encoded into the AI solution?Does the supplier offer training or have an awareness programme to ensure its team understands the potential impact of creating an AI system that produces an incorrect or d

171、isproportional output?If not,would the supplier be open to AI ethics training offered by your organization?Have there been any instances in the past when the AI has not met performance requirements and delivered unintended results?What were the corrective measures taken for better outcomes?Is there

172、any redressal mechanism if individuals are affected negatively?Does the supplier consider user accessibility needs during design thinking?For example,many facial recognition technology(FRT)cameras are ineffective for individuals using wheelchairs as they cannot be lowered to an appropriate height.Ad

173、opting AI Responsibly:Guidelines for Procurement22SpecificationsKey considerations4.3 How is the suppliers approach toward ethics at an organizational level?Diving deeper to understand how the supplier organization looks at ethics and sustainability.Does the suppliers code of conduct(or equivalent)s

174、pecifically highlight the ethical use of digital technology,including AI ethics?Does the supplier employ a chief AI ethics officer?What is their expertise and experience in this space?Has the suppliers senior leadership endorsed AI ethics?Are there public statements or other evidence to demonstrate

175、the suppliers engagement with ethics?Does the supplier have an AI ethics ambassador network?Did diverse teams build the AI system?Does the supplier have a policy on inclusivity or diversity,equity and inclusiongoals?Does the supplier use an AI ethics impact assessment?If not,can they provide relevan

176、t information so your organization can complete its assessment?Does the supplier have an AI ethics board?Does it have the enforcement power to prevent the engineering of AI systems that are inconsistent with its organizational values?Does the supplier have a whistle-blower protection policy?Does the

177、 supplier work with law enforcement and/or military agencies?What type of AI is being supplied to these organizations?For instance,providing a biased facial recognition system for policing can create problems.Does the supplier provide responsible use guidelines?Does the supplier have a glossary of t

178、erms for ethics?4.4 Is the supplier concerned about the environmental impact caused by its AI solutions?What steps are being taken to mitigate the impact?Evaluate whether the solution provider has a clear road map in place to mitigate the environmental impact caused by the AI technology.What is the

179、suppliers monthly(average)energy consumption(kWh)?How much energy(kWh)is spent every month by the supplier for training its AI solutions?What are the measures or techniques the supplier has adopted to optimize its overall energy consumption?Does the supplier have a road map to become energy efficien

180、t in the next three to five years?What are the hardware systems the supplier uses to power its AI solution?Is the hardware infrastructure optimized to reduce energy consumption?Does the supplier use renewable energy sources to meet its energy requirement for training the AI solution?If yes,whats the

181、 breakup of the percentage of energy derived from different energy sources?Has the supplier considered possible environmental harm during data collection?For example,collection and processing of large amounts of data can require significant amounts of energy,particularly if the data is stored in dat

182、a centres or on cloud servers.Building an ethical and sustainable future:questions to ask continuedTABLE 7Adopting AI Responsibly:Guidelines for Procurement23Governance,risk and compliance5Build and integrate risk management plans with the AI solution and improve business resilience,especially to AI

183、-related cyber risks.Adopting AI Responsibly:Guidelines for Procurement24While AI can deliver an incredible increase in benefits,the consequence of its errors can be just as huge.Every industry faces a different kind of risk based on target data collection points,regions,data safety and legal compli

184、ance,among other things.Data sourcing is one of the key focus areas to manage the risk associated with deploying an AI solution.Today,enterprises have a vast volume of data to exploit,but this comes with issues such as data privacy and geopolitical and cybersecurity risks.Data breaches can destroy b

185、usiness reputation,lead to legal actions and result in huge loss of revenue.The average cost of a data breach in the US has almost doubled over the last 10 years jumping from$5.4 million in 2013 to$9.4 million in 2022.4 Another damaging risk related to data is bias in algorithms.The AI model learns

186、from the dataset it is trained on,and bias in the dataset can result in undesirable outputs.Cyberattacks are another significant risk.These have the potential to impact the integrity of the AI models decisions and predictions;hackers can take control of the solution or deliver manipulated and/or mal

187、icious inputs that result in the poisoning of the AI algorithm.Used incorrectly or negligently,AI can leave an organization exposed to operational,financial,regulatory and reputational hazards.Safeguards must be put in place to ensure that the AI solution functions as intended,and this calls for a s

188、trong governance system designed with AIs unique characteristics and capabilities in mind.AI governance is the process of setting policies and establishing accountability to drive the development and deployment of AI systems in an organization.Its a broad framework that uses a variety of processes,a

189、pproaches and tools to ensure that an organization uses AI technology favourably and responsibly.AI governance encompasses risk management,regularity compliance,contractual agreements and ethics.When done properly,AI governance fosters agility and trust in an organization.Capturing and managing meta

190、data on AI models creates transparency into how AI systems are built and deployed,which is a critical prerequisite for most regulatory concerns.Many tools and calibrated assessments are available in the market that can help assess the impact of an AI system,areas of risk and performance.Many of thes

191、e tools have been developed with academia and governments to encourage transparency in model reporting and drive assurance that the AI systems are aligned with existing and emerging regulations and standards.These assessment tools can help enterprises create consistent and well-defined practices tha

192、t prioritize attributes such as accuracy,bias,consistency,transparency,interpretability and fairness in the AI model being developed.Managing governance,risk and compliance:questions to askTABLE 8SpecificationsKey considerations5.1 What is the target demographic for data collection for the model?As

193、AI modelling depends on data,it is imperative to consider the impact of data use on the target,especially in terms of privacy,consent and regulations related to individuals.Is the data being collected from vulnerable demographic groups of the target countries?If personal data is being used,is it bei

194、ng collected in compliance with the data protection and privacy laws of the country(GDPR,Health Insurance Portability and Accountability Act(HIPAA),regional laws etc.)?Does the supplier have the informed consent of the individuals whose data has been collected,i.e.have the individuals been provided

195、all the necessary information?Can individuals withdraw their consent to the data collected?If so,will the collected data be withdrawn from the AI model?What are the relevant PII categories for the data collection process?Is the content moderated?(e.g.for sexuality/violence)5.2 How has the supplier a

196、ccounted for managing cybersecurity risks?An important concern to be addressed early on is cybersecurity failures,given its potential to have a serious impact on the output of the AI solution.What are the proactive measures the supplier has taken to detect and tackle cyberattacks?How does the suppli

197、er minimize the effect of an attack?Does the supplier actively perform vulnerability management to address common and frequent threats?Adopting AI Responsibly:Guidelines for Procurement25SpecificationsKey considerations5.3 Has the supplier reviewed the potential geopolitical risks that arise from op

198、erating in different physical locations?For a company looking to embed a disruptive technology like AI into its systems,the impact of geopolitics must be considered.Has the supplier accounted for the geopolitical risks associated with collecting data from certain disputed territories?Have the risks

199、of storing or processing data in unstable regions been considered?Will AI pose any risk if used in such territories?e.g.can it heighten instability in regional politics,affect peace and security?Are data collectors at any physical risk during the process of data collection?5.4 Have the risks related

200、 to the project been defined clearly?Due to the uncertainty of AI/ML work,an experienced AI modelling supplier is expected to identify possible risks and ways to mitigate them.Is the scope for the AI defined clearly in terms of deliverables/outcomes to be achieved?How does the supplier manage unsupp

201、orted content types?How does the supplier define hard performance metrics with AI?To what extent is the AI solution reproducible?Will the AI model be covered by intellectual property policy?Who has legal ownership of source data,models and resell rights?5.5 Is the supplier compliant with rules and r

202、egulations associated with building an AI model?AI offers great value to businesses,but it comes with a strategic risk for all stakeholders.Governments and institutions are actively taking measures to prevent the misuse of the technology and to build trust in AI tools.*The AI explainability statemen

203、t is a public document released by an AI organization that outlines how its AI algorithms work,its intended use,technology infrastructure,model accuracy,bias detection and mitigation,system maintenance,risk management,ethical principles,and data sources.*The EU AI Act establishes a horizontal set of

204、 rules for developing and using AI-driven products,services and systems within the EU using a risk-based approach of evaluating each AI system.The NYC Law on Automated Employment Decision Tools states that an automated hiring system used on or after 1 January 2023 must undergo a bias audit consistin

205、g of evaluation by an independent auditor,including testing to assess disparate impact to certain groups.Has the supplier proactively prepared to ensure compliance with regulations?Does the supplier provide an explainability statement*outlining the critical dimensions of the AI solution?Does the sup

206、plier comply with data-related regulations such as General Data Protection Regulation(GDPR),California Consumer Privacy Act(CCPA),Health Insurance Portability and Accountability Act(HIPAA),Childrens Online Privacy Protection Act(COPPA)to name a few?Is the model compatible with emerging algorithmic c

207、ompliance regulations,e.g.the EU AI Act5 and the NYC Law on Automated Employment Decision Tools6(effective July 5,2023)*?5.6 How does the supplier prepare for audits and compliance requirements?A risk management process that captures the policies,processes,procedures and practices across the organiz

208、ation involved in the development,testing,deployment,use and auditing of AI systems should be in place.It must be implemented effectively as well as be transparent.Does the supplier conduct mandatory conformity assessments?At what frequency?Has the supplier clearly defined the systems in place for i

209、nternal audits?What are the audit artifacts that it can share with you?How does the supplier ensure that compliance is met on its side as well as the buyers side after implementation of the AI model?If access to legal support is limited(in case of a smaller buyer),how can the supplier assist you in

210、ensuring compliance?Has the AI model been assessed for its performance with algorithm assessment tools,model cards etc.,to prevent biases and undesirable outputs?5.7 Is the supplier implementing international standards and certifications in the model?Standards and certifications can form a part of t

211、he initial guiding principles of governance to help AI developers and its users in their journey to build a responsible AI model.Does the supplier follow any AI governance standards set by international standard bodies(such as the International Organization for Standardization(ISO)and the Institute

212、of Electrical and Electronics Engineers Standards Association(IEEE)and others)to ensure that best practices are being followed?Will the AI system be accredited by any recognized institute that provides a calibrated conformity assessment of the model?Managing governance,risk and compliance:questions

213、to ask continuedTABLE 8Adopting AI Responsibly:Guidelines for Procurement26Managing governance,risk and compliance:questions to ask continuedTABLE 8SpecificationsKey considerations5.8 What are the organizational practices recommended by the supplier?Organizations should establish practices that can

214、characterize the AI model specifications paying special attention to attributes such as accuracy,bias,consistency,transparency,interpretability and fairness.Is the risk-based approach developed by the supplier based on the AI model and the industry in which it is being implemented?Has the supplier c

215、onducted an AI impact assessment of the buyer organization at an early stage of the procurement process?5.9 Do the contractual agreements include all compliance-related factors?A good agreement will not only involve basic contractual terms and project management details but also have protective meas

216、ures against non-compliance.*RAII is a non-profit organization offering certification programmes that align with AI laws,regulations,principles,research and practitioner insights.Can the supplier develop capacity in the form of new contract requirements?Is there a supplier compliance statement(e.g.R

217、esponsible Artificial Intelligence Institute(RAII)certification*)that organizations can include in their master service agreements?Are there contractual agreements surrounding restricted use or prohibited forms of use?Has the buyer developed well-defined KPIs and compliance metrics to track performa

218、nce during the AI life cycle?Does the supplier offer support beyond contractual agreements to assist in governance,maintenance and change management?Adopting AI Responsibly:Guidelines for Procurement27ConclusionThe history of AI in enterprises dates back to the early days of computing when rule-base

219、d systems and expert systems were used to perform specific tasks,such as analysing data and making predictions.However,these systems were limited in their capabilities and required significant amounts of human input to function effectively.The growth of AI has been exponential in the past decade,and

220、 AI technology is now achieving goals that seemed very distant just a few years ago.Organizations worldwide are increasingly adopting AI and ML to support business growth,improve efficiency and gain a competitive edge.Businesses are investing in AI-powered systems to automate repetitive and time-con

221、suming tasks,optimize pricing,identify new opportunities and make investment decisions.In many cases,these AI solutions are deeply embedded into the organizations enterprise applications,making recommendations and predictions and influencing critical decision-making.That said,its crucial to approach

222、 the procurement process in a comprehensive and collaborative manner.The procurement team needs to maintain a high degree of control to ensure that the AI tool delivers ethical,responsible and reliable solutions,instilling confidence.The guidance on industry-standard practices and ways to minimize o

223、rganizational risks while adopting AI technologies is very limited,and theres a pressing need for a responsible AI procurement toolkit.This toolkit offers a holistic procurement process beyond the technical evaluation of capabilities of the solution providers that can be tailored as per the needs of

224、 the organization and the industry it operates in.It enables a clear understanding of the business strategies that are driving the need to procure an AI solution;the commercial implications of the decision;quantity,quality and security of the data involved;the governance necessary for accountability

225、 and risk management;and the impact on the organizations ethics and sustainability policies.Further,it explores the role of various internal stakeholders from procurement to product to legal to IT and others in addressing the five key considerations in this toolkit,which will steer an organizations

226、procurement team in the right direction.Overall,this AI procurement toolkit provides comprehensive,non-prescriptive and industry-agnostic guidelines that will establish a procurement process to facilitate the selection of a suitable,ethical AI solution.Adopting AI Responsibly:Guidelines for Procurem

227、ent28AppendixWhen a company ventures into the market to explore AI offerings for the first time,it is beneficial to begin with a few preliminary questions for all potential suppliers.These questions,meant for preliminary understanding and assessment,will help the organization understand the applicab

228、ility and fitment of the suppliers and their solutions to requirements.Typically,commercial strategy questions are not included in this evaluation stage as they are addressed in subsequent steps of the procurement process after the scope of requirement is clearly defined,along with a narrowed-down l

229、ist of comparable AI solutions.This exercise will help streamline and focus the effort required in the detailed assessment.The following list of questions will help buyers in their initial assessment.To this,they can add more solution domain/industry-specific questions to drive a better understandin

230、g of the potential solutions.A1 Preliminary understanding and assessmentThe first step:questions to askTABLE 9SpecificationsPreliminary considerations 1.Business strategy What kind of assurance or warranty does the supplier offer regarding the process and business outcomes?How is the proposed AI/ML

231、solution an optimal approach to meet requirements?Can the supplier provide case studies to support their response?Does the solution measure business results and user satisfaction?If so,how are these insights used for product modifications or upgrades?Does the supplier provide key performance indicat

232、ors(KPIs)for monitoring performance drifts that may prompt retraining of the model?2.Data strategy What data quality assurance processes and frameworks(storage,management,transfer etc.)does the supplier follow?In the case of external data ingestion,how does the supplier ensure complete,relevant,unbi

233、ased and timely data?Who is responsible for ensuring the quality,usability and reliability of third-party data?3.Ethics and sustainability What methods are used to train the AI model?Do the training methods uphold the principles of ethics(fairness,interpretability,privacy,security etc.)?Can the supp

234、lier identify possible sources of bias?What are the checks in place within the model to prevent biases from creeping in?Does your solution address new and emerging ethical risks such as misinformation,over-reliance and loss of skills?4.Governance,risk and compliance Does the supplier comply with dat

235、a-related regulations(e.g.General Data Protection Regulation(GDPR),California Consumer Privacy Act(CCPA),Health Insurance Portability and Accountability Act(HIPAA),Childrens Online Privacy Protection Act(COPPA)etc.)?Does the supplier follow any AI governance standards and best practices set by inter

236、national standards organizations(such as the International Organization for Standardization(ISO)and the Institute of Electrical and Electronics Engineers Standards Association(IEEE)?Adopting AI Responsibly:Guidelines for Procurement29ContributorsAcknowledgementsThe report was co-created with active

237、guidance and support from experts and diverse stakeholders in the World Economic Forum project community on Shaping the Future of Technology Governance,who shared insights and lessons learned through interviews,workshops and consultation sessions.The opinions expressed herein do not necessarily refl

238、ect the views of the individuals or organizations involved in the project or listed below.Sincere thanks are extended to those who contributed their insights via interviews and workshops,as well as those not captured below.World Economic Forum Benjamin Larsen Lead,Artificial Intelligence and Machine

239、 LearningGEPAditya Desai Director,ConsultingJoanne Jacob Senior Consultant,ConsultingSitharthan K Senior Consultant,ConsultingMudit Kumar Vice President,ConsultingRaina Saxena Senior Director,ConsultingHemanty Tudu Senior Consultant,ConsultingSincere appreciation is extended to the following working

240、 group members,who spent numerous hours providing critical input and feedback to the drafts.Their diverse insights are fundamental to the success of this work.Uthman Ali Digital Ethics Lead and Senior Product Analyst,bpChloe Autio Director,Policy,The Cantellus Group Ivar Beljaars Senior Vice-Preside

241、nt,Sales,CUJO AIMegha Chawla Manager,Open Innovation and Tech Evangelist,HCLTechAmy DeCicco Vice-President,Marketing,retrain.ai Sachin Dev Duggal Chief Wizard,Builder.aiDean Horton Digital Security Risk Manager,Procurement,bpAustin Imperato Chief of Staff,Government and Regulatory Affairs,IBMIgor Ja

242、blokov Chief Executive Officer,PryonMukund Kalmanker Vice-President and Global Head,AI Solutions,WiproVijay Karunamurthy Field Chief Technology Officer,Scale AI Paige Kassalen Head,Customer Operations,CrowdAIAbhinav Khare Head,Tech Venturing and Open Innovation Ecosystem,HCLTech Simon Greenman Partn

243、er,Best Practice AI Abhishek Gupta Senior Responsible AI Leader and Expert,Boston Consulting Group(BCG)Meirav Oren Chief Executive Officer and Co-Founder,VersatileClaudia Perry Digital Graduate,Digital Science,bpDevaki Raj Chief Executive Officer,CrowdAINikita Sejpal-Shah Cloud and IoT Lead,Procurem

244、ent,bp Adopting AI Responsibly:Guidelines for Procurement30ProductionStudio MikoLaurence Denmark Creative DirectorMartha Howlett EditorOliver Turner DesignerKaren Silverman Chief Executive Officer and Founder,The Cantellus Group Navrina Singh Founder and Chief Executive Officer,Credo AI Jascha Stein

245、 Co-Founder and Chief Executive Officer,OmniBot.aiAndy Smith Principal Security Architect,bpLaura Taddei Director,Business Affairs,BenevolentAIErik Vogt Vice-President,Enterprise Solutions,AppenAdopting AI Responsibly:Guidelines for Procurement31Endnotes1.Precedence Research,Artificial Intelligence(

246、AI)Market,2023,https:/ Out of the Data Swamp With a Governed Data Lake”,Dataeaze,2019,https:/www.dataeaze.io/get-out-of-the-data-swamp-with-a-governed-data-lake/.3.Gonfalonieri,Alexandre,“Do You Need Synthetic Data For Your AI Project?”,Towards Data Science,21 October 2019,https:/ Cost of a Data Bre

247、ach in the United States From 2006 to 2022 Graph,https:/ Commission,Proposal for a REGULATION OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL LAYING DOWN HARMONISED RULES ON ARTIFICIAL INTELLIGENCE(ARTIFICIAL INTELLIGENCE ACT)AND AMENDING CERTAIN UNION LEGISLATIVE ACTS,2021,https:/eur-lex.europa.eu/le

248、gal-content/EN/TXT/?uri=celex%3A52021PC0206.6.New York City Department of Consumer and Worker Protection,Notice of Adoption of Final Rule,2023,https:/rules.cityofnewyork.us/rule/automated-employment-decision-tools-updated/.Adopting AI Responsibly:Guidelines for Procurement32World Economic Forum9193

249、route de la CapiteCH-1223 Cologny/GenevaSwitzerland Tel.:+41(0)22 869 1212Fax:+41(0)22 786 2744contactweforum.orgwww.weforum.orgThe World Economic Forum,committed to improving the state of the world,is the International Organization for Public-Private Cooperation.The Forum engages the foremost political,business and other leaders of society to shape global,regional and industry agendas.