《波士顿咨询:2023利用人工智能加速气候行动报告(英文版)(52页).pdf》由会员分享,可在线阅读,更多相关《波士顿咨询:2023利用人工智能加速气候行动报告(英文版)(52页).pdf(52页珍藏版)》请在三个皮匠报告上搜索。
1、Accelerating Climate Action with AINovember 2023By Amane Dannouni,Stefan A.Deutscher,Ghita Dezzaz,Adam Elman,Antonia Gawel,Marsden Hanna,Andrew Hyland,Amjad Kharij,Hamid Maher,David Patterson,Edmond Rhys Jones,Juliet Rothenberg,Hamza Tber,Maud Texier,and Ali ZiatBoston Consulting Group partners with
2、 leaders in business and society to tackle their most important challenges and capture their greatest opportunities.BCG was the pioneer in business strategy when it was founded in 1963.Today,we work closely with clients to embrace a transformational approach aimed at benefiting all stakeholdersempow
3、ering organizations to grow,build sustainable competitive advantage,and drive positive societal impact.Our diverse,global teams bring deep industry and functional expertise and a range of perspectives that question the status quo and spark change.BCG delivers solutions through leading-edge managemen
4、t consulting,technology and design,and corporate and digital ventures.We work in a uniquely collaborative model across the firm and throughout all levels of the client organization,fueled by the goal of helping our clients thrive and enabling them to make the world a better place.Contents01 Foreword
5、02 Executive Summary05 The Climate Action Imperative and the Promise of AI09 How AI Can Help Accelerate Climate Action22 Navigating AIs Potential Risks28 AI for Climate:A Summary of Critical Policy Outcomes41 Endnotes43 About the Authors45 Acknowledgements47 ReferencesContents1 ACCELERATING CLIMATE
6、ACTION WITH AIForewordMore specifically,its goals are to highlight AIs significant potential to help address our environmental challenges,to shed light on climate-relevant AI risks,and to offer policymakers a streamlined framework for desirable policy outcomes.Throughout the report,we share examples
7、 of successful early applications of AI for climate and of instances in which policymakers have already taken the initiative to enable,promote,or guide the use of AI for climate action across sectors.This work draws on interviews with a range of climate change and AI experts,builds on previous resea
8、rch from organizations including Climate Change AI and the AI for the Planet Alliance,and leverages BCGs analysis and client experience as well as Googles technical and operational expertiseand its experience in developing solutions.This report aims to provide policymakers,corporate decision makers,
9、and climate leaders with a clear and concise understanding of the role that artificial intelligence(AI)can play in climate action.BOSTON CONSULTING GROUP GOOGLE 2Accelerating climate action is imperative,as we are on a path to fall short of the Paris Agreements goal to keep warming under 1.5 Celsius
10、.The United Nations Intergovernmental Panel on Climate Change(IPCC)estimates that,based on action to date,the world will likely see warming of 2.8C with cata-strophic consequences.The IPCC forecasts that in order to meet the 1.5C goal,the world will need to reduce emissionsfrom the base-line of 2010
11、 levelsby 43%by 2030.By scaling currently proven applications and technology,artificial intelligence(AI)has the potential to unlock insights that could help mitigate 5%to 10%of global greenhouse gas(GHG)emissions by 2030and significantly bolster climate-related adaptation and resilience initiatives.
12、87%of executives view AI as having the potential to address climate issues.AIs positive impact will be multiplied should it contrib-ute to scientific breakthroughs that open new pathways for climate action.AI can contribute to climate action by reducing emissions,guiding adaptations to unavoidable c
13、limate change impacts,and providing foundational capabilities that enable climate action.Mitigation.Helping with both the reduction and remov-al of emissionsand with the underlying measurement needed to size the challenge and track progress Adaptation and Resilience.Aiding countries,regions,cities,c
14、itizens,and businesses in forecasting climate-related hazards,developing plans to address them,and responding in real time to crises Foundational Capabilities.Enabling climate-related modeling,research into climate economics,and new approaches to climate education and supporting break-throughs in fu
15、ndamental researchWhile AI is only just starting to be applied to climate challenges,leading-edge organizations and use cases are already delivering resultsand demonstrating the promise of AI for climatealong three dimensions.Information.AI-curated information sources are aiding nations in shaping t
16、heir climate strategyand in re-sponding to emergencies such as wildfires.Prediction.AIs predictive power is helping save lives by offering advance warning of floods.Optimization.AI applications are enabling organiza-tions to understand and reduce their Scope 1,2,and 3 carbon footprints.1AI also pose
17、s risks that must be considered and managed thoughtfully to ensure its use has a net positive impact on climate.Energy-Related GHG Emissions.A 2022 paper in Nature Climate Change estimates that cloud and hyperscale data centers are responsible for 0.1%0.2%of global GHG emissions and that roughly 25%
18、of data center workloads are related to machine learning(ML).Yet,newer and more complex AI models may require more energy.At present,robust forecasts for AIs future energy requirements remain elusive given uncertain adoption rates and the broad spectrum of potential technical advancements with the p
19、otential to decrease AIs energy intensity.Nonetheless,AI providers are already striving to enhance energy efficiency and integrate clean energy sources.Water Use.Water-based cooling remains the most energy-efficient option for data centers,and its overall impact on water consumption is low.In 2016 i
20、n the US,data centers were estimated to have used less than 0.02%of the countrys water consumption for cooling.Nevertheless,in some cases,water-based cooling can put pressure on local water resources.Data center oper-ators have begun to address this issue by providing more disclosure,exploring new c
21、ooling techniques,and invest-ing in replenishment initiatives.Executive Summary3 ACCELERATING CLIMATE ACTION WITH AI Waste.While data centers currently account for only a small fraction of the worlds e-waste challenge,there is an opportunity for tech firms to build on early circularity successes and
22、 take a more thoughtful approach embrac-ing more recycling and reuse.Other Potential Risks.AI applications should be sustainable and equitable by intention.AI can be applied to both climate-friendly and climate-unfriendly appli-cations,can narrow or widen disparities between the Global North and the
23、 Global South,and can be trained on data sets that reflect the worlds diversity.Leaders and model builders need to be mindful in their design choices.Policymakers have a critical oversight role to play in maximizing the benefits from AI-driven climate action while minimizing its risks.Critical polic
24、y outcomes to pursue include the following:Enabling AI for climate progress by encouraging data sharing,ensuring affordable technology access,building awareness,and investing in talent Accelerating the deployment of AI for climate by defining public and private sector priorities,delivering on public
25、 sector use cases,and encouraging private sector action Promoting environmentally and socially responsible deployment of AIBOSTON CONSULTING GROUP GOOGLE 4According to the Massachusetts Institute of Technology,AI is defined as the ability of computers to imitate human cognitive functions such as lea
26、rning and problem-solving,using math and logic to simulate the process of reasoning that helps humans learn from new information and make decisions.For the purposes of this report,we are using a broader definition of AI that comprises a set of mathematical and computer science techniques aimed at an
27、alyzing data to help understand and navigate real-world phenomena through:providing better information(descriptive use cases),delivering improved predictions(predictive use cases),and suggesting optimization moves and recommendations to reach targets(prescriptive use cases).These goals can be attain
28、ed by applying wide range of techniques including those in the table below-all of which we include in this reports definition of AI.Applying AI to real-world problems is common practice today.The technology has proven its ability to help public and private organizations have a better understanding o
29、f their context,provide better services,and improve their operational performance.How We Define Artificial IntelligenceTechnologyGeneral ExampleClimate-Related ExampleAdvanced AnalyticsThe use of advanced mathematical and statistical techniques to develop insights from structured and unstructured da
30、ta.Supermarket Inventory Management.Advanced analytics can identify best sellers and demand dynamics,enabling more efficient shelving and restocking strategies,thereby reducing waste and ensuring popular items are always in stock.Energy Consumption Optimization.Advanced analytics can optimize a buil
31、dings carbon footprint by adjusting heating,cooling,and lighting systems in response to real-time data from sensors and weather forecasts.Machine LearningTraining computers to learn and make predictions from data.Historical data constitutes the inputs,while predictions based on new or unseen data ar
32、e the outputs.Credit Card Fraud Detection.Machine learning helps banks and credit card companies detect unusual transactions,enabling them to alert card holders and minimize fraud losses.Predicting Wildfires.Machine learning models can analyze weather data,satellite imagery,and terrain information t
33、o predict the likelihood of wildfires,helping authorities take preventive measures and optimize resource allocation.Deep LearningA specialized form of machine learning that uses artificial neural networks to generate hierarchical insights from diverse data sets,such as images,text,or audio.These mod
34、els are able to recognize patterns or features within the data,for example,by identifying objects in images.Medical Image Evaluation.Applied to the analysis of medical images such as X-rays and MRIs,deep learning helps doctors diagnose diseases and other abnormalities more accurately,enabling more t
35、imely and effective treatments.Extreme Weather Prediction.Deep learning can analyze vast amounts of historical and real-time meteorological and satellite data,leading to more accurate forecasts for hurricanes,tornadoes,and typhoons.Large Language Models Advanced AI models trained on vast amounts of
36、text dataand able to generate human-like text as output,such as for Generative AI use cases.Customer Service Chatbots.Large language models enable companies to automate the process of answering customer questions and helping troubleshoot issues,enhancing the efficiency of,and satisfaction with,custo
37、mer service operations.Green Technology Innovation.Large language models can accelerate innovation by digesting research papers and patent applications and rapidly surfacing ideas and identifying knowledge gaps.5 ACCELERATING CLIMATE ACTION WITH AIDespite significant progress over the last several y
38、ears in mobilizing the global community to intensify its climate actions and ambitions,the world is not on track to meet the Paris Agreements target to limit tem-perature rise to 1.5C.This target was selected because scientists believe that above that level,the effects would be catastrophic and pote
39、ntially irreversible.At presentbased on updated national pledges since COP26 in 2021the United Nations Environment Programme currently estimates that we are on a path to warming by 2.8C.2Even if the world succeeds in limiting warming to 1.5C,there will still be adverse impacts.Already today at 1.1C,
40、the IPCC reports that over 3 billion people live in areas highly vulnerable to climate impacts.We are already seeing the impact on weather,agriculture,water security,and migration.If we overshoot the target,the picture becomes increasingly dire:seas will rise further,droughts will be worse,and extre
41、me weather events will be more common.The Climate Action Imperative and the Promise of AIBOSTON CONSULTING GROUP GOOGLE 6In a 1.5C world,the IPCC forecasts that 48%of the worlds population will be exposed to deadly heat levels for more than 20 days a year.In a 3 to 4C world,that number increases to
42、74%.If we stay on our current trajectory,the World Bank estimates an additional 143 million peoplemore than the combined populations of the United King-dom,Morocco,and Malaysiacould be displaced.3 And,absent significant investments in resilience,major global citiesfor example,Tokyo,Osaka,Mumbai,Bang
43、kok,New York,London,and Lagoswill find themselves partly under water.We urgently need new tools to accelerate the reduction and removal of GHG emissionsand to help citizens,cities,regions,countries,and businesses make plans to adapt to the inevitable impacts of warming.AI offers much promise.Climate
44、 Action Has an Analytical Challengeand AI Can HelpLeaders increasingly understand the urgency.So far,194 parties to the Paris Agreement have developed Nationally Determined Contributions(NDCs)each representing detailed commitments for how their country will help the world meet the Paris Agreements 1
45、.5C goalup from 75 parties in February 2021.4But avoiding the most catastrophic impacts of warming requires more than political will.To achieve real progress,we need to develop a much richer analytical understanding of a complex system comprising many variables and feed-back loops.(See Exhibit 1.)Ex
46、hibit 1-Climate is an interlinked,multi-parameter systemSource:Philippe Rekacewicz,Emmanuelle Bournay,UNEP/GRID-Arendal;BCG analysis.Human activities,such as fossil fuel burning and land use changes,create significant volumes of greenhouse gas.Emissions have varying impacts on core climate character
47、istics,and changes in these processes can worsenthe greenhousegas effect.Fluctuations in climate characteristics drive major impactsnatural,physical,and socioeconomicat both local and global scales.Land use changesDisasters(enhanced)Greenhouse effectFossil fuel burningGreenhouse gas emissionsGulf St
48、reammodificationHuman activitiesClimate change processesCore climate characteristicsMajor impactsUrbanizationDeforestationTransportHeating IndustryAgricultureIncrease in impermeable surfacesCH4N2OEurope coolingHeat wavesCyclonesDroughtsFoodLoss of traditional lifestylesCloudsSalinityWater tempera-tu
49、reCarbon cycle disturbanceIce cap meltingSea level riseDiseasespreadCasualtiesFaminesEconomic lossesBiodiversity lossesAbrupt climatechangeChanges in precipitationOcean circulation upheavalAverage temperature rise GlobalwarmingCO27 ACCELERATING CLIMATE ACTION WITH AIDeveloping models is essential to
50、 understanding the rela-tionships among variablesand to anticipating the likely impact of different strategies and choices.But modeling these complex interconnections on a local and global scale is a huge challenge.It requires assembling massive,longi-tudinal,and real-time global data sets.Informati
51、on is need-ed on climate(for example,temperatures,ocean process-es,and meteorological phenomena)and on human activities(for example,emissions,and land use changes).And not all the necessary data is even available.But understanding the complex systems that drive climate-relevant outcomes is exactly t
52、he kind of challenge at which AI excels.By amalgamating and processing massive data sets,AI can reveal elusive patterns and valuable insights,facilitate scenario development and prediction,accelerate the evaluation of multiple courses of action,enable operational optimizations,and help monitor progr
53、ess toward predefined goals.Business leaders agree.In a 2022 BCG survey of senior executives with leadership roles related to climate or AI(see AI is Essential for Solving the Climate Crisis),87%viewed AI as a helpful unlock for climate issues.They saw supporting emissions reduction as the top clima
54、te use case for AI in their organizations,but expressed interest in other applications as well.(See Exhibit 2.)Estimating AIs Potential ContributionBased on our research and experience,the three broad areas in which AI can accelerate climate progress are the following:Mitigation.Helping with both th
55、e reduction and remov-al of emissionsand with the underlying measurement needed to size the challenge and track progress Adaptation and Resilience.Aiding citizens,countries,regions,cities,and businesses to prepare for and respond to the inevitable impacts of a warming planet Foundational Capabilitie
56、s.Enabling climate action via improvements in climate modeling,climate eco-nomics,and climate education,as well as accelerating breakthrough innovations that will open new horizons for climate actionExhibit 2-Leaders believe AI can play a role in climate action,especially in helping to reduce emissi
57、onsSource:BCG Climate AI survey 2022.All respondents have decision-making authority over climate or AI topics at their organizations.Respondents were permitted to give more than one answer.87%of respondents say that AI is a helpful tool to address climate changeIn which areas of climate-related adva
58、nced analytics and AI do you see the greatest business value for your organization?(%)Reducing emissions Measuring emissionsPredicting hazardsManaging vulnerabilitiesRemoving emissionsMitigationAdaptation&Resilience61%57%44%42%37%28%Foundational CapabilitiesFacilitating climate research,climate econ
59、omics,and educationBOSTON CONSULTING GROUP GOOGLE 8Regarding emissions reduction potential,a 2021 BCG study(see Reduce Carbon and Costs with the Power of AI)estimates that currently proven AI-enabled use cases could reduce emissions by 5%to 10%by 2030.If that potential is fully realized,AI-driven ap
60、plications would be responsible for achieving roughly between 10%and 20%of the IPCCs 2030 interim emissions-reduction target for the world to achieve net zero by 2050.5 Similarly,a Microsoft/PwC study looking at four sectors(agriculture,energy,transport,and water)estimates that AI has the potential
61、to reduce global GHG emissions by 4%.6 Further,respondents in a Capgemi-ni survey of companies that had leveraged AI for climate action reported that their efforts to date had achieved GHG reductions of between 11.3%and 14.3%depending on the sectorand these executives believe that AI could reduce ov
62、erall GHG emissions by 15.9%in the next three to five years.7On adaptation and resilience,AI can help cities forecast their climate vulnerability,develop estimates of the cost of inaction,and model the impact of different climate inter-ventions.These insights can aid them in identifying the actions
63、with the greatest benefit,generating private-sector enthusiasm for funding investable projects,and securing public and philanthropic support for essential,but non-bankable,adaptations.It also can help guide real-time decision-making in agriculturefor example,increasing crop production through intell
64、igent irrigation systemsor in fast-moving crises such as wildfires.And AI offers many foundational capabilities that sup-port both short-term and long-term climate action.For example,it can support todays climate research with higher-fidelity climate change simulations.But it also has the potential
65、to accelerate breakthrough innovations in domains such as physics,chemistry,biology,and material science that could“bend the curve”on climate progress.All of our estimates are based on the current state of AI technologyand thus speak primarily to AIs potential in currently proven applications.Today,
66、we are in the early stages of the adoption curve.Transforming potential to achievement will require that all organizations fully em-brace AI as an essential enabler of their climate actions.And it is important to note that our assessment does not encompass major AI-driven disruptions and break-throu
67、ghsfor example,new materials for batteries,new drought-resistant crops,novel carbon removal technolo-gies,and scalable approaches to nuclear fusionthat could unlock massive positive impact.The promise of AI is real.While we are already seeing benefits,we need to accelerate its contribution to planet
68、-saving climate impact.The next chapter offers a deeper dive into the primary known climate-related use cases for AIand highlights some examples of how and where AI is already making a positive difference.9 ACCELERATING CLIMATE ACTION WITH AIAI has demonstrated the potential to enable and catalyze c
69、limate progress in three broad areas:taking emissions mitigation to the next level,shap-ing strategies for adaptation and resilience,and supporting both climate research and reinforcing technologies.Some AI applications are in early stages,some are being tested,and others are already being scaled.Bu
70、t all will need to be embraced more broadly if we are to fulfill the promise of AI to limit warming to less than 1.5C.Exhibit 3 summarizes the most promising of the currently known AI use cases for climate.The rest of this chapter will offer more detail on each,along the way highlighting inspir-ing
71、examples of how AI is helping unlock and accelerate climate progress.AIs Role in Emissions MitigationGetting smarter on reducing and removing emissions is essential.And AI is already delivering significant wins that need to be scaled.Its contributions fall into two broad areas:measurement and monito
72、ring,and reduction and removal.8Measurement and MonitoringWithout reliable,clean,and independently verifiable data,effective climate action is difficult.Countries and compa-nies need to know their baselines and track their progress,both at the macro level(“What are our total GHG emis-sions?”)and the
73、 micro level(“Which aspects of our opera-tions and broader supply chain are the big drivers?Are our efforts at reduction or removal delivering the expected results?”).How AI Can Help Accelerate Climate ActionBOSTON CONSULTING GROUP GOOGLE 10Effective measurement and monitoring solutions leverage AI
74、to process and analyze data from multiple sources such as satellite data,weather data,sensors,and other heavy data setswhich can,for example,help an organization develop a baseline for its Scope 1,2,and 3 emissions.AI can also deliver insights,revealing patterns in emissions and suggesting the best
75、ways to prioritize abatement efforts.In the domain of macro-level measurement,Climate TRACE has been an early mover.This nonprofit offers free emissions data for more than 80,000 individual sources and facilities around the globe,providing a data foundation to help organizations get started with mit
76、igation plans.Its data could,for example,assist countries seeking to transi-tion away from coal and other fossil-fuel based electricity generation by pinpointing the largest emitters and reveal-ing the mix of power sources by region.(See the sidebar Climate TRACE:Providing Timely,Independent Emissio
77、ns Datafor Free.)Solutions are emerging for micro-level measurement as well.Googles Environmental Insights Explorer(EIE)uses machine learning to offer city planners annual estimates of emissions from buildings and transportation,tree canopy status,and emissions reduction opportunities such as the po
78、tential for expanded rooftop solar.Houston,Texas,used EIE to perform a solar assessment and inform the development of its 5 million MWh Solar Energy Target Plan.Similarly,CO2 AI,a novel SaaS platform,enables business leaderstogether with their value chain partnersto develop an accurate estimate of t
79、heir organizations Scope 1,2,and 3 emissions down to the product level.It also helps them to model and evaluate emissions reduction opportunities.(See the sidebar CO2 AI:Helping Business Ecosystems Reduce their Carbon Footprints.)Exhibit 3-Key AI applications to accelerate climate progressSource:BCG
80、,AI for the Planet Alliance.MitigationFoundational CapabilitiesMeasurement&MonitoringMacro-level measurement e.g.,calculating carbon footprint at the country levelEnabling emissions reductione.g.,integrating renewable energy into smart grids,optimizing transportation of goodsBuilding early warning s
81、ystemse.g.,predicting near-term extreme events such as flooding,drought,and cyclonesResponding to crisese.g.,monitoring drought and wildfire spreadClimate modelinge.g.,monitoring drought and wildfire spreadClimate economicse.g.,developing cost of inaction assessmentsEducation&behavioral change e.g.,
82、developing recommendations for climate-friendly consumptionInnovation&breakthroughs e.g.,supporting research on fusionMicro-level measuremente.g.,calculating carbon footprints of individual productsSupporting nature-based&technological removale.g.,assessing natural carbon stocksProjecting long-term
83、trends e.g.,modeling localized sea-level rise and drought frequencyBuilding resilient infrastructure&protecting biodiversity e.g.,intelligent irrigation,monitoring of endangered speciesHazardPredictionVulnerabilityManagementReduction&RemovalAdaptation and Resilience11 ACCELERATING CLIMATE ACTION WIT
84、H AIMaking real progress on climate requires timely and accu-rate data on emissions to inform government policy and business action.But historically,emissions data has been based on self-reporting,calculated using varying algo-rithms,and submitted years after the fact.Climate TRACEa global coalition
85、 of nonprofits,tech startups,and researchersoffers a powerful,free,and independent alternative:the first comprehensive source-level global inventory of GHG emissions.Supported by Google.org,among others,Climate TRACE uses AI and machine learning to calculate GHG emissions on a global scale,with the
86、goal of moving toward real-time precision.To achieve this,its model analyzes more than 59 terabytes of data from over 300 satellites and more than 11,000 sensors to create highly granular emissions data for over 80,000 sources globally.That number is expected to grow to more than 70 million sources
87、by the end of 2023.Application area:Macro-Level MeasurementClimate TRACE:Providing Timely,Independent Emissions Datafor FreeClimate TRACE tracks global emissionsSource:Climate TRACE.Used with permission.BOSTON CONSULTING GROUP GOOGLE 12In order to make real progress on decarbonization,organi-zations
88、 need a more granular and actionable view of their carbon footprints,both across their Scope 1,2,and 3 emis-sions and at the level of individual product areas.Until now,that kind of single source of truth has not been avail-able to help operations leaders understand emissions hot spots and explore p
89、otential solutions.CO2 AI,an innovative SaaS platform,helps organizations seamlessly map emissions across their value chains and leverage those insights to drive climate action.AI plays a central role in both assembling emissions data and match-ing it to activities and productsand in simulating solu
90、-tions and building decarbonization roadmaps.In one example,a global health care company seeking to reduce its Scope 3 emissions by 20%by 2030 embraced CO2 AI.The platform enabled it to incorporate 50 times more factors into its calculations and to develop an activity based emissions baseline that w
91、as 20%more precise.And CO2 AIs simulation and roadmapping tools enabled it to identify decarbonization opportunities that would deliver 120%of its emission reduction target.Application area:Micro-Level MeasurementCO2 AI:Helping Business Ecosystems Reduce their Carbon FootprintsMeasuring and managing
92、 emissions with CO2 AISource:CO2 AI.Used with permission.13 ACCELERATING CLIMATE ACTION WITH AIReduction and RemovalAI has the potential to aid organizations in reducing and removing emissions in two ways:enabling emissions reduction and supporting nature-based and technology-based carbon removal.En
93、abling Emissions Reduction.AI can contribute to the creation of more efficient and cleaner energy systems in multiple ways.It can,by consolidating information from dozens of different organizations and grid components,provide insights on how to optimize electric grid opera-tionsand support informed
94、decision-making on grid planning.It can also help speed transition from fossil fuels to alternative energy sources through better supply and demand forecasts that reduce the need for battery storage and standby power and enable more efficient real-time balancing of electric grids.For example,Tapestr
95、y,an Alphabet project,is creating a single virtualized view of the electricity system with the goal of lowering emissions,minimizing outages,shortening interconnection queues,and integrating more renewables into the grid.AI is at the heart of its computational and simulation tools.Relatedly,on the s
96、ubject of renewables,Frances Engie has partnered with Google Cloud to develop and pilot an AI-powered tool that can provide grid opera-tors with more accurate real-time forecasts of wind energy supplies.9In Africa and India,Husk Power Systems provides“pay-as-you-go”100%renewable power to off-grid an
97、d weak-grid communities that is 30%cheaper than the alternative:diesel generation.Husk estimates that its AI model en-ables it to predict user demand with 80%accuracy across its microgrids,thereby improving capacity utilization,re-ducing costs,delivering lower prices,and guiding capital investment i
98、n additional capacity.Moreover,AI-driven insights can also enable people and organizations to make smarter decisions that decrease emissions.For instance,as a result of using AI to improve demand forecasting,manufacturers can avoid both over-production and the emissions those unsold goods would prod
99、uce.Similarly,AI-optimized transportation can reduce emissions by identifying and directing drivers to the most efficient routes.As of September 2023,Google Maps fuel-efficient routing feature was estimated to have helped prevent more than 2.4 million metric tons of CO2e emis-sions since its launch
100、in October 2021equivalent to taking approximately 500,000 fuel-based cars off the road for an entire year.10 And between 2011 and 2022,Googles Nest thermostats are estimated to have helped customers cumulatively save 113 billion kWh of energythe rough equivalent of double Portugals annual electricit
101、y useby proposing thermostat adjustments based on customer behavior,such as automatically adjusting temperatures when customers are away from home.11In the realm of agriculture,the integration of AI tools with technologies such as drones can help farmers monitor their crops in real time for better f
102、ield management,thus enhancing agricultural productivity while minimizing GHG emissions.Moreover,AI-driven precision farming helps empower farmers to make well-informed,data-driven decisions regarding farming practices,crop selection,irrigation,fertilizing,pest management,and harvesting.This approac
103、h streamlines resource utilization and,if done purposefully,can minimize the environmental impact associated with farming practices.For example,Alphabets project Mineral is using robotics,AI,and computer vision to create a more sustainable food production system.It is developing perception-powered s
104、olutions with partners across the agriculture value chainfrom grocery retailers and enterprise farms to equipment manufacturers and crop protection companiesto develop a better under-standing of the complex interactions of plants,their grow-ing environments,and farm management practices.12Another in
105、teresting use case involves using AI to reduce contrails.Contrails,the white clouds that sometimes form behind airplanes when they fly,are responsible for about 35%of the aviation sectors global warming impact.AI solutions developed by Google Research in partnership with Breakthrough Energy have ena
106、bled airline pilots in trial studies to reduce contrails by up to 54%.13(See the sidebar The Contrails Impact Task Force:Addressing Avia-tions Other Contribution to Warming.)Supporting Nature-Based and Technology-Based Removal.According to the IPCC,limiting warming to 1.5C by 2100 will require an ex
107、tensive deployment of CO2 removal measures,of which there are two broad types:nature-based removal in which carbon is removed by and stored in natural sinks such as forests,algae,and wetlands,and technology-based removal via approaches such as direct air capture(DAC).14 AI can play a supporting role
108、 in both types of removal.In nature-based removal,AI-based solutions can help quantify and verify the level of carbon sequestration achieved in an ecosystem,enabling public and private sector leaders to make informed decisions regarding the deployment of natural solutions,including land manage-ment
109、and reforestation efforts.One actor in this space is Albo Climate.Using myriad remote sensing imagery and proprietary AI algorithms,Albo Climate monitors and quantifies environmental metrics such as above-and below-ground carbon sequestration and land-use dynamics in forestry and agricultural ecosys
110、tems and provides trans-parent and reliable data to various stakeholders of the nature-based markets.BOSTON CONSULTING GROUP GOOGLE 14Aviation is one of the hardest sectors to decarbonize.According to the IPCC,contrailsthe thin,white lines you sometimes see behind flying aircraftaccount for roughly
111、35%of aviations global warming impact.Contrails are created in certain atmospheric conditions that enable water droplets to condense and freeze around the soot particles from jet engine exhaust.Some dissipate quickly,while others form into persistent contrail-cirrus clouds that can last for hours,tr
112、apping heat that would otherwise escape into space.To mitigate this,Google Research teamed up with Ameri-can Airlines and Breakthrough Energy to combine AI and huge amounts of data to predict where contrails will form and how planes can avoid making them.The trial reduced contrails by 54%across 70 l
113、ive American Airlines flights.As part of the initiative,American Airlines integrated contrail-likely zones into the tablets their pilots used,enabling them to make in-flight altitude adjustments to avoid creating contrails,just as they do to avoid turbulence.In October 2023,a new partnership was ann
114、ounced with EUROCONTROLs air traffic control center that manages the airspace over Belgium,the Netherlands,Luxembourg,and northwest Germanyone of the busiest airspaces in the world.With this partnership,EUROCONTROL will be able to provide aircraft flying through its airspace with information about h
115、ow to avoid making contrails.15Application area:Enabling Emissions ReductionThe Contrails Impact Task Force:Addressing Aviations Other Contribution to WarmingSource:How AI is helping airlines mitigate the climate impact of contrails,Google Blog,August 2023.Used with permission.Contrails detected ove
116、r the United States15 ACCELERATING CLIMATE ACTION WITH AIIn the sphere of technology-based removal,advanced technologies such as DAC can filter and capture CO2 from the air as it passes through a device.The captured CO2 can then be stored underground in,for example,saline aquifersor prepared for ind
117、ustrial applications.This technology currently faces questions about its energy efficiency,which may hinder scalability,but ongoing re-search and development efforts may resolve these chal-lenges.Another approach is bioenergy with carbon capture and storage,which generates energy from biomass such a
118、s wood and agricultural waste and captures the resulting CO2 for underground storage or industrial use.AI can play a role in assessing optimal capture and storage locations,monitoring for potential leaks,and optimizing the industrial processes and materials used in carbon capture.AI as an Enabler of
119、 Adaptation and ResilienceEven if mitigation efforts enable us to achieve the Paris Agreement goal of limiting warming to 1.5C,communities will need to adapt and build resilience to the consequences of a warming planet.AI-driven solutions can play a critical role in two specific areas:predicting haz
120、ards and manag-ing vulnerability.Hazard PredictionSome climate impacts,such as sea-level rise,are slow moving.Others,such as flooding,are fast.The incidence of extreme weather eventsfor example,heatwaves,heavy precipitation,droughts,and severe stormsis increasing.A CDP analysis estimates that S&P 50
121、0 companies face a potential$40 billion to$50 billion impact from physical climate risks by 2026.16 In light of that,government and business leaders need to double down on preparationand AI can make a significant difference in two areas:enabling early warning and projecting long-term trends.Building
122、 Early-Warning Systems.AI can save lives and minimize property damage by predicting devastating weather events and giving governments and people time to prepare.Riverine floods offer an example.Extreme rains take time to flow down river and that time can be put to good use taking action to mitigate
123、their impact on commu-nities along the banks.Flood Hub,an initiative of Google Research,enables governments,aid organizations,and individuals to take timely action and prepare for riverine floods via locally relevant flood data and forecasts up to seven days in advance.(See the sidebar Providing Ear
124、ly Warning of Rising River Levels with Flood Hub.)Projecting Long-Term Trends.The impacts of climate change vary across regions and cities due to the complex interplay of local geography,weather patterns,ocean cur-rents,and other variables.AI can play a pivotal role in developing powerful climate mo
125、dels that can anticipate adverse impacts such as rising sea levels and droughtand assessing their implications for local communities on factors such as economic development,infrastructure,and agricultural and fishing output.These insights enable the development of thoughtful resilience strategies to
126、 mitigate the effects of climate change.Vulnerability ManagementPrediction is undeniably a crucial aspect of preparedness;however,it should be complemented by proactive efforts at the local level to fortify communities against both sudden and protracted crises.AI is showing significant promise in he
127、lping manage crises and build forward resilience for physical infrastructure and living creatures.Responding to Crises.When a crisis strikes,it is incredi-bly valuable to have tools that help to ensure the right resources are allocated to the right tasks and in the right locations.AI can help crisis
128、 managers by combining data from scattered sources and offering a consolidated,real-time view of facts on the ground.One example is Googles Wildfire Boundaries tracker that uses satellite imagery and AI models to detect the location of wildfire boundaries in real time and then display them in Search
129、 and Maps to support both responders and residents in making informed decisions.17Another example is in helping governments and NGOs to prepare and provide support and shelter to people displaced by extreme weather events.(See the sidebar Helping UNHCR Get Ahead of Forced Displacement in Somalia.)Bu
130、ilding Resilient Infrastructure and Protecting Biodiversity.AI can also support risk assessment and remediation planning for critical infrastructure,helping localities model vulnerabilities,prioritize resilience-build-ing investments,and pressure-test plans.In India,Google is helping the government
131、address food and water security challenges by interpreting satellite data to track farm boundaries,measure forest and woodland acreage,and identify improvements to irrigation structures to prepare for droughts.18BOSTON CONSULTING GROUP GOOGLE 16Annually,floods cause thousands of fatalities worldwide
132、,disrupt the lives of millions,and lead to significant financial costs.They are one of the deadliest natural disasters,and climate change is increasing their frequency and severity.Better prediction of impending flooding has the potential to save lives and mitigate the extent of property damage.Floo
133、d Hubpowered by AI models developed by Google Researchaims to predict when and where riverine flood-ing will occur in order to provide timely warnings to govern-ments,organizations,and the people likely to be affected,empowering them to act before the flood strikes.Real-time flood forecasts and visu
134、alizations are available on the Flood Hub platform and,in many cases,also on Search and Maps.As of 2023,Flood Hub covers more than 80 countries,providing forecasts up to seven days in advance.It offers alerts for geographies across Africa,Europe,South and Central America,and the Asia-Pacific region
135、that com-bined are home to 460 million people.In October 2023,Flood Hub expanded to the US and Canada,covering more than 800 riverbanks where over 12 million people live.The goal is to bring flood forecasting to every country and to include more types of floods through ongoing collaborations with go
136、vernments,communities,academics,and organiza-tions such as the World Meteorological Organization.19Application area:Early Warning SystemsProviding Early Warning of Rising River Levels with Flood HubFlood Hubs heat map of flood riskSource:Google.Used with permission.17 ACCELERATING CLIMATE ACTION WIT
137、H AIMillions of people in Somalia face displacement as a consequence of climate-driven natural disasters,resource shortages,and violent conflict.In 2022 alone,climate-driven eventsdrought,floods,wildfires,and storms displaced 1.2 million Somalis,representing nearly two-thirds of all internal displac
138、ements for the year.On point to support those refugees,UNHCR,the United Nations refugee agency,wanted a better way to forecast where and when to deploy its resources.It partnered with Omdena,a global,crowdsourced commu-nity of AI experts,to develop AI solutions that could predict displacement a mont
139、h in advance.Omdena developed machine learning models that helped predict areas for intervention based on identified conflict“hot zones”com-bined with drought and agricultural production metrics.The insights from these models are enabling UNHCR to optimize the assignments of its personnel and deploy
140、ment of resources.Application area:Adaptation and ResilienceHelping UNHCR Get Ahead of Forced Displacement in SomaliaBOSTON CONSULTING GROUP GOOGLE 18Moreover,AI can help people understand and mitigate the impacts of a warming planet on agriculture,fisheries,and the broader natural world.In agricult
141、ure,it can control intelligent irrigation systems,identify early signs of crop diseases,predict future yields based on current trends,and promote knowledge transfer across biomes.It can also help protect wilder places.Global Forest Watch,for exam-ple,uses AI and satellite imagery to create a real-ti
142、me tool to monitor and combat deforestation.AIs Foundational Capabilities to Support Climate ActionAI can also help turbocharge foundational capabilities that are essential to shaping an effective response to the climate crisis.Climate Modeling.AI can strengthen climate models by filling in data gap
143、s,enabling the incorporation of additional variables,and navigating data sets too large for human analysis.It can also yield more accurate estimates by modeling multidimensional complex systems and the feedback loops among,for example,climate and socioeco-nomic variables.The greater accuracy and pre
144、cision of the AI models in turn enhances the impact of AI-supported climate applications.Climate Economics.AI can improve estimates of the financial implications of climate-related impacts and re-sponse measures,enabling policymakers and private-sector leaders to better understand which investments
145、today are likely to yield the greatest benefit.For example,government leaders in Lagos,Nigeriaone of the African continents most vulnerable citiesleveraged AI to model the potential financial and socioeconomic impacts of rising sea levels and evaluate those investments with the highest impact on ada
146、ptation and resilience.(See the sidebar Helping Lagos Shape Its Strategy for Climate-related Adaptation and Resilience.)Another example is Sprout,an insurtech startup that provides coffee farmers with crop insurance based on weather fluctuations.(See the sidebar Sprout:Helping Smallholder Coffee Far
147、mers Navigate Climate Change.)Education and Behavioral Change.AI can help influ-ence climate-friendly behavior.After all,people may wish to make climate-friendly choices but lack the information to know,for example,the relative carbon footprint of two pairs of jeans.By providing relevant information
148、,AI can help consumers make environmentally conscious choices.For example,Google Maps uses AI to suggest fuel-efficient routes that have fewer hills,less traffic,and constant speeds with the same or similar ETA.The feature is avail-able in the US,Canada,Europe,and Egypt and will be rolling out in In
149、dia and Indonesia during 2023.In India and Indonesia,the feature will be expanded to two-wheelers,helping even more people to travel more sustainably.20Innovation and Breakthroughs.AI can also accelerate breakthrough innovation in domains that could open new frontiers in the battle against climate c
150、hange.For instance,in the field of material science,AI has already aided the discovery of a new family of solid-state materials that conduct lithium.These solid electrolytes will be helpful in the development of solid-state batteries that offer longer ranges and increased safety for electric vehicle
151、s.In the field of clean energy,Google DeepMind has pioneered a deep reinforcement learning system that helps researchers better control nuclear fusion plasma,opening new avenues to advance fusion research and thus clean energy alterna-tives.(See the sidebar Accelerating Fusion Science through Better
152、 Plasma Control.)AI is clearly poised to play a pivotal role in shaping positive climate outcomes.But,alongside its benefits,it is crucial to also acknowledge the potential risks associ-ated with its implementation.The next chapter offers some perspective.19 ACCELERATING CLIMATE ACTION WITH AILagos
153、is one of Africas most populous urban areas.It is also home to the continents fourth-busiest port and con-tributes around 30%of Nigerias GDP.At an average eleva-tion of only 1.5 meters above sea leveland with much of the city at or below sea levelit is also one of the cities most endangered by risin
154、g sea levels.But Lagos has set itself on a journey to build a more resil-ient future,leveraging data and analytics to build a robust adaptation and resilience plan that addresses the impact of extreme weather on its key systems.The plan is guiding the city to mobilize financing and implement effecti
155、ve governance,legislation,and monitoring capabilities.AI-enabled tools have played a central role in guiding Lagoss journeyfirst,in risk assessment and then in shaping and prioritizing strategies.Lagoss risk assessment was based mainly on four key indicators:flood intensity,capital cost to damaged i
156、nfrastructure,impact of extreme events on GDP,and the number of citizens affected.The data is calibrated in increments of only 500 square meterssmall enough to show fine-grained patterns of climate vulnerability.Overall,the model estimated 165 square kilometers would likely be inundated,700,000 resi
157、dents would need to be relocated,and there would be$5 billion in damage to transport,communications,and power infrastructure.The heat map below shows the models prediction of inundation risk,with the red areas the most vulnerable.These AI-powered tools then enabled Lagoss leaders to simulate the rel
158、ative contribution of different strategies under multiple scenarios.It also helped estimate the cost of inaction,which,at about$30 billion,represents 12 times the budget of the Lagos metropolitan area.By utilizing these tools,policymakers have curated a priori-tized portfolio of projects spanning th
159、ree broad areas.One of these areas concentrates on enhancing infrastructure resilience,exemplified by initiatives such as the construc-tion of an 18 km embankment and sea walls designed to safeguard more than 2.7 million individuals,including 700,000 from vulnerable populations.Another revolves arou
160、nd bolstering community resilience and safeguarding vulnerable groups,with initiatives such as the establish-ment of infectious disease surveillance systems.And the third factor centers on proactively addressing risks and enhancing crisis response capabilities,such as the estab-lishment of well-defi
161、ned post-disaster procedures.The city has actively engaged with a diverse array of stake-holders from the public,private,and social sectors.Cur-rently,it is in the process of implementing its Climate Adaptation and Resilience Plan while securing funding from various sources including the private sec
162、tor,public sector,NGOs,and others.Application area:Climate Modeling,Climate Economics,and Adaptation and ResilienceHelping Lagos Shape Its Strategy for Climate-Related Adaptation and ResilienceInundation risk heat map for LagosSource:BCG Climate Impact AI Platform.BOSTON CONSULTING GROUP GOOGLE 20Co
163、ffee production faces a significant threat from the extreme temperatures and unpredictable rainfall that are a result of climate change.Small farms are particularly vulnerable.And many are seeing declines in yields of up to 15%as a consequenceyet few have access to or can afford crop insurance.Sprou
164、t,an insurtech startup,offers an innovative,AI-driven solution that helps farmers navigate risks and withstand full or partial crop failures resulting from extreme climate conditions.Sprout brings its own proprietary data on coffee farming together with data from satellites and other sources to do t
165、wo things that support smallholder farmers.First,via the Sprout mobile app,it offers locally customized agronomy advice to farmers on how best to navigate unex-pected weather trends.This reduces intra-seasonal risks,thereby requiring less insurance cover.Second,via an innovative index insurance offe
166、ring,Sprout helps alleviate downside impacts.Premiums are paid by the farmers direct customersand ultimately by developed-world coffee consumers.Index insurance doesnt require farmers to file claims to receive compensation.Instead,it provides automatic payouts should an index variable for the local
167、area(for example,rainfall)fall below a target value.Sprout is piloting its Climate Smart CoffeeTM program with support from USAID Development Innovation Ventures in Kenya.Sprout aspires to offer protection to over 1 million farmers worldwide by 2030.Application area:Climate EconomicsSprout:Helping S
168、mallholder Coffee Farmers Navigate Climate ChangeSprouts AI-enabled crop insurance offeringSource:Sprout.Used with permission.21 ACCELERATING CLIMATE ACTION WITH AINuclear fusion has the potential to be a source of abun-dant,clean energy.It is the reaction that powers the stars of the universe.But h
169、uge breakthroughs will be required for it to become cost effective and scalable.A critical aspect of fusion research involves learning how to control and sustain a hydrogen“plasma”that is hotter than the core of the sun.One tool researchers use is a tokamak,a doughnut-shaped vacuum that aims to cont
170、ain the plasma by making thousands of adjustments per second to a set of powerful magnetic coils.The magnets seek to keep the plasma from touching the vessel walls,which would,at a minimum,dissipate heat,or worse,damage the tokamak.Since the worlds tokamaks are in high demand,one way to advance and
171、accelerate fusion research is through simulation.A collaboration of Google DeepMind and the Swiss Plasma Center at EPFL,a Swiss research university,has leveraged AI to create the first deep reinforcement learning system for fusion research.It simulates EPFLs Variable Configura-tion Tokamak(TCV)and h
172、as successfully modeled ways to stabilize and sculpt plasma that have subsequently been validated in the actual TCVopening new avenues to advance nuclear fusion research.Application area:Innovation and BreakthroughsAccelerating Fusion Science through Plasma Control SimulationBOSTON CONSULTING GROUP
173、GOOGLEAs governments and businesses increasingly rely on AI to help mitigate emissions,build resilience,and adapt to a changing climate,a critical question arises:what risks are associated with the rise of AI?Under-standing and navigating these risks is essential if we are to scale AI responsibly an
174、d manage its environmental foot-print.GHG emissions,water,and waste management are three key areas that will be addressed in this chapter.Issue 1:AIs Energy-Related GHG EmissionsIn 2022,global data center electricity consumption ac-counted for 1.0%to 1.3%of global final electricity de-mand.21 Furthe
175、r,a 2022 paper in Nature Climate Change estimates that cloud and hyperscale data centers are responsible for 0.1%to 0.2%of global GHG emissions and that roughly 25%of their workloads are related to machine learning.22A critical question is howwith AI at the start of a new innovation and adoption cur
176、vedata center electricity use and related GHG emissions will evolve going forward.There is a critical need for more deep research on this topic,but at this stage,we can make some important observations.Navigating AIs Potential Risks23 ACCELERATING CLIMATE ACTION WITH AIHistorically,data center energ
177、y consumption has grown much more slowly than demand for computing power.Between 2015 and 2018,for example,the IEA reports that data center electricity use was flat despite a doubling of compute demand and tripling of internet traffic.Further-more,a paper in Science reports that between 2010 and 201
178、8a period during which data center compute de-mand increased sixfold and internet traffic tenfolddata center electricity use grew just 6%as a result of a shift from inefficient on-premises data centers to the highly energy-optimized cloud.23 And we still have optimization potential in the future.In
179、2022,the average annual power usage effectiveness for Googles global fleet of data centers was 1.10,compared with the industry average of 1.55.24Looking forward,Exhibit 4 illustrates the key drivers of AIs electricity use and GHG emissions throughout its life cycle.At the level of the grid,emissions
180、 will be determined by the power sources selected to power a data center and the efficiency of the distribution system.Compute machines and data center design choices also can have a powerful influence.The choice of more energy-efficient servers optimized for AI models can make a significant differe
181、nce.AI model developers can be more mindful of the energy intensity of their programming choices.And a critical unknown is how quickly and how much end-user demand for AI-enabled products and services increases compute demand.So how might these factors evolve?A 2022 paper by researchers from UC Berk
182、eley and Google,which studied the energy requirements for machine learning training,identifies four best practices with the potential to reduce energy use and emissions.25 Lets take each in turn:Energy Source Carbon Intensity.In data center emissions profiles,such as in real estate,location matters.
183、In 2022,Norways electric grid had an average carbon intensity of 29 g CO2e/kWh,compared with an average of 102 in Brazil,367 in the US,489 in Singapore,and 709 in South Africa.26 And even within a country,the share of carbon-free energy can vary significantly from region to regionmeaning that many o
184、rganizations operate in jurisdictions without easy access to clean energy.And while trade associations such as the Clean Energy Buyers Association,RE-Source,and the Asia Clean Energy Coalition are working to increase access,changes are still needed.Other choices matter too.Information and communica-
185、tions technology(ICT)companies have led the world in adopting renewable energy power purchase agreements(PPAs),with Amazon,Microsoft,Meta,and Google having signed renewable energy agreements totaling almost 50 GW of clean energy generation capacity through 2022,equal to the generation capacity of Sw
186、eden.27 And many tech leaders are going further.Google,for example,has a target to run on carbon-free energy in every grid where it operates by 2030,is already procuring energy from geothermal power plants and battery-based backup power systems,and is piloting the use of AI algorithms to better pred
187、ict wind production and facilitate its integration into the grid.28Exhibit 4-Critical factors determining AIs emissions footprintSource:BCG analysis.AI-related energy flowAI-relatedemissionsPowerplantsEnergy source carbon intensityDatacenterinfrastructureModeldevelopment&codingAI adoption&usageEnerg
188、ydistributionData center operationsDevelopers buildingAI modelsEnd-userdemand forAI-enabledproductsBOSTON CONSULTING GROUP GOOGLE 24 Data Center Infrastructure.Data center energy needs are driven by server and infrastructure choices.Servers specifically designed for machine learning can run AI model
189、s faster while using less power.But servers arent the only source of energy demand.Additional power is needed to operate the facilityfor example,for cooling and lighting.And specific data center equipment choices matter.A study by researchers at Berkeley and Google finds that Google owned,designed,a
190、nd operated cloud data centers,for example,can be 1.42.0 times more energy efficient than traditional data centers,and hard-ware specifically designed to support AI and machine learning can be 25X more efficient than off-the-shelf systems.29 Googles machine-learning optimized Tensor Processing Units
191、(TPUs)are an example:TPU version 4 has proven to be one of the fastest,most efficient,and most sustainable machine learning infrastructure hubs in the world with the potential to generate 93%fewer emissions compared with unoptimized servers using P100 GPUs.30However,new processors require new manufa
192、cturing practices and therefore attention must be paid to their embodied carbon.Future hardware design should look at optimizing full life cycle,instead of just operational,GHG emissions.And relatedly,more effort is needed from the semiconductor industry to understand and reduce embodied emissions.M
193、odel Development and Coding.State-of-the-art approaches to developing AI models are varied and evolving,but one thing is clear:the desire for more precise and accurate model outputs has been leading to more complex models that rely on larger sets of training data and require more processing power.31
194、 These more complex models may lead to higher energy consumption,all other factors being equal.Nonetheless,it is important to note that AI model design is an evolving field,and new releases and versions of complex models consistently demonstrate improved energy efficiency while maintaining model per
195、formance.Indeed,ongoing improvements in software and algorithmic optimization have the potential to significantly enhance efficiency and decrease computational requirements.One example is the development of evolved sparse models for deep neural networks,which can reduce computations by approximately
196、 5 to 10 times while maintaining the same level of output quality compared with denser baseline models.32 Adoption and Usage.AI use is forecast to grow.In an IBM Global survey,35%of companies reported already using AI in their business in 2021,with an additional 42%stating that they are piloting AI
197、applications for lat-er adoption.33 Some of the growth in AI adoption might replace existing workloads in todays data centers,but given the expansion of use cases,aggregate demand will grow.At present,however,there is no recent,peer-reviewed research that forecasts future AI workloads and energy nee
198、ds.Any AI model has two phases to its life cycle:training and inference.Training gives the model its smarts and happens sporadically,while inference is the production phase in which the model is used by customers.Infer-ence workloads are therefore driven by customer adop-tion and usage.And today,inf
199、erence processing accounts for 80%90%of AI and machine learning workloads.34These factors are synthesized in Exhibit 5,which makes a critical point:while there is uncertainty on future AI emissions,our decisions matter.We are not suggesting that these are equally important,only making the point that
200、 already today there are technology,architecture,and location options that can significantly mitigate AIs GHG impact.And,beyond these active measures,there are other factors that will influence the speed and scale of AI deployment(for example,the economics of data center development).Data centers ca
201、n take years to build and require significant capital investment.Issue 2:AIs Water UseData centers generate insight,but they also generate heat.This heat must be dissipated to protect the servers,com-munication equipment,and storage devices they contain.For warehouse-scale data centers,water-based c
202、ooling is the most common approach.And among water-cooling solutions,evaporative coolingin which water absorbs ambient heat through evaporationis frequently both the cheapest and the most energy efficient.Google found that its water-cooled data centers use about 10%less energy than its air-cooled da
203、ta centers.35 But water cooling has the potential to exacerbate pressure on water resources in specific locations.A Virginia Tech study estimates that one-fifth of US data centers (primarily located in Western states)draw their water from moderately to highly stressed watersheds.3625 ACCELERATING CL
204、IMATE ACTION WITH AIIn the data center sector,water use is not widely report-edand actual volumes will vary widely based on the centers size,location,local weather conditions,and the use of its infrastructure.A 2016 study from the US Department of Energy estimates data center water consumption at 1.
205、7 billion liters/day,of which 0.3 billion liters/day is used on site for cooling,or 0.02%of total US water consumption of 1,218 billion liters/day.37Clearly,data center operators need to be mindful of and manage a tradeoff between energy use and water use.Google began disclosing water use for each o
206、f its owned US data center locations in 2021 and for global owned data center locations in 2022.A Google analysis of the 2021 data finds that its embrace of water cooling across its data centers reduced its carbon emissions by roughly 300 kilotonsthe emissions equivalent of about 64,000 passenger ve
207、hicles.38 The 5.2 billion gallons of water required to do that in 2022 was comparable to the water needed to irrigate 34.8 of the more than 11,000 golf courses in the United States or the annual water consumption of 69,800 average American homes.39,40Tech players are also starting to explore new,mor
208、e wa-ter-efficient approaches.In a data center,temperature,airflow,and relative humidity(RH)are the three critical factors to manage.Based on guidance from the American Society of Heating,Refrigerating,and Air-Conditioning Engineers,Meta has experimented with shifting the low-er-bound for RH in its
209、data center from 20%to 13%.The nine-month pilot yielded water savings of 40%.41Data center operators are also taking steps to replenish their water consumption.Google,for example,has a target to replenish 120%of the freshwater volume it consumes on average by 2030 through initiatives including wetla
210、nd restoration,rainwater harvesting,and land conservation.Exhibit 5-Critical choices can shape AIs emissions footprintNote:The figure is based on the most recent data available on actual emissions profiles for different choicesand is not intended to suggest that these factors are equally important.1
211、Patterson,D.et al.,2021.Carbon Emissions and Large Neural Network Training.2Patterson,D.et al.,2022.The carbon footprint of machine learning training will plateau,then shrink.3Energy Institute Statistical Review of World Energy.4Europe Data Centers main locations:Frankfurt,London,Amsterdam,Paris,and
212、 Dublin.Factor and Baseline MetricAI adoption&usageReduced emissions100%ML-oriented1(TPU v4,2021)NorwayML-oriented1(TPU v2,2019)Evolved(2019)and Primer(2021)modelsCloud data centersIncreased emissions+100%BaselineUnknown8090%3050%93%94%Brazil77%USSingaporeSouth Africa16%+12%+60%FLAP-Dcountries4 22%2
213、3%Baseline:Current AI compute consumptionModel development&codingBaseline:Transformer(2017)Data center infrastructure(servers)Baseline:Unoptimized systems energy use(P100 from 2017)Data center infrastructure(non-IT equipment)Baseline:Traditional data center energy needsEnergy source carbon intensity
214、3Baseline:Global grid emission intensity in 2022BOSTON CONSULTING GROUP GOOGLE 26Issue 3:WasteThe United Nations estimates that globally 53.6 million metric tons of electronic waste was generated in 2019.This figure represents a 21%increase in just five years.And the volume is projected to increase
215、to 74.7 million metric tons by 2030representing a near doubling of e-waste over a 20-year period.While there is currently a lack of specific data regarding e-waste generated by data centers,they clearly only ac-count for a fraction of the broader e-waste challenge.None-theless,there is a clear imper
216、ative for tech firms to take a smarter,more circular approach to waste.Circular economy principles emphasize the opportunity to maximize the lifespan of products and materials through practices such as reuse and recycling.For tech companies,this entails designing products and data center infrastruc-
217、ture with an eye toward longevity and ease of upgrading or repurposing.Moreover,it involves establishing efficient systems for recycling and refurbishing electronic equip-ment,ensuring that valuable materials are reclaimed and that hazardous substances are disposed of responsibly.Microsofts Circular
218、 Centers,for example,focus on finding productive uses for decommissioned equipment,including new homes for older equipment,such as in schools.They break servers down into components that can be reused by others,and return other items to suppliers for recycling and reclamation.42 The first of these c
219、enters opened in 2020 in Amsterdam and has been able to channel 83%of e-waste into reuseand 17%into recycling.Based on that initial success,five other Circular Centers were established in 2022.In 2016,Google announced a“Zero Waste to Landfill”goal for its data center operations,which it defines as m
220、ore than 90%of waste diverted from landfill.43 In 2022,38%of Google owned and operated data centers had achieved Zero Waste to Landfill.Other firmsfor example,Iron Mountain through its 2021 acquisition of ITRenew,which specialized in refurbishing and repurposing used data center equipment from hyper
221、-scale operatorsare pursuing circularity as a business opportunity.Other Potential Risks The rise of AI also brings a set of societal and ethical risks that must be managed vigilantly.It is important to have a set of clear principles that guide the development and use of AI.For example,Google has ou
222、tlined the following seven key principles for its development of AI applications:441.Be socially beneficial2.Avoid creating or reinforcing unfair bias3.Be built and tested for safety4.Be accountable to people5.Incorporate privacy design principles 6.Uphold high standards of scientific excellence 7.B
223、e made available for uses that accord with these principlesApplication Selection and Optimization Metrics.There are any number of climate-unfriendly applications to which AI could be applied,for example,oil and gas explora-tion.Google,for instance,has pledged not to develop customized AI/machine lea
224、rning solutions to facilitate upstream extraction for oil and gas.In addition,algorithms could be designed to optimize for financial or other out-comes over environmental ones.For example,travel web-site algorithms could steer customers to the cheapest flights regardless of carbon emissions instead
225、of guiding them toward an informed compromise between price and emissions.It is also important to guard against unintended conse-quences.While AI can help optimize resource use and curtail emissions,there are scenarios in which it could influence consumer behaviors that lead to unintended increases
226、in emissions.For instance,AI-powered autono-mous vehicles and smart transportation systems can optimize routes and reduce fuel consumption,but their convenience could lead to increased vehicle use,which could potentially increase overall emissions if the vehicles are not predominantly electric or po
227、wered by renewable energy sources.Looking beyond the risk of negative environmental impact,we must guard against unethical applications,for example,the spreading of misinformation and disinformation.27 ACCELERATING CLIMATE ACTION WITH AIEquity and Bias.Clearly,AI models need to be trained on diverse
228、 data sets that reflect the worlds range of people to ensure both fairness and the accuracy of model output.And beyond that,it is essential that we take positive steps to ensure that the growth of AI does not exacerbate region-al disparities.Today,the majority of advanced climate modeling and AI dev
229、elopment occurs in the Global North.This divide arises from several factors,including the Global Norths more extensive technological infrastructure that allows for richer data collectionfor instance,through satellites and dronesand its greater computing power.Additionally,AI expertise and resources
230、tend to be more readily available in these affluent regions when compared with the Global South.However,the concentration of AI development and application in certain regions has significant repercussions.Climate AI models primarily trained on data from the Global North may inadvertently neglect vit
231、al information about the Global South,with its distinct climate patterns,vulnerabilities,and emissions sources.A Google Research team in Ghana,for example,is focused on leading many sustainability initiatives of particular interest to Africa in collaboration with local universities and research cent
232、ers.45 Additionally,historical data often mirrors emissions and climate impacts in more industrialized regions,further skewing data representation.When such biased AI models inform climate assessments or policymaking for the Global South,they risk yielding flawed and inaccurate outcomes.Excluding sp
233、ecific regions or historical periods can result in inaccurate predictions and evaluations of carbon emissions,environmental consequences,and climate trends in these underserved areas.This undermines climate science accuracy and obstructs effective decision-making and planning for climate adaptation
234、and mitigation in the Global South.Privacy and Security.The large data sets that train AI algorithms will at times include personal data.It is essential to ensure that AI applications are aligned with established privacy standards and regulations to protect individuals.Given the promise of AI to add
235、ress the climate crisis,policymakers will want to encourage its usebut also mitigate its potential risks.The next chapter offers a sum-mary of critical policy outcomes.BOSTON CONSULTING GROUP GOOGLEPolicymakers play a central role both in harnessing the potential of AI for climate action and in ensu
236、ring its sustainable and equitable use.In this chapter,we share a set of suggestions for policymak-ers that synthesizes the convergences and complementari-ties across more than 30 expert interviews and a compre-hensive review of the literature(See the References section for more detail).We also draw
237、 on best-practice approaches from related policy domains such as energy,transport,and buildings.Policy can make a difference in ensuring and accelerating the following three critical outcomes:Enabling the use of AI for climate action by building awareness and ensuring equal access to data,tech-nolog
238、y infrastructure,and talent around the globe Deploying public-sector solutions on priority use cases while catalyzing private sector action through the right incentives Promoting the responsible use of AI in climate action,taking into account its potential environmental as well as social impactsAI f
239、or ClimateA Summary of Critical Policy Outcomes29 ACCELERATING CLIMATE ACTION WITH AIExhibit 6 offers a menu of possible policy moves in support of these desirable outcomes.The rest of the chapter pro-vides examples of how policymakersas well as business and social-sector leadershave already taken s
240、teps to con-tribute to these outcomes.While policy priorities must be tailored to the specific circumstances and capabilities of each country and region,we believe all leaders should adopt clear AI principles to ensure the responsible devel-opment and application of the technology.We expand on the t
241、hinking in the remainder of the chapter,and hope our framing provides inspiration and a starting point.Enable AI for ClimateIf we are to be successful in maximizing AIs contribution to climate action,one critical area for policy focus is on the supply side,ensuring that AIs critical inputshigh-quali
242、ty data,technology infrastructure,and talentare available wherever needed.We discuss each in turn.Encourage Data Collection and SharingAI impact starts with good data.Without it,algorithms cannot generate accurate and effective insights or recom-mendations.Actionable wildfire alerts,for example,cann
243、ot be developed without access to high-quality,real-time satellite dataand good agricultural advice is impossible if data is available only for a limited number of crops or geographies.Moreover,to yield useful insights,many AI applications for climate will need to tap into multiple data sources.Lond
244、on Transport,for example,leverages data on emissions sources,road traffic,air quality,and population density to monitor air pollution challenges and identify highly exposed locations.Making the data available is not enough;it also needs to be accessible in standard formats that allow data from diffe
245、rent sources to be merged safely and efficiently.It is therefore essential that policymakers take steps to make data accessibleat a minimum in the climate sphere,given the urgent need for rapid and effective action to reduce emissions and build resilience globallywhile also protecting trade secrets
246、and intellectual property.Exhibit 6-AI for Climate:A summary of critical policy outcomesPolicy outcomesPolicy movesEnable AI for climateEnsure the availability of data,technology infrastructure,and talentEncourage data collection and sharingPromote the principle ofclimate-related data as a common go
247、odDefine and deliver on publicsector prioritiesShape and execute AI for climate strategies and demonstration casesAddress environmental impacts of AI operationsEnhance transparency and streamline the adoption of sustainable AI practicesEncourage private sector adoptionCreate incentives to accelerate
248、 use of AI for climateCultivate awareness and build expertiseInvest in knowledge and talent to driveAI for climate solutionsPromote socially responsible use of AI for climateEncourage fairness,inclusiveness,safety,and data privacyEnsure technology access and affordabilityEncourage and invest in esse
249、ntial technology infrastructure to support AIPromote environmentally and socially responsible use of AIDrive AI solutions for climate in public sectorand catalyzeprivate sector actionDeploy AI for climateGuide AI for climateSource:BCG analysis.BOSTON CONSULTING GROUP GOOGLE 30Leaders could:Encourage
250、 collection and sharing of climate-related data and tools across public and private organizations.Examples:The UK Climate Change Statistics Portal gives open access to climate-related data and statistics(weather,emissions of GHG,status of surface water,renewable energy share,etc.).This data is compi
251、led from various government departments,agencies,and public bodies.The US Climate Resilience Toolkit,developed by a collaboration of multiple federal agencies and organizations led by the National Oceanic and Atmo-spheric Administration,provides essential climate information,projections,and tools to
252、 help organizations enhance their resilience to climate-related challenges.Sign Smart,also known as the National Greenhouse Gas Inventory System,is an application system developed by the Indonesian government.Its objec-tive is to provide valid,accurate,and up-to-date data and information about GHG e
253、missions,while also enhancing the effectiveness of data processing and GHG estimation at the national,provincial,and district/city levels.Data Commons is an initiative led by Google designed to centralize and streamline publicly available data from diverse sources.It provides a wide range of climate
254、 and sustainability-related data,covering areas such as emissions,natural disasters,and waste.Accessing this data is made simple through an interface that supports natural language searches.The AI-driven query function retrieves results directly from Data Commons,providing links to the original sour
255、ces of information and data.Define standard processes and protocols for climate-relevant data gathering and sharing to ensure data is robust,trustworthy,safe,and respectful of privacy.Examples:The European Space Agencys Climate Change Initiative(CCI)aims to deliv-er a consistent,satellite-derived da
256、ta set of Essential Climate Variables(for example,greenhouse gases,sea level,glacier status,etc.)to aid climate modelers and researchers over the long term.To guarantee consistency among the various projects within the program,it has released data stan-dards outlining the minimal requirements for cl
257、imate data producers.The World Meteorological Organization has published the Technical Regula-tions,an international framework for data standardization and interopera-bility in the fields of meteorology,hydrology,climatology,and related envi-ronmental disciplines.These standards enable the continuou
258、s operation of global systems,ensuring 24/7 observations,data exchange,management,forecasting,and the delivery of authoritative scientific assessments and standardized services.Create data catalogs46 across priority sectors for all climate-relevant data categories(for example,weather,water,agricultu
259、re,energy use,and socioeconomic factors).Examples:The World Meteorological Organization has not only published a set of data standards but also a Catalogue for Climate Data,a curated listing of global,regional,and national data sets of climate-related data that meet its stan-dards for data quality a
260、nd stewardship.Climate Change AI,a nonprofit,has published the CCAI Dataset Wishlist that enumerates currently unavailable data that could accelerate AI-driven climate progress.It classifies desirable data by topic as well as by its current state of availability(for example,public data needing struc
261、ture,private data needing release,scattered data needing collation,and scarce data needing collection).31 ACCELERATING CLIMATE ACTION WITH AIEnsure Technology Access and AffordabilityWithout devices collecting data from the real world(for example,sensors,drones,and satellites)and without connectivit
262、y(for example,fiber and 5G)and computing infrastructure(for example,cloud data centers)AI algo-rithms are powerless.Consider an application such as AI-driven smart irrigation.Field sensors need broadband to send data to servers and storage devices in the cloud.The servers then process the data,compa
263、ring it with insights from a training set,and formulate recommendations that need to be communicated back to the irrigation system in the field.This critical technology infrastructure,while widely avail-able in the developed world,is not currently accessible in many less-developed regions,particular
264、ly the Global South.Affordability is also a concern.For example,customers in less-developed countries pay 6%of per capita gross nation-al income for mobile broadband service,while those in high-income countries pay just 0.4%.47 Leaders could:Build private-public partnerships to ensure affordable acc
265、ess and deploy locally or regionally critical AI technology infrastructure (for example,cloud data centers,satellites).Examples:Japans Ministry of Economy,Trade,and Industry is partnering with Sakura Internet to expand Japans computational capacity.It is providing half of the needed investment to bu
266、ild an AI supercomputer,aiming to accelerate AI development and implementation in Japan.The European High-Performance Computing Joint Undertakinga 7 billion initiative over 20212027 to build petascale and pre-exascale supercomput-ing infrastructure in Europeaspires to accelerate European innovation
267、by expanding access to state-of-the-art tools.The US Community Infrastructure for Research in Computer and Informa-tion Science and Engineering(CIRC)program seeks to increase researcher access to critical technology infrastructure by funding the development and improvement of top-tier research infra
268、structure.Empower the private sector to build and/or expand AI technology infrastructure.Examples:Norways data center strategy,introduced in 2018,features public invest-ments in fiber infrastructure as well as tax incentives(such as property tax reductions)aimed at attracting data center operators a
269、nd ensuring the accessibility of computing infrastructure.The South African government aims to launch a National Cloud and Data Policy.This initiativecurrently under consultationseeks to provide policy certainty for investments in data centers and cloud services and to reinforce South Africas leadin
270、g position in Africa.The draft policy includes provisions for establishing a Special Economic Zone for Digital and ICT,as well as policies addressing data protection,data localization,and cross-border data transfers.Support R&D for AI technology infrastructure across the public sector,private sector
271、,and academia.Examples:The US Networking and Information Technology Research and Develop-ment(NITRD)program is a federally funded R&D initiative focused on advanced IT,covering computing,networking,and software nationwide.Its 25 member agencies invest about$9.6 billion annually in various R&D progra
272、ms,including high-capability computing systems,advanced communi-cation networks and systems,and more.The Quantum Technologies Flagship is a 10-year,1 billion EU initiative that aims to advance Europes leadership in quantum technologies by bridging research with practical applications.It brings toget
273、her research institutions,industry partners,and public funders.BOSTON CONSULTING GROUP GOOGLE 32Cultivate Awareness and Build ExpertisePeople need to remain at the center of policy and climate action.They must be aware of AIs potential and support its use for climate actionand people with AI skills
274、will be needed to help address the climate challenge.It is important to build awareness among all stakehold-erspolicymakers,corporate decision-makers,civil ser-vants,and the broader publicof AIs potential contribu-tions to climate solutions as well as of its risks.With awareness of the challenges,of
275、 key achievements,and of best practice comes support for AI-enabled climate action.Translating that support into climate progress will require,in addition to more AI computing infrastructure,more talent.Without technical experts such as data scientistsand domain experts such as climatologists and cl
276、imate economistsit will be impossible to develop,deploy,and govern climate-related AI applications.In a recent BCG survey,78%of business leaders with responsibility for climate,AI,or both cited insufficient access to qualified talent as a barrier to using AI to address their climate-relat-ed challen
277、ges.And critical talent is not only in short supply,but also heavily concentrated in the developed world.According to the OECDs AI Policy Observatory,North America is home to 30%of the worlds data scientists and machine learning experts,while sub-Saharan Africa hosts under 2%.Leaders could:Establish
278、 AI and climate training and literacy programs for policymakers and the broader public sector workforce.Examples of AI-related trainings:Government AI Campus,created by Google.org and the Rockefeller Foundation,is an online career-development initiative for government staff that prepares them to lea
279、d in the age of AI.AI4Gova European Union-funded masters degree program offered by four leading universities,focuses on AIs public sector application.It is part of a broader European initiative to create AI-related masters programs to build skills in areas such as AI ethics and AIs application to he
280、alth care.The UNESCO-developed Artificial Intelligence and Digital Transformation Competency Framework provides guidance on the essential AI and digital competencies for civil servants.This initiative responds to a significant de-mand for efforts to enhance the digital skills of government officials
281、,particu-larly in Africa.48 UNESCO conducted workshops in Africa and India to gain a deeper understanding of the challenges before formulating its solutions.49Examples of climate-related trainings:The UN Climate Change Learning Partnership(UN CC:Learn)offers a range of introductory and advanced onli
282、ne courses for policymakers to learn how to address climate change and apply an integrated approach to climate action throughout the various stages of the policy cycle.Climate Change and Energy:Policymaking for the Long Term is an executive education program for policymakers developed by Harvards Ke
283、nnedy School of Government.It seeks to equip them with the knowledge,analyti-cal tools,and frameworks needed to comprehend climate science and eco-nomicsand to craft policies and adaptation strategies.33 ACCELERATING CLIMATE ACTION WITH AISupport the creation and expansion of AI and climate-related
284、upskilling programs for corporates(for example,climate change training for AI experts,AI introduction for climate experts).Example of AI for Climate trainings:The Climate Change AI summer school aims to equip individuals who have a background in AI and/or climate change with the knowledge and skills
285、 necessary to address significant climate challenges using AI.Examples of AI-related trainings:Quebecs Ministry of Employment and Social Solidarity has granted$23.4 million in funding to SCALE AI to upskill over 25,000 professionals,execu-tives,and managers in AI between 2019 and 2023.The EU-funded
286、project Artificial Intelligence Skills Alliance(ARISA)unites 20 partners for a four-year period(20222025)to create the European strat-egy for AI skills development,including up/reskilling curricula and learning programs.Examples of Climate Change trainings:Climate-KIC,a knowledge and innovation comm
287、unity supported by the European Institute of Innovation and Technology,provides education pro-grams,at the intersection of zero-carbon,climate resiliency,and innovation,in Europe and online for postgraduates and professionals.The national business support agency of Ireland,Skillnet,offers the Climat
288、e Ready Academy to help Irish businesses develop climate-related skills,including sector-specific programs such as the Energy Leaders Programme.Build AI and climate modules within education curricula(for example,early-stage initiation for K-12,cross-skills for AI students or climate students,etc.).E
289、xamples of climate-related curricula:Since September 2020,Italian students in every grade have spent 33 hours each year learning about climate change and sustainability,bringing Italy to the forefront of environmental education worldwide.In the US,the National Oceanic and Atmospheric Administrations
290、 Collec-tion of Climate and Energy Educational Resources(CLEAN)provides open access to over 700 validated,ready-to-use teaching materials and guidelines suitable for secondary through higher education classrooms.Examples of AI-related curricula:Countries such as Finland,the UK,Japan,and Singapore ha
291、ve introduced computational thinking and programming to pedagogical courses to increase students exposure to coding and computing at early stages of education.50Moroccos Ministry of Higher Education,Scientific Research,and Innovation has initiated the“Code 212”project,designed to help students acqui
292、re skills in coding,programming,big data,AI,and related fields.A core objective is to establish Code 212 centers in all national universities,thereby enhancing students digital competencies alongside their specialized studies.BOSTON CONSULTING GROUP GOOGLE 34Deploy AI for ClimateWhile enabling AI is
293、 necessary to help society mitigate and build resilience to climate impacts,policymakers can play an important role in accelerating the deployment of AI technologies in both the public and private sectors.We discuss each in turn.Define and Deliver on Public Sector PrioritiesThe dozen AI use cases fo
294、r climate outlined in Exhibit 3 are all important,but that doesnt mean that theyre always equally important.Every country and region is different and each confronts its own distinct set of climate challenges given its geography,industry mix,human resources,and relative wealth.Therefore,every country
295、 or region needs to develop its own climate priorities and action plan.For Germany,an industrial giant where the Climate Action Act mandates carbon neutrality by 2045,the priorities for AI might be accelerating the decarbonization of industries and driving energy efficiency.51,52 By contrast,Bangla-
296、deshgiven its significant vulnerability to sea-level rise and extreme weathermight prioritize accelerating its National Adaptation Plan and developing early warning systems for communities.In addition to identifying their high-priority use cases,policymakers must also expedite the application of AI
297、to climate challenges.The public sector can serve as a pivotal catalyst for accelerating AI-supported climate action and set a compelling example for the entire economy.This is particularly important since,while 87%of organizations acknowledge the role of AI in climate action,only 40%can envision pr
298、actical applications within their own operations.Leaders could:Integrate AI solutions and expertise into government strategic planning on climate priorities(for example,Nationally Determined Contributions,National Adaptation Plans,and sec-tor-specific transition plans).Example:AI4PublicPolicyan init
299、iative of the EUs R&D program Horizon 2020is creating an open cloud platform for automated,scalable,transparent,and citizen-centric policy management.Lisbon,for example,is using it to map its current inventory of,and expansion potential for,solar panelswith the goal of developing renewables supply f
300、orecasts to inform building codes and incentive budgets.Define priority sectors or use cases for AI to support climate action at local and regional levels(for example,as part of Nationally Determined Contributions,National Adaptation Plans,or National AI strategies).Examples:53Denmarks AI strategy(2
301、019)identifies three climate-related priority areas:energy efficiency,precision agriculture,and traffic optimization.The Netherlands Strategic Plan for AI(2019)includes commitments to leverage AI in agriculture and in accelerating the energy transition.The UKs AI Roadmap advocates for AI use to addr
302、ess climate change,particularly in the energy sector.The government of the Republic of the Philippines is advocating using AI to tackle climate change adaptation challenges and disaster risk reductionin particular,through programs led by the Department of Science and Technology,such as AI for a Bett
303、er Normal.Implement priority public sector AI solutions for climate.Examples:The US National AI Initiative Act of 2020 formalizes the National Oceanic and Atmospheric Administrations(NOAAs)role in coordinating AI applica-tions for climate,ocean,Earth,and space sciences.NOAA has created the NOAA Cent
304、er for Artificial Intelligence(NCAI),which collaborates across scientific fields to promote responsible and equitable AI use for environmen-tal research.Singapore is using AI to predict floods and test flood-resilient infrastructure.For example,a partnership between the island nations National Resea
305、rch Foundation and the Hydroinformatics Institute(H2i)has developed Virtual Water,a surface water simulation toolbox that can predict and simulate floods resulting from heavy rainfall events.35 ACCELERATING CLIMATE ACTION WITH AIEncourage Private Sector AdoptionPolicymakers could also play a catalyt
306、ic role in accelerating private sector AI adoption by addressing major challenges hindering the at-scale implementation of AI for climate.Globally,we see the following five key hurdles:Unclear Goals.In the absence of clear regional,national,or sector-specific objectives for climate action,AI-driven
307、innovation may become fragmented,and resource allocation inefficient.The establishment of priority innovation domains for climate action can be a significant unlock.Regulatory Limitations.In some cases,particularly in heavily regulated sectors such as energy and trans-port,policymakers could adopt a
308、 thoughtful,risk-based governance framework for AI that ensures sufficient protections and safeguards while taking care not to stifle innovation.And,conversely,the lack of clear regulatory frameworks in other areas might also impede invest-ment.Infrastructure Challenges.Many high-emission sectors re
309、ly on legacy infrastructure that cannot readily support AI technologies.For example,outdated road net-works may struggle to accommodate smart traffic man-agement systems powered by AIor limited penetration of smart meters could hinder the at-scale deployment of AI-based energy efficiency solutions.I
310、nnovation Costs.Developing and implementing AI solutions often demands significant financial resources and time.Giving access to specific innovation platforms,such as high-performing computing infrastructureor to targeted innovation fundingcan remove significant barriers to innovation for the innova
311、tion ecosystem broadly,and particularly for small and medium-sized enterprises.Deployment Costs.In many areas,there may be sig-nificant entry or adoption costs,including capacity and capability needs,that pose challenges for companies,their stakeholders,and customers.For example,the adoption and imp
312、lementation at scale of precision agri-culture could be held back because of farmer concerns regarding the expense of installing the Internet of Things hardware needed to support it.To address these challenges,policymakers could activate the following five corresponding key levers:Create public/priv
313、ate partnerships to drive AI adop-tion in key sectors and applications Remove regulatory barriers to and provide regulatory support and clarity for AI adoption Accelerate infrastructure modernization and digitalization to enable AI use Support AI research and innovation in climate-relevant domains b
314、y involving both academia and the private sector Optimize incentives to drive AI adoption at scale for climate action priorities when existing mechanisms are insufficient(for example,public funding,tax credits,etc.)Nonetheless,these levers need to be tailored to the distinct challenges of each juris
315、diction and sector.These challenges are shaped by both global and context-dependent factors such as market intricacies,complexity,regulatory frameworks,and innovation dynamics.There-fore,policymakers could adopt a strategic approach by identifying the most effective measures to address these challen
316、ges and determining the appropriate level of intervention,whether through advocacy,incentives,or binding measures.They could also draw insights from best practices within each sector and jurisdiction,as well as across sectors and jurisdictions,in order to learn from one another.For example,consider
317、the implementation of AI in grid planning and management.AI solution developers may encounter a constraint that necessitates holding specific licenses to test or develop their solutions.In such circum-stances,policymakers might consider issuing regulatory waivers or establishing controlled testing e
318、nvironments.Similarly,in regions with limited smart meter adoption,policymakers could take action by allocating funding for infrastructure modernization or offering incentives to encourage electricity providers to replace aging meters.These measures would facilitate the integration of AI application
319、s designed to improve the energy efficiency of buildings.To underscore sectoral distinctions,in the following,we provide examples to illustrate how policymakers have begun implementing policies in sector-specific contexts.BOSTON CONSULTING GROUP GOOGLE 36Leaders could:Create public/private partnersh
320、ips to drive AI adoption within and across key sectors and applications.Examples:Buildings:DO IT SMARTera public-private partnership between the Tech-nical University of Cluj-Napoca,sfold University College,the Norwegian company NxTech,Romanias Alba Iulia municipality,and the NGO Center for the Stud
321、y of Democracyseeks to develop AI-driven energy efficiency solutions for public buildings in Romania and Norway.Energy:The US Department of Energys Princeton Plasma Physics Labora-tory is partnering with the Renaissance Fusion startup on AI-driven efforts to accelerate the development of carbon-free
322、 fusion energy.Transport:In Project Green Light,Google is collaborating with 12 cities including Manchester,Rio de Janeiro,Jakarta,and Abu Dhabi to reduce stop-and-start events through AI-supported traffic light management.Early indicators show a potential for up to a 30%reduction in stops,which cou
323、ld reduce emissions at intersections up to 10%.This could have a significant impact,as cars at city intersections generate 29 times more pollution than do cars on the open road.Remove regulatory barriers to and provide regulatory support and clarity for AI adoption.Examples:Electricity:Spain,Brazil,
324、and Australia have launched regulatory“sandboxes”for their electricity sectors that can offer waivers for climate-related pilots.In Australia,for example,grid participants can be granted exemptions from registering as a network service provider.Agriculture:Key agro-food stakeholders have co-signed a
325、 set of non-binding guidelines entitled the EU Code of Conduct on Agricultural Data Sharing.Similarly,The American Farm Bureau Federation has worked with stake-holders to establish the Privacy and Security Principles for Farm Data.Accelerate infrastructure moderniza-tion and digitalization to enable
326、 AI use.Examples:Buildings:In 2022,the UK government imposed new binding requirements that energy suppliers install smart meters in homes and small businesses.The target,currently under negotiation,aims for 80%smart meter coverage in homes and 73%in small businesses by 2025.Energy:The Modernization
327、Fund is a dedicated funding program to support 10 lower-income EU Member States in modernizing their energy systems and improving energy efficiency(for example,modernization of energy networks,district heating,etc.).Similarly,adopted in late 2022,the Digitaliz-ing the energy sector EU action plan,wh
328、ich is characterized by cybersecurity,efficiency,and sustainability,aims to cultivate a competitive marketplace for digital energy services and digital energy infrastructure.Support AI research and innovation in climate-relevant domains by involving both academia and the private sector.Examples:Ener
329、gy:In May 2023,the US Department of Energy announced a$40 mil-lion investment in 15 projects focused on developing high-performance,energy-efficient cooling solutions for data centers.Agriculture:The EUs Horizon 2020 program for research and innovation has allocated more than 200 million to support
330、the deployment of precision farming.37 ACCELERATING CLIMATE ACTION WITH AIGuide AI for ClimatePolicymakers play a crucial role in shaping the evolution of AInot just in accelerating its positive contributions,but also in minimizing its potential negative impacts.With regard to AI for climate,clear p
331、rinciples and guidelines are essential in two areas:first,maximizing the environmental friendliness of AI applications,and second,ensuring that AI applications respect the privacy and diversity of people and do not inadvertently exacerbate inequalities.We dis-cuss each in turn.Address Environmental
332、Impacts of AI OperationsAs discussed in the previous chapter,AI has its own environmental footprint.And while it has not yet been comprehensively measured,it is essential that we monitor and manage it.AI developers recognize the necessity of minimizing AIs environmental footprint,especially in the c
333、ontext of expanding the use of AIand are already working to make algorithms and data centers more energy efficient and to increase their commitment to renewables.For instance,in terms of clean energy utilization,several organizations such as the Clean Energy Buyers Association(CEBA)in the US,RE-Source in the EU,and the Asia Clean Energy Coalition(ACEC)in APAC(each with members including prominent