McKinsey：2022年人工智能现状：五年回顾.pdf

The state of AI in 2022and a half decade in reviewDecember 2022The results of this years McKinsey Global Survey on AI show the expansion of the technologys use since we began tracking it five years ago,but with a nuanced picture underneath.1 Adoption has more than doubled since 2017,though the pro-portion of organizations using AI has plateaued between 50 and 60 percent for the past few years.A set of companies seeing the highest financial returns from AI continue to pull ahead of competitors.The results show these leaders making larger investments in AI,engaging in increasingly advanced practices known to enable scale and faster AI development,and showing signs of faring better in the tight market for AI talent.On talent,for the first time,we looked closely at AI hiring and upskilling.The data show that there is significant room to improve diversity on AI teams,and,consistent with other studies,diverse teams correlate with outstanding performance.This marks the fifth consecutive year weve conducted research globally on AIs role in business,and we have seen shifts over this period.First,AI adoption has more than doubled.In 2017,20 percent of respondents reported adopting AI in at least one business area,whereas today,that figure stands at 50 percent,though it peaked higher in 2019 at 58 percent.Meanwhile,the average number of AI capabilities that organizations use,such as natural-language generation and computer vision,has also doubledfrom 1.9 in 2018 to 3.8 in 2022.Among these 1 In the survey,we defined AI as the ability of a machine to perform cognitive functions that we associate with human minds(for example,natural-language understanding and generation)and to perform physical tasks using cognitive functions(for example,physical robotics,autonomous driving,and manufacturing work).2 In 2017,the definition for AI adoption was using AI in a core part of the organizations business or at scale.In 2018 and 2019,the definition was embedding at least one AI capability in business processes or products.In 2020,2021,and 2022,the definition was that the organization has adopted AI in at least one function.Five years in review:AI adoption,impact,and spend2The state of AI in 2022and a half decade in reviewResponses show an increasing number of AI capabilities embedded in organizations over the past ve years.Average number of AI capabilities that respondents organizations have embedded within at least one function or business unit Share of respondents who say their organizationshave adopted AI in at least one function,%of respondents who say given AI capability is embedded in products or business processes in at least one function or business unitMcKinsey&CompanyThe number of capabilities included in the survey has grown over time,from 9 in 2018 to 15 in the 2022 survey.Question was asked only of respondents who said their organizations have adopted AI in at least one function.TransformersGenerative adversarial networks(GAN)Transfer learningNatural-language generationFacial recognitionReinforcement learningPhysical roboticsNatural-language speech understandingDigital twinsRecommender systemsKnowledge graphsDeep learningVirtual agents or conversational interfaces Natural-language text understandingComputer visionRobotic process automation39343333302525242320201818161111201720182019202020212022204758505650201820192020202120221.92.33.13.93.8capabilities,robotic process automation and computer vision have remained the most commonly deployed each year,while natural-language text understanding has advanced from the middle of the pack in 2018 to the front of the list just behind computer vision.3The state of AI in 2022and a half decade in reviewThe most popular AI use cases span a range of functional activities.Most commonly adopted AI use cases,by function,%of respondentsTop use casesUse cases by functionMcKinsey&CompanyOut of 39 use cases.Question was asked only of respondents who said their organizations have adopted AI in at least one function.Service operations optimizationCreation of new AI-based productsCustomer service analyticsCustomer segmentationNew AI-based enhancements of productsCustomer acquisition and lead generationContact-center automation Product feature optimizationRisk modeling and analyticsPredictive service and intervention24201919191716161514Service operationsProduct and/or service developmentMarketing and salesRiskEg,eld services,customer care,back o ce.The top use cases,however,have remained relatively stable:optimization of service operations has taken the top spot each of the past four years.Second,the level of investment in AI has increased alongside its rising adoption.For example,five years ago,40 percent of respondents at organizations using AI reported more than 5 percent of their digital budgets went to AI,whereas now more than half of respondents report that level of investment.Going forward,63 percent of respondents say they expect their organizations investment to increase over the next three years.4The state of AI in 2022and a half-decade in reviewThird,the specific areas in which companies see value from AI have evolved.In 2018,manufacturing and risk were the two functions in which the largest shares of respondents reported seeing value from AI use.Today,the biggest reported revenue effects are found in marketing and sales,product and service development,and strategy and corporate finance,and respondents report the highest cost benefits from AI in supply chain management.The bottom-line value realized from AI remains strong and largely consistent.About a quarter of respondents report this year that at least 5 percent of their organizations EBIT was attributable to AI in 2021,in line with findings from the previous two years,when weve also tracked this metric.Lastly,one thing that has remained concerningly consistent is the level of risk mitigation organizations engage in to bolster digital trust.While AI use has increased,there have been no substantial increases in reported mitigation of any AI-related risks from 2019when we first began capturing this datato now.The most popular AI use cases span a range of functional activities.Top use casesUse cases by functionMcKinsey&Company Question was asked only of respondents who said their organizations have adopted AI in at least one function.Eg,eld services,customer care,back o ce.Most commonly adopted AI use cases within each business function,%of respondentsService operations optimization24Contact-center automation16Service operationsCustomer service analytics19Customer segmentation19Marketing and salesSales and demandforecasting10Logistics network optimization9Supply chain managementPredictive maintenance13Simulations(eg,using digital twins,3 D modeling)11Yield,energy,and/or throughput optimization11ManufacturingOptimization of talent management10Optimization of workforce deployment5Human resources Risk modeling and analytics15Fraud and debt analytics11RiskCapital allocation7M&A support4Treasury management4Strategy and corporate fnanceCreation of new AI-based products20New AI-based enhancementsof products19Product and/or service development5The state of AI in 2022and a half decade in reviewCost decrease and revenue increase from AI adoption in 2021,by function,%of respondents1Question was asked only of respondents who said their organizations have adopted AI in a given function.Respondents who said“no change,”“cost increase,”“not applicable,”or“dont know”are not shown.AI-related cost decreases are most often reported in supply chain management and revenue increases in product development and marketing and sales.Service operationsDecreaseby 10%Increaseby 610%Increaseby 5%ManufacturingHuman resourcesMarketing and salesRiskSupply chain managementProduct and/or service developmentStrategy and corporate fnanceAverage across all activites45422928435230433229322521304120312310734874866313546435761587048597065631010149101413881018132011172416193733314127283341366The state of AI in 2022and a half decade in reviewOver the past half decade,during which weve been conducting our global survey,we have seen the “AI winter”turn into an“AI spring.”However,after a period of initial exuberance,we appear to have reached a plateau,a course weve observed with other technologies in their early years of adoption.We might be seeing the reality sinking in at some organizations of the level of organiza-tional change it takes to successfully embed this technology.In our work,weve encountered companies that get discouraged because they went into AI thinking it would be a quick exercise,while those taking a longer view have made steady prog-ress by transforming themselves into learning organizations that build their AI muscles over time.These companies gradually incorporate more AI capabilities and stand up increasingly more applications progressively faster and more easily thanks to lessons from past successes as well as failures.They not only invest more,but they also invest more wisely,with the goal of creating a veritable AI factory that enables them to incorporate more AI in more areas of the business,first in adjacent ones where some existing capabilities can be repurposed and then into entirely new ones.There is,at a high level,an emerging playbook for getting maximum value from AI.Each year that we conduct our research,we see a group of leaders engaging in the types of practices that help execute AI successfully.Its paying off in the form of actual bottom-line impact at significant levels.We also see it every day as we guide others on their AI journeys.Its not easy work,but as has been the case with previous technologies,the gains will go to those who stay the course.McKinsey commentaryMichael Chui Partner,McKinsey Global InstituteThose taking a longer view have made steady progress by transforming themselves into learning organizations that build their AI muscles over time.7The state of AI in 2022and a half decade in reviewAI use and sustainability effortsThe survey findings suggest that many organizations that have adopted AI are integrating AI capabilities into their sustainability efforts and are also actively seeking ways to reduce the environmental impact of their AI use(exhibit).Of respondents from organizations that have adopted AI,43 percent say their organizations are using AI to assist in sustainability efforts,and 40 per-cent say their organizations are working to reduce the environmental impact of their AI use by minimizing the energy used to train and run AI models.As companies that have invested more in AI and have more mature AI efforts than others,high performers are 1.4 times more likely than others to report AI-enabled sustain-ability efforts as well as to say their organizations are working to decrease AI-related emissions.Both efforts Types of sustainability e orts in which respondents organizations are using AIImproving the organizations environmental impact(eg,improving energy e ciency,optimizing transportation)Evaluating sustainability e orts(eg,benchmarking)Improving the organizations governance(eg,regulatory compliance,risk management)Improving the organizations social impact(eg,sourcing ethical products)62514534Only asked of respondents whose organizations have adopted AI in at least one function who said that their organizations are using AI in sustainability e orts;n=302.Includes respondents in India,Latin America,Middle East,North Africa,and sub-Saharan Africa.Includes respondents in Hong Kong SAR and Taiwan China.Only asked of respondents whose organizations have adopted AI in at least one function.For organizations based in Greater China,n=102;for AsiaPaci c,n=74;for developing markets,n=118;for Europe,n=260;and for North America,n=190.Greater China61AsiaPaci c54Developing markets44Europe39North America30Organizations using AI in their sustainabilityeforts,%of respondents Greater China46AsiaPaci c47Developing markets53Europe36North America31Organizations taking steps to reduce carbonemissions from their AI use,%of respondents Organizations are using AI within sustainability e orts and are working to reduce the environmental impact of their AI use.Exhibit are more commonly seen at organizations based in Greater China,AsiaPacific,and developing markets,while respondents in North America are least likely to report them.When asked about the types of sustainability efforts using AI,respondents most often mention initiatives to improve environmental impact,such as optimiza-tion of energy efficiency or waste reduction.AI use is least common in efforts to improve organizations social impact(for example,sourcing of ethically made products),though respondents working for North American organizations are more likely than their peers to report that use.8The state of AI in 2022and a half decade in reviewMind the gap:AI leaders pulling aheadOver the past five years,we have tracked the leaders in AIwe refer to them as AI high performersand examined what they do differently.We see more indications that these leaders are expanding their competitive advantage than we find evidence that others are catching up.First,we havent seen an expansion in the size of the leader group.For the past three years,we have defined AI high performers as those organizations that respondents say are seeing the biggest bottom-line impact from AI adoptionthat is,20 percent or more of EBIT from AI use.The proportion of respondents falling into that group has remained steady at about 8 percent.The findings indicate that this group is achieving its superior results mainly from AI boosting top-line gains,as theyre more likely to report that AI is driving revenues rather than reducing costs,though they do report AI decreasing costs as well.Next,high performers are more likely than others to follow core practices that unlock value,such as linking their AI strategy to business outcomes.Also important,they are engaging more often in“frontier”practices that enable AI development and deployment at scale,or what some call the“industrialization of AI.”For example,leaders are more likely to have a data architecture that is modular enough to accommodate new AI applications rapidly.They also often automate most data-related processes,which can both improve efficiency in AI development and expand the number of applications they can develop by providing more high-quality data to feed into AI algorithms.And AI high performers are 1.6 times more likely than other organizations to engage nontechnical employees in creating AI applications by using emerging low-code or no-code programs,which allow companies to speed up the creation of AI applications.In the past year,high performers have become even more likely than other organizations to follow certain advanced scaling practices,such as using standardized tool sets to create production-ready data pipelines and using an end-to-end platform for AI-related data science,data engineering,and application development that theyve developed in-house.Hig