1 AI & THE SUSTAINABLE DEVELOPMENT GOALS: THE STATE OF PLAY

Global Goals Technology Forum 2 3

INTRODUCTION

In 2015, 193 countries agreed to the United Nations (UN) 2030 Agenda for Sustainable Development, which provides a shared blueprint for peace and prosperity for people and the planet, now and into the future. At its heart are the 17 Sustainable Development Goals (SDGs), which are an urgent call for action by all countries – developed and developing – in a global partnership. Achieving the SDGs is not just a moral imperative, but an economic one. While the world is making progress in some areas, we are falling behind in delivering the SDGs overall. We need all actors – businesses, governments, academia, multilateral institutions, NGOs, and others – to accelerate and scale their efforts to deliver the SDGs, using every tool at their disposal, including artificial intelligence (AI). In December 2017, 2030Vision 2030Vision AI & The Sustainable Development Goals: The State of Play published its first report, Uniting to “WE ARE AT A PIVOTAL Deliver Technology for the Global ABOUT 2030VISION Goals, which addressed the role MOMENT. AS ARTIFICIAL of digital technology – big data, Founded in 2017, 2030Vision INTELLIGENCE BECOMES robotics, internet of things, AI, and is a partnership of businesses, MORE WIDELY ADOPTED, other technologies – in achieving NGOs, and academia that aims to the SDGs. transform the use of technology to WE HAVE A TREMENDOUS support the delivery of the SDGs. OPPORTUNITY TO In this paper, we focus on AI for 2030Vision serves as a platform REEVALUATE WHAT WE’VE the SDGs. AI extends and amplifies to convene cross-sector leaders CREATED AND ENSURE the capacity of human beings to understand and solve complex, to raise awareness of the SDGs, to THAT AI IS DEVELOPED showcase thought-leadership on dynamic, and interconnected the role of technology in addressing COLLABORATIVELY AND systems challenges like the SDGs. the SDGs, and to stimulate APPLIED TO THE UN Our main objective was to survey partnerships for action. SUSTAINABLE DEVELOPMENT the landscape of research and GOALS TO ACHIEVE A BETTER initiatives on AI and the SDGs to identify key themes and questions AND MORE SUSTAINABLE in need of further exploration. We FUTURE FOR ALL.” also reviewed the state of AI and Tabitha Goldstaub the SDGs in two sectors – food and agriculture and healthcare – to Co-Founder at CognitionX & Chair understand if and how AI is being of the UK Government’s AI Council deployed to address the SDGs and the challenges and opportunities in doing so. 4 5

AI AND THE SUSTAINABLE DEVELOPMENT GOALS:

As illustrated by the quotes opposite, there are a variety of DEFINING ARTIFICIAL perspectives about AI and its 1 potential impact on sustainable INTELLIGENCE development. For some, AI is a Whether we know it or not, AI is job killer, a tool that will benefit THE now part of the fabric of our daily wealthier nations and citizens. lives. For example, AI helps power Worse, AI could pose existential predictive Google searches, Spotify threats, for example Stephen music recommendations, Waze Hawking warned that AI could driving directions, and Facebook lead to the end of humanity. POTENTIAL facial recognition. However, AI promises – and is There are a variety of definitions of starting to deliver – benefits AI, which we paraphrase as follows: across sectors and geographies, and holds great potential to help solve complex and interconnected AI IS AN FOR GOOD sustainable development challenges such as climate change, access to health care, OVERARCHING TERM and inequality. While 2030Vision “PERHAPS THE MOST recognizes the potential risks of FOR A COLLECTION IMPORTANT QUESTION AI, we believe AI will be a positive OF TECHNOLOGY “...CLIMATE CHANGE IS A WE HAVE LOOKED AT IS force for global sustainable development. Guided by the 2030Vision AI & The Sustainable Development Goals: The State of Play MASSIVE PROBLEM ACROSS WHETHER AI WILL POSE SDGs, we must now work together ALGORITHMS AND NEARLY EVERY SECTOR A THREAT – OR PROVIDE to accelerate and scale the AND MEASURE OF HUMAN NEW OPPORTUNITIES – FOR development and use of AI. APPROACHES THAT DEVELOPMENT. TO ADDRESS DEVELOPING REGIONS SUCH IT AT THE SPEED AND SCALE AS AFRICA. OPTIMISTS SAY ALLOW MACHINES TO THAT CURRENT CONDITIONS THAT SUCH PLACES COULD PERFORM HUMAN- REQUIRE, WE’LL NEED USE RAPIDLY ADVANCING TO TAKE A MORE DATA- AI SYSTEMS TO BOOST LIKE COGNITIVE DRIVEN APPROACH – ONE PRODUCTIVITY AND THAT HARNESSES THE FULL LEAPFROG AHEAD. BUT I AM FUNCTIONS SUCH POWER OF ARTIFICIAL BECOMING INCREASINGLY INTELLIGENCE AND OTHER CONCERNED THAT AI AS REASONING AND ADVANCED TECHNOLOGIES WILL, IN FACT, BLOCK THE LEARNING. TO ACCELERATE DISCOVERY TRADITIONAL GROWTH PATH AND INNOVATION AT A BY REPLACING LOW-WAGE AI is also being used to address TRULY PLANETARY SCALE.” JOBS WITH ROBOTS.” sustainable development challenges: improving the Lucas Joppa Ian Goldin diagnosis of various diseases, Chief Environmental Scientist, Professor of Globalisation and fighting wildlife poaching, and Microsoft Development, Oxford University improving crop yields to name a few. Other examples of SDG- related applications of AI are included throughout this report. 6 7

PROGRESS ON THE SDGS

AI AND THE SUSTAINABLE In its most recent progress report However, progress has been on the SDGs, the UN cited success insufficient in other areas, DEVELOPMENT GOALS: in some areas, for example: for example:

GOAL 1: NO POVERTY GOAL 4: QUALITY GOAL 14: LIFE BELOW EDUCATION WATER 2 Since 1990, the percentage of people living in extreme poverty GOAL 15: LIFE ON LAND CAPTURING (living on less than $1.90 per 617m UN scientists recently day) has declined from children of primary and warned that secondary school age do not 33% TO 9% meet minimum proficiency in million mathematics and reading. 1 THE STATE out of the world’s eight million GOAL 3: GOOD HEALTH species are at risk of extinction AND WELL-BEING GOAL 5: GENDER EQUALITY due to human activities. Since 2000, maternal Globally, women represent just OF PLAY mortality has declined by 23% OF SEATS 37% in single or lower houses of national parliaments. and under-five mortality has declined by Organizations such as McKinsey Over the last two and PwC have published reports 2030Vision AI & The Sustainable Development Goals: The State of Play GOAL 13: CLIMATE ACTION years, we have seen a with a wealth of information – 47% including use cases, economic In May 2019, the concentration considerable increase impacts, risks, enabling factors, of CO2 in the atmosphere and more. We’ve seen a number reached in discussion about GOAL 7: AFFORDABLE With 10 years of non-profit organizations and AND CLEAN ENERGY remaining to achieve the role that AI can events emerge at the intersection 415ppm play in sustainable of AI and the SDGs, for example the ambitions of the AI for Good Foundation and 87% for the first time in human development issues. ITU’s AI for Good Summit. And history, and it continues to rise. the 2030 agenda, every day there are multiple media of the global population has pieces on the role of AI and some access to electricity, up from it is up to all of sustainability topic. 78% in 2000. us – businesses, governments, academia, multilateral institutions, NGOs, and others – to strengthen and quicken our efforts to build a better future for everyone. 8 9

DIAGNOSIS AND TREATMENT FINANCIAL CRIME A growing number of researchers Financial institutions are using AI are exploring the use of AI in to analyze ever larger and more the diagnosis and treatment of varied data sets to spot fraud and various diseases and the design of money laundering. Banks spend health interventions. For example, £5 billion per year on fighting researchers in Germany, the US, financial crime in the UK, and AI and France found that AI is on can reduce the cost and increase par or better than dermatologists the effectiveness of such efforts. at diagnosing skin cancer. Researchers from IBM and New York University recently published a paper on how AI can be used to detect glaucoma, the second leading cause of blindness in the world. Such AI-based systems hold THE HIGH COST OF ILLEGAL promise to make diagnostic and TRADE IN WILDLIFE Given how much positive work has been other health services more readily Illegal and unsustainable available in regions that have a lack completed and is ongoing in this space, wildlife trade is a threat to the of health care workers. we reviewed this work to understand: existence of certain species COMBATTING FRAUD AND and human livelihoods. The UN OTHER FINANCIAL CRIMES estimates that illegal wildlife What ground has been covered with respect to AI and the SDGs? trade worldwide is worth $8 HSBC is using AI to help spot What questions are in need of further exploration? billion to $10 billion per year, money laundering, fraud, and the value of the ivory trade and terrorist funding. AI What frameworks are useful for connecting AI to the SDGs? alone is roughly $1 billion. Non- enables the bank to screen the vast amounts of data it Can AI be used to better understand the interaction between the SDGs? profit RESOLVE is deploying cameras enhanced with image holds on customers and their How is AI being used to address the SDGs in both the food and agriculture processing and deep neural transactions against publicly and health sectors? network algorithms across a available data, in the search for suspicious activity. What should different stakeholders do to foster the use of AI for the SDGs? REDUCING MATERNAL range of reserves in Africa to MORTALITY IN UTTAR allow rangers to detect and PRADESH, INDIA respond to poachers in near real-time. THE COMMON THREAD From our research across a variety of reports, organizations, articles, Researchers at the University ACROSS THESE EXAMPLES and other sources, we identified a number of consistent themes. of Sussex have partnered with IS THAT AI BRINGS NEW the Surgo Foundation, the Bill CAPABILITIES AND and Melinda Gates Foundation, URBAN PLANNING EFFICIENCIES TO ADDRESS GNS Healthcare, and the COMPLEX GLOBAL AI is increasingly used for urban University of Manitoba to better CHALLENGES SUCH AS planning and management, and to 2030Vision AI & The Sustainable Development Goals: The State of Play understand the causes of high THE SDGS. help build smart and resilient cities maternal mortality rates in around the world. Uttar Pradesh, India. Through the application of machine AI extends and amplifies our capacity to learning algorithms to data about mothers’ perceptions solve complex challenges like the SDGs. and behaviors, the partners hope to design evidence based, This is beneficial for picking up the pace effective, and acceptable towards the SDGs. interventions that will reduce the number of deaths in this community. The various forms of AI – – have enormous potential to machine learning, natural address a wide range of sustainable WIND AND SOLAR TOWARDS RESPONSIVE AND language processing, predictive development issues. In combination AI can be applied to the design RESILIENT URBAN SYSTEMS WILDLIFE CONSERVATION analytics, and more – can bring with other technologies (e.g. and operation of wind and solar great efficiencies, speed, and sensors), tools (e.g. data analytics), PetaBencana.id combines AI, farms to make them more efficient, AI is being used to conserve intelligence to tasks and processes. and infrastructure (e.g. data citizen reporting via social for example orienting turbine wildlife. For example, Wildbook is centers), AI enables us to access, media, government flood alerts, The capabilities of AI – including heads to capture a greater portion a software that allows researchers analyze, and act on ever-growing and hydraulic sensor data to automating routine tasks, analyzing of incoming wind. Google and to more efficiently assess animal sets of data. help megacities in South and big data, and bringing intelligence DeepMind are deploying machine populations by using AI to help Southeast Asia respond to and learning to various processes To illustrate: learning to manage 700 MW of identify individual animals from and manage emergencies and wind power capacity in the US, photos by their unique features disasters. resulting in a 20% increase in value. and markings. 10 11

The SDGs are dynamic, complex, and While AI promises benefits for the SDGs, interconnected – AI can help us better it will only be as positive and impactful as understand and achieve them. designed and utilized.

The SDGs are inherently a complex, The researchers also suggested Much has been written about China is using facial recognition Bias also shows up in realms such systems problem that will require ANALYZING TRADE-OFFS it may be possible to identify how AI reflects the experiences, and surveillance cameras to identify as financial services, where AI is collaboration across subject BETWEEN SDG’S IN URBAN appropriate data and build intentions, and biases of its and in some cases detain Uighurs, being used for credit decisions matters (climate, water, poverty, ENVIRONMENTS AI models that improve designers, and this is certainly true a Muslim minority group, in the (for purchasing property, starting etc.), sectors (private, civil society, In 2017/18, the Sustainability our understanding of the in the context of the SDGs. western region of the country. companies, etc.). While AI allows government, etc.), and industries interconnections between the Amazon and others have faced financial institutions to evaluate Research Programme at the An often-cited example is how facial (finance, healthcare, food, etc.) SDGs and their 169 targets, criticism for holding discussions more data points, if the algorithms University of Sussex conducted recognition technology can be to understand and solve. We will whether they are synergies or with US customs officials about are based on biased data (i.e. a research project to better more accurate for white men than need to avoid the temptation trade-offs. To illustrate, a billion using facial recognition technology people of color have historically understand the trade-offs for men of other ethnicities and to implement fixes to individual people globally do not have to identify immigrants crossing been denied credit), then the AI between development, women – a result of machines being goals without addressing the access to electricity in their the US / Mexico border. Such model will be biased. food security, and poverty trained on images of white males connections between goals. homes, and three billion rely on applications of AI can have alleviation. Focused on Wuhan, (presumably by white males). This There are a number of organizations dirty fuels such as charcoal and profound human rights implications, Academic institutions, China and Varanasi, India, has serious implications for privacy working to address bias in AI, for animal waste for cooking. This lack which is one reason why some especially those that encourage the researchers applied deep and security, for example African example AI4All, which aims to of access to clean energy (Goal 7) technology executives are calling interdisciplinary working, are learning techniques to map American males may be improperly increase diversity and inclusion has negative impacts on education for clearer regulatory guidance on effective platforms for this type and better understand trade- identified as crime suspects via in AI, Humans for AI (similar (Goal 4), economic status (Goals the technology. of approach. For example, in offs in peri-urban agriculture poorly trained AI. This challenge mission), and the Partnership on AI. 1 and 10), and gender equality 2018, the Sussex Sustainability systems, as well as inform urban is one reason why the city of San Companies such as Google (AI hub (Goal 5). In specific contexts such Research Programme convened sustainability policy. Francisco recently became the in Ghana) and Microsoft (4Afrika) as this, the researchers suggested a workshop with diverse experts first US city to ban the use of facial “FACIAL RECOGNITION are investing in AI capabilities away it may be possible to use AI to and stakeholders that explored recognition technology. from their home markets to ensure improve our ability to understand TECHNOLOGY RAISES ISSUES how targeting efforts around the that local experiences are reflected In conversation, Sussex researchers and prioritize our efforts, THAT GO TO THE HEART interactions between the SDGs, in the design of AI. emphasized the following elements ultimately helping to accelerate OF FUNDAMENTAL HUMAN whether synergies or trade-offs, LACK OF DIVERSITY as being key to establishing an progress towards the SDGs. RIGHTS PROTECTIONS LIKE will be critical for driving progress effective AI approach for the PERPETUATES BIAS IN AI forward. SDGs, grounded in interdisciplinary PRIVACY AND FREEDOM OF thinking: EXPRESSION. THESE ISSUES SDG-related problems must HEIGHTEN RESPONSIBILITY be articulated and rephrased FOR TECH COMPANIES THAT such that they can be solved 80% CREATE THESE PRODUCTS. by AI, for example drought can 2030Vision AI & The Sustainable Development Goals: The State of Play IN OUR VIEW, THEY ALSO be rephrased as a problem of CALL FOR THOUGHTFUL water management; GOVERNMENT REGULATION Experts from different areas According to an April 2019 AND FOR THE DEVELOPMENT should agree a shared language report from the AI Now OF NORMS AROUND and understanding of research Institute, 80% of AI professors opportunities; ACCEPTABLE USES.” are male, and women comprise Steps should be taken to ensure only 15% of AI research staff at Brad Smith that AI is applied appropriately Facebook and 10% at Google. President, Microsoft, July 2018 in specific contexts; Provide robust interpretations of research results, built on an understanding of the technology as well as the challenge, area or context in which it is applied. 12 13

More work is needed to bring together the current applications of AI, map them against the

SDGs, and assess how this work can be scaled. TRILLIONS OF VALUE IN AI FOR ENVIRONMENTAL Researchers examining the A report from Google and the APPLICATIONS connection between AI and the Ellen MacArthur Foundation According to PwC, by 2030, SDGs are fortunate to have a has examples specific to AI and the use of AI for environmental growing number of use cases to the circular economy, while the applications could contribute draw upon. International Telecommunications Looking across these reports and up to Union (ITU) has started an open For their 2018 report Applying AI for others, we are keen to explore a source repository of AI and SDG Social Good, McKinsey developed number of areas. For example, use cases. Across these resources a library of over 160 instances in looking at the McKinsey use case $5.2tn and others, there are plenty framework (Figure 1), it appears which AI is being used for social to the global economy, and of examples that illustrate the that AI is more beneficial or useful impact, and created a number of greenhouse gas emissions potential of AI and the SDGs. for certain SDGs – good health useful frameworks, including the one reductions of in Figure 1 that categorizes the use and well-being (Goal 3) has 29 cases by Goal. As illustrated in the use cases, while life below water FIGURE 1: Mapping AI Use Cases for the SDGs chart, there is an imbalance across (Goal 14) has one use case. While Gt CO2e 2030Vision AI & The Sustainable Development Goals: The State of Play 2.4 Source: McKinsey Global Institute, Applying AI for Social Good, December 2018 the SDGs, which is something we this pattern may not hold for a discuss below. broader sample size, we are keen to understand if AI is inherently AI could also create A report from PwC and the World MICROSOFT AI FOR GOOD better suited for certain SDGs, or Economic Forum, Fourth Industrial if other factors are at work such as million Revolution for the Earth, is full Through its AI for Good capital investment or availability of 38.2 initiative, 2030Vision partner of use cases for AI and various academic research. net new jobs globally. sustainability topics, primarily Microsoft provides technology, environmental. It describes how resources, and expertise Lastly, we recognize that the field of AI for the SDGs is new and we AI can be deployed across six to address environmental, MAPPING DATA are eager to see more work on the environmental impact areas – accessibility, and humanitarian AVAILABILITY ACROSS cumulative benefits of AI and the climate change, biodiversity and challenges around the world. THE SDGS conservation, healthy oceans, water The website includes a variety SDGs. PwC has done valuable work security, clean air, and weather and of use cases, for example to estimate the potential climate Researchers from Sussex disaster resilience – and illustrates Seeing AI, a free app that change and economic benefits offered that a systematic each with use cases. helps the blind and low of applying AI to environmental mapping of data availability vision community access the applications, while Google and across the 17 SDGs (and visual world, and SilvaTerra, the Ellen MacArthur Foundation the 230 indicators) would a satellite and machine have quantified the financial and be a beneficial first step in learning system that allows environmental benefits of applying determining how AI can be forest managers to take more AI, with a circular economy lens, to applied to the SDGs. This would accurate and efficient surveys the food and electronics sectors. be a significant undertaking, of forests. Microsoft also has Yet, more work is needed on this as and one that academic an online map where users can AI matures, and the use cases get institutions would likely be well search over 300 projects. more significant (many to date are suited to address. pilot or small scale). 14 AI’s benefits will flow to the “have’s” and AI will result in economic dislocation – 15 not the “have not’s” – unless it is designed we must manage it. and deployed deliberately. PwC and Microsoft also found Much has been written about the that AI applied to four sectors potential economic dislocation and As mentioned above, AI is a tool – agriculture, water, energy and job loss to which AI could contribute, that takes on the biases, intentions, SDGS AND NATIONAL AI transport – will result in a net gain particularly to jobs that are based and objectives of its developers – it STRATEGIES of up to 38 million jobs globally. on routine tasks. For example, the is not inherently good or bad. A number of countries have Brookings Institute found that 25% However, these added jobs will be Much of the AI that we see in developed national AI strategies of jobs in the US face high exposure The potential to “make the rich created outside of the four sectors the world today tends to benefit or positions, yet the SDGs to automation in the coming decades. richer” is illustrated in PwC and studied (these four sectors will wealthier individuals and nations. are generally not included in McKinsey found that roughly Microsoft’s latest report, How AI actually experience a decline in Applications such as predictive such documents. For example, half of the activities people are can enable a Sustainable Future jobs). We presume the added jobs search (e.g. Google), facial China’s plan does not mention paid to do in the global workforce that finds that the economic and will come in higher-value areas, recognition for mobile device the SDGs or individual issues could be automated by adapting greenhouse gas reduction benefits for example AI programmers security (e.g. Apple iPhone), and like climate change, though currently available technology. These of AI applied to four sectors will be for autonomous vehicles, while image recognition and computer it does have a section on organizations and many others have greater in North America, Europe, jobs such as taxi drivers will vision for autonomous vehicles intelligent environmental written about this potential dislocation and China than for the Middle East diminish. Higher skilled jobs such tend to benefit those individuals, protection. The UK’s AI Sector and possible ways to address it. and North Africa, Sub-Saharan as technicians and managers will countries, or companies with access Deal and the US’s executive Africa, and Central and South increase, while agricultural and to technology and the resources to order on AI similarly do not America. Again, this may be in part unskilled jobs will decline. power it. mention the SDGs. Given the AUTOMATION’S IMPACT central role that governments because the former regions have ON EMPLOYMENT IN will play in delivering the SDGs, the enabling environments that SOUTHEAST ASIA ACCORDING TO PWC, it will be important for such AI allow them to reap the benefits FIGURE 3: AI Impact on Employment by Region and Skill Level of AI, but more attention to and Source: Microsoft & PwC UK, How AI can enable a Sustainable Future, April 2019 According to a 2016 ILO report, ALMOST strategies and plans to connect the dots between AI and the investment in the latter regions more than half of workers in is needed to avoid worsening five Southeast Asian nations SDGs. Impact by region inequalities. – Cambodia, Indonesia, the 70% Philippines, Thailand, and Vietnam – are at risk of losing OF THE GLOBAL ECONOMIC FIGURE 2: Economic and greenhouse gas reduction of AI their jobs due to automation IMPACT FROM AI WILL Source: Microsoft & PwC UK, How AI can enable a Sustainable Future, April 2019 in the next two decades, with ACCRUE TO CHINA AND those in the garment sector particularly vulnerable (many of NORTH AMERICA. whom are female).

This is not unexpected, given that The Brookings Institute highlighted wealthier nations and actors (e.g. the potential for AI to exacerbate companies and universities) have geographic divides, with more been first movers on AI. Countries economic gains and influence 2030Vision AI & The Sustainable Development Goals: The State of Play such as the US, China, Germany, the accruing to digitally-oriented UK, and France are out in front on metropolitan areas compared with Impact by skill level AI, and the application of AI is thus lower tech locales. With lower- and aimed at challenges these countries middle-wage jobs such as some face. Such countries also have the types of manufacturing being enabling environments – computing replaced with automation (enabled infrastructure, robust electricity by AI), regions risk falling behind. systems, better data sets, skilled This in turn could increase the degree workers (for development and use of political polarization we are seeing of AI), and so on. Oxford Insights around the world, that is reducing publishes an annual index on AI trust in institutions, and represents government readiness for countries “TRANSFER LEARNING” AS A – thus reducing the development an obstacle to achieving the shared members of the Organization MEANS TO DEMOCRATIZE AI costs. For example, a team vision of the SDGs by 2030. for Economic Co-operation and from Penn State University has Development (OECD), and the In discussing how to get AI into applied transfer learning to train top performing countries are the the places where it is needed a deep convolutional neural INVESTING IN AI STARTUPS wealthier ones. most, researchers from the network to identify different OUTSIDE OF MAJOR CITIES University of Sussex suggested types of cassava disease, using The Rise of the Rest® Seed “transfer learning” as one a dataset of images taken in the Fund is a $150 million fund that solution. Transfer learning is a field in Tanzania. The best model invests in startups (including AI) machine learning method where achieved an overall accuracy outside of large US coastal cities a model trained on one problem of 93% for data not used in the like New York San Francisco with is reused as the starting point training process. for a model on another problem the goal of catalyzing economic growth in these regions. 16 17

The need to anticipate and manage trade-offs DATA PRIVACY and unintended consequences in the SDGs. While AI is fueled by data, there is the critical need to protect consumer privacy – particularly While AI can benefit the SDGs, it Similar unintended consequences around sensitive topics such as can also negate progress. might arise as AI enables the health and finances. Interestingly, deployment of autonomous For example, while AI can help The challenges and well-intentioned government vehicles, potentially exacerbating INFRASTRUCTURE AND make solar and wind power more privacy regulations could hinder congestion and crowding out RELATED TECHNOLOGY efficient and productive, it brings enablers are multiple the development of AI. For investment in public transportation. the same supply benefits to oil example, the Brookings Institution The computational needs for AI applied to farming techniques – and solvable. and gas production. AI can make it noted that the European Union AI, and the needed storage for could allow big agriculture to get easier to identify and map oil and General Data Protection Regulation the underlying data, require even bigger at the expense of gas reserves, thus lowering the Other authors have articulated the (GDPR) prohibits the transfer of robust computing and electricity smallholder farmers. And while AI cost to access them. The Brookings key challenges and enablers for AI personal data to countries with infrastructure. Also, AI typically might allow companies to better Institution suggests that AI might including: inadequate protections, and works in concert with other digital manage supply chains that are ever benefit oil and gas development generally limits how data can be technology such as sensors and expanding geographically, this may more so than renewable energy. used. This limits the potential for cameras. As mentioned above, run counter to the shorter value ACCESS TO HIGH QUALITY DATA this data to be used to train AI. countries that would benefit most chains that are needed to deliver from the application of AI to responsible consumption and AI is fundamentally dependent EASIER FOSSIL FUEL address the SDGs may not have production (Goal 12). on high quality data, for example DEVELOPMENT THROUGH AI image recognition that helps “THERE’S GROWING the necessary infrastructure and In the absence of an enabling supporting technology, meaning Through AI and other digital identify pest damage in crops is RECOGNITION OF A environment that is pro-SDGs – investment will be needed in technology, ExxonMobil projects reliant on having clear images of PARADOX. IF WE FOCUS including policies, trade rules, and these areas. that it will generate billions of the damage. The image recognition financial mechanisms – AI will be ONLY ON PRIVACY, THEN WE dollars in value over the next that supports autonomous vehicles used for whatever outcome its DON’T INNOVATE AND FIND decade and expand production is based on the technology being designers and users have in mind. In AI IN THE PALM OF YOUR by as much as 50,000 oil- fed thousands of images that WAYS TO USE THE DATA. the example above of ExxonMobil HAND equivalent barrels a day by 2025 enable the car to distinguish YOU HAVE TO BALANCE THE using AI and other technology to in the Permian Basin, USA. between objects. Good data may RISK OF MISUSE AGAINST Google is working to embed drill for more oil and gas, if we are not exist for certain SDGs, thus machine-learning models to achieve Goal 13 and the targets THE RISK OF MISSED USE. limiting the capabilities of AI, directly on remote devices. set forth in the Paris Agreement, THE OPPORTUNITY COST and concerns about security and This has the potential to reduce we need robust policies that limit privacy may limit the extent to OF NOT USING THIS DATA reliance on central data centers, how much fossil fuels we combust which data can be collected. FOR THE PUBLIC GOOD IS making the technology more – regardless of how easily AI might REALLY QUITE HIGH. WE ARE useful in rural regions, and to allow us to access them. We also INCURRING IT EVERY DAY, protect data privacy. need more AI developers and users IMPROVING ACCESS TO to be aware of the imperative to DATA FOR THE SDGS AND EVERYWHERE, LARGELY meet the SDGs so that they are UNBEKNOWNST TO ANYONE.”

2030Vision AI & The Sustainable Development Goals: The State of Play a starting point for the creation The Global Partnership for ELEVATING SKILLS IN AI of new AI, not issues that get Sustainable Development Robert Kirkpatrick DEVELOPMENT AND USE addressed after the fact. Data is an independent, multi- Director, UN Global Pulse stakeholder network that aims Training significantly more people to improve data access and in the development and use of AI quality in support of advancing is essential for its further evolution. sustainable development. We need to ensure we are training and supporting people to apply AI For example, the Global EXPLAINABILITY Partnership recently announced to the SDGs, as these fields are less a collaborative initiative with the The ability to clearly convey what, lucrative than more commercially- Government of Kenya and other why, and how AI makes decisions oriented sectors (e.g. defense and partners to mine two decades and achieves its objectives – medicine). We will also need to of satellite imagery for insights opening up the “black box” – is ensure that we train AI specialists on farming strategies to help essential to building trust with in the regions in most need of cope with climate stresses. the public. Such transparency is support to address the SDGs. particularly critical in matters of human health and well-being, and in other areas such as criminal INVESTING IN AI IN AFRICA justice. Without this trust and In May, Microsoft announced it acceptance, the application of AI would invest $100 million in its across the SDGs will be hindered. Africa Development Centres and The AI community might take hire 500 full-time developers lessons from other technologies by 2023 to work on AI, machine that are beneficial yet have faced learning, and mixed reality public opposition (e.g. genetically innovation. modified foods). 18 19

AI AND THE SUSTAINABLE DEVELOPMENT GOALS: 3 THE SECTORAL FOOD AND AGRICULTURE Building a food system that FIGURE 4: Food and Agriculture can meet the nutritional needs Connections to the SDGs of nine billion people by 2050, while staying within the carrying VIEW capacity of the planet, will take an enormous effort. Such a food system is needed to deliver a number of SDGs, as Figure 4 illustrates. For example, PwC examined A range of the greenhouse gas reduction Given these interconnected organizations have benefits of AI in energy, transport, challenges, AI promises to help agriculture, and water, while make food production and 2030Vision AI & The Sustainable Development Goals: The State of Play explored how AI can Google and the Ellen MacArthur consumption more efficient drive sustainable Foundation looked at how AI and sustainable. We offer a few can enable the circular economy examples below. outcomes in different in food and agriculture and electronics. industries. PRECISION / SMART Given the growing interest in AGRICULTURE companies using the SDGs as the basis for their sustainability efforts, AI is increasingly being deployed and the proliferation of AI for a with other digital technology growing universe of commercial to enable more efficient food purposes, we were keen to explore production (i.e. more food with how companies are linking the fewer inputs). AI helps farmers two. In our first report, we looked better predict weather patterns, broadly at the role of digital adjust planting based on supply technology and the SDGs. In this and demand, fight pests, and report, we specifically look at how optimize inputs – water, fertilizer, two sectors – food and agriculture pesticides, etc. AI can be used to and healthcare – are starting to analyze chemical and biological use AI to address the SDGs. Our systems, and increase efficiency intent is to work with our partners in plant breeding, biotechnology, to explore other sectors in future and agrochemical discovery. With research. the growing amount of data being generated on farms, AI will be essential for farmers to utilize this data to make better decisions. 20 21

The use of AI in food and ONE ESTIMATE IS THAT BY EMPOWERING FARMERS BRINGING TRANSPARENCY IKEA REDUCING FOOD agriculture is still in its early days, PARTNERING TO SOLVE THE 2050, THE TYPICAL FARM THROUGH AI AND SECURITY TO FOOD WASTE WITH AI but holds tremendous potential for AGRICULTURAL DATA GAP SUPPLY CHAINS WILL CREATE IBM’s Watson Decision Platform Winnow Vision deploys AI delivering commercial benefits and Noting that many governments, for Agriculture leverages AI ImpactVision has developed a and computer vision to help addressing the SDGs. To realize food companies, and farmers to help farmers make more platform that provides real- restaurants and food service these aims, a number of barriers have incomplete agricultural 4.1million informed choices about time insights into the quality companies to better understand need to be overcome: information as they make policy their crops. For example, and characteristics of food the volume and economic value While larger food and and investment decisions, one application uses big using image recognition and of food waste. The camera sits DATA POINTS EVERY DAY. agricultural companies are the Global Partnership for data and machine learning predictive learning. It uses above a waste bin and through starting to deploy AI, we need Sustainable Development Data to predict crop yields two to hyperspectral sensors to take image recognition, companies to improve the understanding is currently developing the three months out with limited images of foods and uses are made aware of the value DEPLOYING AI TO COMBAT of AI and its benefits and 50 X 2030 Initiative, which data and computing power classification software and of food waste. WinnowVision CROP PESTS AND DISEASE risks across the value chain, aims to significantly improve requirements. Another allows algorithms to evaluate different has been installed in over 20 especially with small and mid- data quality and availability by According to the FAO, between farmers to model the probability characteristics (e.g. tenderness IKEA stores across the UK and sized companies and farmers. conducting regular surveys of 20 to 40% of crop yields are of disease and pests. Both of of meat, existence of foreign Ireland. This supports IKEA’s We need to do this in a way farming households in 50 low destroyed each year by pests, these applications help farmers objects). goal to reduce food waste by that respects the experiences and lower income countries and this could worsen due to improve profitability and a 50% by 2020. and values of these small and by 2030 and sharing the data climate change. A number of range of sustainability factors mid-sized players. widely. organizations are using AI- (e.g. soil health). In May, IBM based systems to fight pests, announced an expansion of the REDUCING FOOD WASTE In agricultural settings, data including Farmwave, which NEW PRODUCT DEVELOPMENT platform, which now includes In our first report, we noted that availability and quality may be uses machine learning to help models for corn, soy, sorghum, poor, as may be the necessary farmers identify pest damage roughly one third of all food is AI is being deployed in the potato, and more. wasted, representing 8% of global development of more sustainable computing infrastructure and disease, and PlantVillage to store data and run AI which has developed Nuru, an greenhouse gas emissions and up products, for example alternative to $405 billion in economic losses protein companies such as NotCo algorithms. As discussed AI platform that allows African above, there are promising farmers to identify disease in SUPPLY CHAIN OPTIMIZATION each year. AI can be deployed and Impossible Foods are using AI across the food and agriculture to identify the right combination new approaches that could cassava, maize, potatoes, and overcome this barrier, for wheat. AI brings efficiencies to global value chain to combat food waste, of ingredients that mimic animal food supply chains, and enables from reducing crop loss at the products. example Google’s work on better forecasting of consumer and farm level, to improving storage, building AI capabilities into retailer demand via historical and distribution, and processing of fresh mobile devices. real-time data (which ultimately food, to reducing food waste at Realizing AI’s full potential will means less food waste). AI can also retail and with consumers. Google require players from across be used to map global food supply and the Ellen MacArthur Foundation the food and agriculture value chains and help optimize them to found that designing out food chain to share data. This will be source and consume ingredients waste could deliver up to $127 challenge for a host of reasons locally where appropriate, and to billion of economic value in 2030. including competition and process food as close as possible mistrust of “big ag.” We will

2030Vision AI & The Sustainable Development Goals: The State of Play to where it is grown and consumed. need to develop the necessary legal frameworks that allow for this, and consider approaches such as making SDG- beneficial data public goods (a topic raised by the Sussex researchers). 22 23

HEALTHCARE

The successful application of AI FIGURE 5: Healthcare DIAGNOSING AND TREATING IMPROVING FETAL HEALTH MONITORING DISEASE ENHANCING MEDICAL to the health sector depends on Connections to the SDGs DISEASE WITH AI OUTBREAKS RESEARCH many factors: AI is being used to augment, GE’s AI-based SonoCNS HealthMap is a website and AI, and in particular machine High-quality datasets can be and in some cases improve, the tool allows gynecologists to mobile app that collates various learning, can lead to a drastic difficult and time-consuming ability of healthcare professionals take better measurements data sources to monitor disease reduction of costs and time to collect, especially in to diagnose and treat diseases of the fetal brain and detect outbreaks and emerging needed for drug discovery and developing countries. Yet they such as cancer and glaucoma. complications earlier. It also public health threats. The site development, through rapid are necessary to train machine- Many of these cases entail using enables gynecologists to delivers real-time intelligence analysis of health data; it can learning algorithms to identify machine learning to diagnose obviate mundane tasks and on a broad range of emerging also support the analysis of drug risk factors, diagnose diseases, disease, for example Google’s work spend more time focusing on infectious diseases for various efficiency in different contexts. and automate planning and on detecting lung cancer. While patients and less on working audiences including local health AI is also used to research the scheduling. human involvement in critical ultrasound machines. departments, governments, and diseases themselves – to find healthcare decisions will likely international travelers. information in unstructured AI applications depend on the always be necessary, AI brings medical literature to support availability of strong electronic efficiency and allows doctors to hypotheses, helping in the health record systems, which spend more time on higher value MODELLING AND PREDICTING discovery of new medical insights. is a challenge considering the activities. AI could be particularly EPIDEMICS AND CHRONIC IMPROVING THE PROVISION diversity of health records beneficial in regions that lack DISEASES OF PRIMARY HEALTHCARE and languages in the world. health care professionals for SERVICES AI BRINGS POWER AND Through AI, it is possible to For example, it is a challenge medical tasks that machines might EFFICIENCY TO DRUG monitor patient populations and Through advances in mobile to adapt natural language perform as well or better than 2030Vision AI & The Sustainable Development Goals: The State of Play DISCOVERY medical data to predict outbreaks connectivity, digitized health processing systems in places humans (e.g. reading CT scans). where health records are of infectious diseases and public systems, and resource allocation A number of companies have handwritten in local languages. health emergencies. Through data, AI is increasingly being developed AI solutions for Delivering Goal 3 – ensuring healthy modelling cause-and-effect used to optimize the allocation drug discovery. For example, While internet connectivity lives and promoting well-being relationships, this can allow public of basic health services and the PrecisionLife has developed is improving throughout the for all at all ages – will require a health agencies to investigate organization of administrative a platform that enables world, many regions lack access healthcare system that is accessible, and mitigate in real-time the tasks. For example, AI planning can faster and deeper analysis of to the bandwidth necessary effective, and affordable. Good progression of epidemics. AI can improve the scheduling of health large genomic, clinical, and to upload large datasets to health and well-being are connected also be used to tackle more chronic visitors, while natural language phenotypic datasets. This the cloud. Some applications, to a range of other SDGs, as Figure diseases, for example using big processing can be used to simplify allows researchers to identify including mHealth tools, have 5 illustrates. data and satellite images to target clinical documentation and enable and validate multiple novel the ability to work offline and As discussed in section 2, AI is being regions where people are most voice-to-text dictation. This can pathways, targets, and drug sync with remote databases deployed in a growing number vulnerable to dying from diseases be valuable in developed countries compounds involved in disease when the connection is of ways in healthcare, often with or health problems. where home care is increasing as in weeks not years using clinical sufficient. In addition, storing positive implications for the SDGs. a result of an ageing population, population datasets. This data locally requires strong as well as in developing countries, provides enormous productivity local IT infrastructure. where healthcare services may be and innovation gain for delivered door-to-door. biopharmaceutical research. Improved diagnosis does not guarantee access to quality treatment options, especially in hard-to-reach areas: remote diagnostics applications may help to identify diseases, but the focus should also be on how to deliver treatment following the confirmation of disease. 24 25

BUSINESSES, AND OTHER assessing the potential benefits DEVELOPERS AND USERS and risks of AI applications – both striving to amplify the benefits Although a growing number of (e.g. improving renewable energy CALL TO ACTION: companies have committed to systems) and minimize the risks addressing the SDGs, few to date (e.g. less human surveillance). are comprehensively looking Organizations developing AI across the SDGs to determine should also consider how models how AI can help advance can be applied to different progress towards them. contexts (e.g. less developed ACADEMIA MAXIMIZING nations) or challenges (e.g. Businesses and other 4 organizations using AI must applying pre-trained AI models Given the wide-ranging engage more in collaborative to new data sets). These efforts nature of the SDGs, academic multi-stakeholder efforts to would help stakeholders with a institutions are well placed to THE POTENTIAL identify opportunities to deploy good understanding of particular bring researchers from across AI to address the SDGs. issues but with access to fewer disciplines to explore how AI resources (e.g. NGOs) ensure the can be used to address the Meanwhile, organizations benefits of AI are more equitably SDGs, and the interconnections developing AI solutions should shared. across them. OF AI FOR THE use the SDGs as a filter for Universities could also play a critical role in educating future generations of AI developers and users. Integrating the SDGs SDGS into curricula could shape how well the SDGs are addressed by AI as it evolves. Academia may also have the ability to explore AI use cases that are further out in time, or that are less commercial in nature. Given the ambition, complexity, “WHAT’S CLEAR IS THAT

2030Vision AI & The Sustainable Development Goals: The State of Play and urgency of the SDGs, we need NO ONE NATION, NO ONE MULTILATERAL GOVERNMENTS AND unprecedented collaboration ORGANIZATION, NO ONE and action from businesses, INSTITUTIONS POLICY MAKERS governments, academia, COMPANY, AND NO ONE Organizations such as the UN, For AI to be trusted by the multilateral institutions, NGOs, and COMMUNITY CAN MEET THESE Organization for Economic public, and to help scale others to engage more with these CHALLENGES ALONE. THE Co-operation and Development efforts for achieving the SDGs, challenges to deliver the world we PATH TO A TRANSFORMATIVE (OECD), and development governments must set the want by 2030. BUT ALSO A SAFE, TRUSTED banks can play a unique right policies for AI’s accepted In itself, AI is a neutral tool that AND INCLUSIVE AI WILL role given their institutional development and use. reflects the intentions, experiences, missions that address SDG- REQUIRE UNPRECEDENTED Tools such as image recognition and biases of the people who related topics and their can advance progress towards the design and use it. Given the moral COLLABORATION BETWEEN frequent access to data on SDGs (e.g. Goal 2 Zero Hunger, and commercial imperative, the GOVERNMENT, INDUSTRY, sustainable development by improving food security) and SDGs must be central to how AI ACADEMIA AND CIVIL issues. impede it (e.g. discriminating is designed from the start. Each SOCIETY.” Such organizations may be well against certain demographics). stakeholder group brings unique placed to be central repositories They can also help set rules perspectives and capabilities to Houlin Zhao for the SDG-related data that AI around data governance, this collaboration. Secretary-General of International developers and users need. including for example, ownership Telecommunications Union (ITU) and access. Lastly, national governments are the signatories to the SDGs, and some have developed national AI strategies. The SDGs should be explicitly addressed in such strategies. 26 27

WE WILL NEED UNPRECEDENTED COLLABORATION TO LEVERAGE THE POWER OF AI TO HELP ACHIEVE THE SDGS. 2030VISION IS PLEASED TO SHARE OUR NEW PAPER AS A MEANS TO HIGHLIGHT THE IMPERATIVES AND OPPORTUNITIES FOR AI TO SUPPORT A MORE PROSPEROUS AND SUSTAINABLE FUTURE 2030Vision AI & The Sustainable Development Goals: The State of Play

2030Vision Global Goals Technology Forum Publication Date: June 2019 Publishers: SustainAbility

Designed by: Wolf&Player | wolfandplayer.com SustainAbility is a think tank and strategic advisory firm which for 30 years has worked to inspire and enable business to lead the way to a sustainable economy. SustainAbility 24 Greville St, London EC1N 8SS sustainability.com PARTNERS

2030Vision is a partnership that connects businesses, NGOs and governments with the technology and expertise they need to realize the Goals. Founded and chaired by Arm, it is a partnership with UN Global Compact, UNICEF, The New York Academy of Sciences, SustainAbility, Fauna & Flora International, Centre for Global Equality, Cambridge Judge Business School, British Council and Be He@lthy Be Mobile and Project Everyone. Our ambition is to transform the use of technology so that all people can live their lives to their full potential by 2030.

2030vision.com