Digitization, Statistics, and AI: A path towards well-informed decision-making (HLPF 2026 Side Event)

Integrated, analyzed, disseminated, and used. These developments create unprecedented opportunities to close data gaps and generate new insights, but also raise fundamental questions about quality, ethics, transparency, trust, and governance. Take, for example, the rapid rise of artificial intelligence. This is a profound technological shift, redefining how humans and machines complement each other. Data is at the heart of this transformation, but nevertheless, this evolution is taking place in an environment saturated with unreliable, sensational, and easily accessible information, increasing risks such as misinformation, incomplete analysis, and fabricated results that are presented convincingly. Mitigating these risks requires ensuring that official, trustworthy data is AI-ready, not only readable by machines, but also understandable by them, and interpreted by them in ways that complement the way humans think. Now, making data AI-ready would allow these current systems to directly interact with official datasets, hopefully helping offset their overreliance on secondary sources and their propensity to misstate authoritative statistics or provide false, misleading, or outdated information. Not preparing official data for AI integration could undermine trust, But embracing AI readiness also offers other major opportunities. When AI connects to reliable data sources, it can broaden public access to official state statistics, support multilingual interaction, guide users through complex datasets, and strengthen evidence-based action. And using AI in the production of official statistics itself can streamline and speed up processing, quality control, and harmonization. Recognizing these opportunities and also the challenges, the Statistical Commission emphasized the importance of making official statistics more interoperable, open where appropriate, AI-ready, and future-ready. And in March 2026, just a few short months ago, at its 57th session, the Commission established a Citi Group on AI Readiness of Official Data and Statistics, hosted by the National Institute of Statistics of Rwanda. The Citi Group will work to establish good practices and guidance for national statistical offices as they navigate this new terrain. It recognizes explicitly that needs are particularly acute in least developed countries, small island developing states, which face heightened vulnerability and also operate often with limited human, financial, and institutional capacity. Overall, this group— the establishment of this group underscores the urgency of strengthening data ecosystems and leveraging technology responsibly. No single institution can achieve this alone, and it remains very important for diverse communities to work together including statistical offices, scientific communities, the private sector, civil society, and users, so that there's meaningful and inclusive participation and also continuing trust in the data that official systems provide. As we begin our session, I am keen to hear our speakers' perspectives on how countries can build stronger and more integrated data ecosystems, How multi-stakeholder and inclusive partnerships can help address national development priorities. What lessons are emerging from efforts to digitize and modernize statistical systems and responsibly leverage AI? And what investments are the most critical to ensure that our data infrastructure, institutions, and capacities remain fit for purpose not just now but also in the future. I would now give the floor to His Excellency Mr. Burhan Ghaffour, who is Permanent Representative of Singapore to the United Nations, for his welcome and opening remarks.

Thank you very much, Shantanu, for the opening remarks. Dear friends, thank you all for accepting our invitation to this side event, and I want to start by thanking the co-sponsors, which are the Permanent Missions of Rwanda and Jamaica, who are represented here very ably by their Deputy Permanent Representatives, Alain, as well as Robert, who's on my left, and Alain on the other side. And also, I want to thank our partners, the UN Statistics Division, which is supporting this event through the Data for Now initiative, as well as our very distinguished panelists who will be introduced shortly. Now, we have a very short amount of time, and I want to start by making a very broad point that digitization and statistics are powerful accelerators for the implementation of SDGs. And here I want to make 3 brief points. Firstly, data is the invisible infrastructure for the 2030 Agenda. And when we diagnose the implementation, we know that the SDGs are off track, and we often speak of financing gaps, capacity gaps, and a gap of political will, but what we do not often talk about is the data gap. When cities, for example, are not able to plan and implement their policies, it is often also because census data is decades old, and newer sources of data such as geospatial data and real-time sensors are not integrated into city infrastructures. Climate action depends on robust scientific statistics, which is then accelerated and amplified by the use of AIs to discern underlying trends and patterns. Ocean governance depends on sustained ocean observation using very high technology, but also using AI to decipher the data that is extracted. And in each case of the examples I provided, the implementation gap is fundamentally an information and data gap, and that is why it is very important that each one of us, as we drive our own national plans and SDG goals, it's important that we have very strong national statistical offices and data systems that are interoperable between our countries. The second point that I want to make is that, in Singapore's experience, data is approached as a system, or as an ecosystem, as you put it, Shantanu, not as a simple collection of different databases. Our national statistical system treats data as a shared national asset, and it is stewarded by our Department of Statistics, and our Chief Statistician is here. He flew in not just for this side event, but also for a range of engagements here with his counterparts around the world. So I will let him elaborate on some of the things that we are doing, but thank you very much for joining us, Dr. Koh. The last point I want to make is that the data and statistical question is going to be at the heart of the post-2030 agenda. Because if we are to decide what we are going to do post-2030, we must have a clear sense of where we are, what the gaps are, and we need to be evidence-driven, we need to be data-driven, and we need to be honest about the indicator framework. I mean, we have more than 230 indicators, many of which obviously are lagging, but also countries are struggling to report against this indicators framework, which means that we should be ready to take another look, revise the indicator framework if it's necessary, so that it allows each one of our countries to do the implementation of the SDG goals in a more evidence-driven and data-driven way. And the final point I would make is that in a very political and polarized context at the UN, I think evidence-based discussion, data-driven dialogue can be foundational to build trust, to build convergence, and to find solutions. So once again, my thanks and gratitude to so many of you for joining us for this event, and I'll now give the floor back to the moderator. Thank you, Shantanu.

Thank you so much, Ambassador, for setting the context for this event, also in the larger framework of discussions at the UN and what we must look towards. Without further ado then, I would invite Dr. Eng Chuan Koh, who is the Chief Statistician of Singapore Department of Statistics, to deliver his keynote address, and this is about strengthening data ecosystems for effective policymaking and SDG delivery. Thank you.

Thank you. Excellencies, distinguished colleagues, ladies and gentlemen, it's my privilege to be able to speak on the topic on strengthening data ecosystems to support effective policymaking and sustainable development. The world today faces a combination of challenges that are complex, interconnected, and rapidly evolving. At the same time, expectations of government have also increased. Citizens expect public services to be responsive, targeted, and effective. Policymakers need better information to understand what is happening and evaluate whether policies are working, both at the micro level of service delivery and at the macro level of national planning and governance. The 2030 Agenda for Sustainable Development, Pact for the Future, Global Digital Compact, all recognize the importance of data. Digital technologies, international partnerships in addressing global challenges. Today, I want to share my thoughts on 3 issues. Firstly, data challenges facing SDG monitoring and delivery. Secondly, opportunities created by digitalization, administrative data, big data, and AI, which we have been talking about so much. Thirdly, how diverse data sources can be transformed into official stats and policy insights for SDG monitoring and decision-making. Next slide, please. This is a quick one. Just take a look at the slide. In 2019, when we first started monitoring SDGs, we were not doing that well. There were a lot of data gaps. Gradually, over time, we have filled it up. It's still not perfect, but we are making good progress. It's just a quick visualization of where we are moving towards. With that in mind, we know that SDG goals are ambitious and comprehensive. While countries have made a lot of progress, there still remains significant data gaps. Several challenges are common across countries. First, timeliness. Traditional surveys and censuses are essential, but they can be costly and time-consuming. So policymakers nowadays need more information faster, more regularly than traditional data cycles. Secondly, granularity. National averages can hide important differences between regions, socioeconomic groups, and population groups. To achieve the SDG commitment of leaving no one behind, finely disaggregated data are needed to help identify vulnerable groups. Thirdly, data fragmentation. Useful data already exists across government agencies and the private sector, but these datasets are often undiscovered, disconnected, or underutilised. Fourth, resource constraints. National Statistical Offices, or NSOs for short from now on, face increasing demand and expectations, yet expertise and budgets are not keeping pace. Access to technology, AI, infrastructure, they all require investments that many countries find difficult to support. So as we approach 2030, there's not much time left. Fortunately, with the rise of big data, digital technologies, AI, this presented an unprecedented opportunity. NSOs today have access to more data than ever before, In addition to surveys and censuses, there are growing volumes of admin data, geospatial satellite accounts, mobile and internet data, digital or online transaction data. Now, the challenge lies in understanding, acquiring, and integrating the— and developing the capability to govern and use the data effectively. So admin data is one of the most valuable but underutilized resources available. And these examples include education records, health records, business registers, tax custom data, utility databases, social security systems, and the list goes on. We need to be able to integrate and make sure that it's interoperable so that we can develop much more detailed and timely indicators. And reduce respondent burden and costs. In Singapore, we introduced the Public Sector Governance Act in 2018 to harness the wealth of admin data in government so that citizens can benefit from the tell us once principle. Under the PSGA, the Singapore Department of Statistics took on the role of a Trusted Data Intermediary, or the Trusted Centre, we call it for short. To integrate and use and share admin data. And we are well-positioned with our wealth of experience and knowledge in data management built up over the decades. And this has extended the NSO's role and relevance, positioning us as the trusted first stop for users to obtain data and data services securely. So AI presents unprecedented opportunities. With the right approach and resourcing, AI can help NSOs throughout the data production chain. For example, web scraping applications have evolved to be able to capture both structured and unstructured data. In Singapore, we established a Digital Transformation Unit in 2022, recognizing the potential of AI and machine learning in our stats processes. And with the work becoming more evident, We saw that the capabilities could not remain siloed, and we need to embed across our entire data value chain. This led to the formation of the Data Products and Innovation Division in 2026, with a mandate to develop and scale digital and AI capabilities. Now, AI can extract information from images, text, process large volumes of structured and unstructured data, help in coding, improve data quality checks, outlier detection, and accelerates statistical data production and reduce actually the cost of production. They are actually our junior statisticians waiting— AI waiting to rise the ranks to become senior statisticians. So there lies a risk Governments must approach this technology responsibly. For example, data privacy and confidentiality concerns, misinformation, hallucination, cybersecurity risks, loss of professional knowledge when we over-rely on AI. So that is a real danger as well. As producers of official stats, we need to set the standards and methods and understand the processes, concepts, ensure the quality of our outputs. So trust still remains the foundation of any data ecosystem, and without it, data cannot effectively support policymaking. I just wanted to share a use case. Take the SDG 3.2.1. Under-5 mortality rate is computed using life tables, which in Singapore's context, used official death registrations. They provide accurate and complete measures of mortality, much more so than just survey reporting. Like Singapore, NSOs and governments can combine vital registration data that universally captures key life events such as births, deaths, marriages, and divorces, with hospital records, census data, and meet data on population and dwelling. And this will create much richer datasets. And we know this has served us well during COVID Where there are data gaps, AI and machine learning models are increasingly capable of integrating and creating network data. And they use diverse data sources, enable a more holistic picture of the healthcare landscape, merged with socioeconomic and geographical location characteristics of the population. Many more possibilities, including mobile data to capture travel patterns, satellite imaging to identify land use patterns, utilities data to determine population dwelling coverage and whether there are people who are staying in the addresses. So all these, the future of SDG monitoring, will not depend on a single dataset, but rather a full data ecosystem, as mentioned before. But there are 4 key elements that are important as we move to using AI, large datasets, private sector data, and mean data. First, governance. Clear legal frameworks, data stewardship arrangements, and responsible data sharing mechanisms are essential. And that's where the NSO is well-positioned to take on the larger data stewardship role, given their experience as primary users of data for stats production and large database management. Second, interoperability. Data must be able to move securely across systems through unique identifiers, common standards, consistent metadata, and streamlined IT architectures. NSOs need to partner, partner, right, their tech partners, counterparts in government to modernize both statistical agency and administrative data agencies to harness its full potential. Third, the investment in capabilities. Governments need to invest in stats skills, data science skills, digital literacy, AI capabilities. So NSOs need a new breed of data stats engineers, professionals who can bridge the world of stats with rigour and modern data engineering. Fourth, trust. Trust can be viewed in 2 ways. Firstly, the citizens must have confidence that their data is protected and used responsibly. NSOs need to be transparent in their methodology as well as the usage of data. Secondly, NSOs must provide data that people need. In a timely manner and in good quality to remain the trusted go-to centre for data. In conclusion, Excellencies and distinguished colleagues, potential benefits of a strong data ecosystem are within reach. By combining the traditional data sources with admin data, big data, geospatial data, digital technologies, and responsible AI, NSOs can significantly improve data available for policymaking and SDG monitoring. To realize these potentials, governments must invest in the right capabilities and ensure that innovation is accompanied by strong governance, robust safeguards, and public trust. At the same time, NSOs need to be innovative in their use of AI and digital tools, and be transparent in their methods and concepts. Harnessing the wealth of data that already exists, transforming them to fit-for-purpose data and indicators, ensuring their reusability— this requires a whole-of-government approach, and it will definitely result in better outcomes and better SDG monitoring. Thank you so much.

Thank you, Anshun, for sharing Singapore's practical experience. but also laying out these broad parameters which have to define the way we move forward. Let me also say that it's nice to have Anshu, the human being, here and not the AI agent.

Not yet.

Not yet, right? Okay. I will now invite Robert Kainamura, who is the DPR of the Permanent Mission to Rwanda to the United Nations. As we mentioned, the Citigroup that the Commission has set up is based in Kigali. We are looking forward to hearing from you, sir, about how countries in the Global South, in particular, can strengthen their capacities to harness data and technology, not just in AI, but also more broadly, to better inform their own policymaking and work. Over to you, sir.

Thank you very much, Mr. Moderator, for giving me the floor. We're happy to have hosted this side event with Singapore and Jamaica and other partners.

Thank you.

Thank you again, Mr. Moderator, and also Dr. Koh for this insightful presentation in Singapore's view, which Singapore will work very closely on these matters with the Statistical Commission as well. One takeaway is that stronger data ecosystems are no longer simply a technical aspiration for countries. It's a key tool for effective governance, and I think this has been a conversation relayed by the speakers who spoke before me. For Rwanda, for this case, digital transformation is not just about technology. It's a national development strategy aimed at strengthening institutions, improving public services in real time, and enabling evidence-based policymaking and decision-making for the data alignment as well as to be informed. But as most speakers mentioned, whilst surveys and censuses remain indispensable, as has been mentioned, they are no longer sufficient on their own. They need to be complemented by administrative records, geospatial information, and responsibly governed real-time data to make decisions. This is very critical from local government bottom-up. For Rwanda's case, in our national strategy for transformation, we have invested in trusted digital identity, digital payment systems, security data exchange, and citizen-centered public records and services supported by a strong broadband connectivity. This was just a baseline to start with, to be able to look far in terms of where we need to be. A practical example is the Irembo platform, through which citizens access hundreds of government services, Online services that took days before and today takes only 24 hours. Improving efficiency while generating variable administrative data that complements official statistics and strengthen policymaking is becoming more easy for our system. Now, coming to the question you mentioned, based on our experience and many other countries, it demonstrates that digital transformation is not simply about digitizing services. It is about creating trusted systems where data flows securely across institutions and informs decisions in real time. This requires strong institutions, common standards, and a whole-of-government approach. The whole entire system has to be aligned. Therefore, artificial intelligence now offers an opportunity to build on this foundation. When deployed responsibly, artificial intelligence can help our governments, and analyze complex data in real time, strengthen disaster preparedness in real time, improve agricultural productivity in real time, optimize health services in real time, and enhance social protection for our citizens. But AI, as mentioned, is only as reliable as the quality of the data behind it, and this is what I think Dr. Koh was emphasizing. For this reason, Rwanda is advancing AI alongside investments in data governance, cybersecurity, privacy protection, and digital skills. Through our national AI policy, the Center for the Fourth Industrial Revolution, and the partnership including with Singapore, we are promoting a trustworthy AI that strengthens public institutions while protecting citizen rights. Now, our experience also reminds that no country can build tomorrow's digital ecosystem alone. I think this has been mentioned. The Pact of the Future and the Global Digital Compact, which we co-facilitated at some point, call for a greater international cooperation not only through financing, but also by sharing digital public good, governance frameworks, institutional capacity, and good ecosystems that are going to help us. Now, for Africa, it represents a unique opportunity to leapfrog legacy systems and build a modern AI-ready digital ecosystems. As we move ahead, we want to offer a few reflections. The true promise of digitization, official statistics, and artificial intelligence lies not in the technologies themselves alone, but in what they enable our government, our systems to achieve. better decision-making, better services, stronger institutions, and a faster progress towards the sustainable development, as mentioned. Now, as we look beyond 2030, we believe our shared responsibility is to ensure that innovations remain people-centered, trusted, and inclusive, so that the technology we're looking at expands opportunities for every citizen and makes the work we look ahead, as the Ambassador mentioned, in the years to come, to help us to achieve the goals we aspire. I thank you very much, Mr. Moderator.

Thank you, Ambassador, for sharing these insights and for emphasizing that it's not technology for technology's sake that we are looking at, but also, it is a question of shared responsibility and mutual accountability. I would now like to invite LL Reid, who's the DPR of Jamaica to the United Nations, and ask about, from Jamaica's perspective, how can these robust national data ecosystems we've been talking about strengthen policymaking, build institutional trust, and accelerate integrated implementation of the SDGs? And of course, as a Small Island Developing State, are there special considerations you would like us to keep in mind? Over to you.

Thank you so much, Mr. Moderator. Let me begin by saying that Jamaica is very pleased to co-sponsor this side event and to help create a space for this important and very timely dialogue. I am especially pleased to join this panel alongside my colleague from Rwanda, which, like Jamaica, is presenting a Voluntary National Review this year. So, colleagues, I'm very happy to also hear what Dr. Koh mentioned in his main wrapping-up points on governance, interoperability, and trust, because this resonates, actually, with a lot of our experience, and you'll hear me return this as I continue my presentation. So let me share briefly from a policy perspective, certainly not from an expert perspective, on how Jamaica views the link between strong data ecosystems, governance, and ultimately the implementation of the SDGs. For Jamaica, the relationship between data, digital transformation, and sustainable development is actually quite clear. Effective policymaking depends on reliable, timely, and accessible data in an increasingly complex global environment. Strengthened data ecosystems are essential not only for measuring progress, but also for anticipating risks and building resilience. This imperative is particularly relevant for a small island developing state like ours. Our countries face unique vulnerabilities arising from climate change, natural disasters, external economic shocks, and capacity constraints. These challenges underscore the importance of modern, resilient, and inclusive national statistical systems that can support evidence-based decision-making and accelerate our implementation of the SDGs. For SIDS, robust national data ecosystems are not simply a technical aspiration; they are a strategic necessity. In a context where resources are limited and vulnerabilities are increasing, governments must be equipped to make timely, informed, and evidence-based decisions. This is only possible when policymakers have access to data that are detailed, timely, relevant, and reliable. At its core, a national data ecosystem is about more than the data collection itself. It is a coordinated system of institutions, governance arrangements, technologies, and partnerships that transform data into actionable evidence. By integrating information from censuses, surveys, administrative records, civil registration systems, geospatial information, and other emerging data sources, governments gain a more complete understanding of their development challenges and also opportunities. Such ecosystems strengthen policymaking by providing the evidence that's needed to design, implement, monitor, and evaluate public policy. Official statistics help governments identify emerging demographic and socioeconomic trends, to assess the effectiveness of programs, to allocate very scarce resources more efficiently, and to anticipate future demands for areas like education, for healthcare, housing, employment, social protection, and even infrastructure. For countries like Jamaica, robust data systems are equally critical for disaster risk management, for climate adaptation, and also resilience building. Strong data ecosystems also help accelerate implementation of the Sustainable Development Goals. The SDGs are inherently interconnected, and so too are the challenges that they're seeking to address. Effectively responding to these challenges, therefore, requires a whole-of-government and whole-of-society approach, supported by data that can be shared, integrated, and analyzed across sectors. This underscores the importance of partnerships among a range of entities, including, at the national level, our national statistical offices, line ministries, academia, civil society, the private sector, but also including regional organizations and our international development partners. Collaboration between data producers and data users also helps to ensure that statistics to respond to policy needs, and that vulnerable and underserved populations are visible in national planning. In this way, robust data ecosystems very much support the central commitment of the 2030 Agenda to leave no one behind. I also wish to highlight the role that data ecosystems play in building institutional trust, mentioned already, of course, by Dr. Koh. Trust is strengthened when public institutions consistently produce high-quality, objective, and transparent statistics, and when citizens also have the confidence that their information is collected, managed, and protected responsibly. When institutions are trusted, people are more willing to participate in censuses, surveys, and organizations are more willing to share data responsibly. Responsibly. This means policymakers can make decisions with greater confidence. Jamaica, like many other countries, is making efforts to modernize its statistical systems, and digitization is a central pillar of that effort. Traditional methods of data collection are increasingly insufficient, and this creates pressure on national statistical systems. To find more efficient and sustainable approaches to data production and data use. From our experience, digitization is not just an end in itself, but rather it's a means of improving decision-making and strengthening development outcomes. Jamaica's 2022 Population and Housing Census, for example, marked a major milestone. It was our first fully digital census, with thousands of enumerators using tablets to collect data electronically. This transition improved data quality, enhanced operational efficiency, and strengthened our ability to generate timely information for national planning. We have also recently launched— and this is earlier this year— the Jamaica Data Exchange Platform. This is an important step towards enabling secure and efficient data sharing across government institutions. This improves interoperability and reduces information silos. The platform supports more integrated policymaking and more responsive public service delivery. This is being referred to as a foundational pillar of our digital transformation agenda. Complementing these efforts is our Vision 2030 Jamaica electronic monitoring platform platform and dashboard, which supports the tracking of national development indicators and progress towards both our national development objectives and the SDGs, which are actually quite closely aligned. This enables policymakers to monitor outcomes more effectively and to make evidence-based adjustments where they become necessary. Colleagues, as discussions increasingly turn to the transformative potential of artificial intelligence, It's important to recognize that AI is only as effective, as my colleague said, as the data that underpins it. Strong national data ecosystems, therefore, provide the foundation for the responsible and effective use of AI in public policymaking. When supported by quality, timely, and representative data, AI can help governments identify trends, improve forecasting, optimize service delivery, and strengthen policy analysis. At the same time, it's essential that the deployment of AI is guided by appropriate governance frameworks that promote transparency, accountability, privacy, and inclusivity, ensuring that technological innovation advances development objectives while maintaining public trust. Jamaica has an established the National Artificial Intelligence Task Force, which is now working towards developing a national AI draft policy by November of this year. This ties in quite neatly to our discussion on enhanced data collection and our objective to prioritize digitization at the national level. Together, colleagues, these initiatives are helping Jamaica to build a stronger, more integrated, and more forward-looking data ecosystem, one that enhances governance, improves public service delivery, and supports sustainable development. Looking ahead, countries must continue to make strategic investments in digital infrastructure, interoperable systems, secure data management, and statistical capacity. These investments are essential if we are to harness the full potential of data, digital technology, and emerging tools such as AI in support of sustainable development. For us, investing in robust national data ecosystems is ultimately an investment in better governance. Better data enables better decisions. Better decisions strengthens public trust, improves policy outcomes, and accelerates implementation of the Sustainable Development goals. Thank you.

Thank you, Ambassador, for sharing Jamaica's forward-looking perspective here and really for emphasizing also this virtuous cycle between trust and data quality. I would now like to invite Maria Dimitridou, who is the World Bank Special Representative to the UN, Head of Multilateral Affairs, And a close— the World Bank is a close partner of the Statistics Division here, so it's a special pleasure to invite her. And it has been advancing a systems-based approach to data and statistics through initiatives such as Data360 and in promoting AI readiness in official statistics. So we'd love to hear from her about the Bank's experience in these areas and how this is contributing to further strengthening this work.

Thank you very much, Santanu. Excellencies, dear colleagues, let me begin by expressing my appreciation to the co-organizers first. I'll start with Singapore, Rwanda, and Jamaica, that are 3 countries from 3 very different places in the world, each one navigating its own development journey, but each one demonstrating that investing in data is not a technical exercise. It's an act of governance, it's an act of ambition, and it's an act of building trust. But also, Santanu, please allow me to express our deep appreciation for the valued, longstanding partnership that we have with the UN Statistical Division for many years now to advance the data agenda globally, but also in support of countries. Now, as you all alluded to, and Talela, you made the point as well, when governments need to make a critical decision, about where to build a school, how to respond to a shock, whether a social protection program is reaching those it has been designed for, they need the data to be there, to be there in time, and to be good enough so that they can act with confidence. But often this is not the case, and this is a gap that is not acceptable, and it's not— but it's not inevitable. So the World Bank Group is in the business of closing it. working with partners. For the World Bank Group's job strategy, investment in data infrastructure sits at the very core of our approach to productive, inclusive employment. Without quality data, that investment is flying practically blind. Data is not a supporting actor. It is the foundation to invest in physical and human infrastructure, promote the right policies, and mobilize private capital at scale. which are the 3 pillars of promoting the creation of more and better jobs. And this is really important. We have also young people in the room. We anticipate 1.2 billion young people coming into the work— to employment in the next 10 to 15 years, but with the current economic prospects, there are jobs for about 400 million. So data is very important in navigating this path. Let me use these 2 examples of why data is also very important with the scale and urgency and speed of World Bank Group initiatives along with our partners. 2 examples that have been discussed this week at the High-Level Political Forum. One is Water Forward, which is an initiative that was launched at the World Bank Group-IMF Spring Meetings, where along with other MDBs and partners, we aim to deliver water security to over 1 billion people by 2030. Or Mission 300, where, again, working with partners, we aim to provide access to electricity to 300 million people in sub-Saharan Africa. And all of that is by 2030. Different challenges, very interconnected, same conclusion: sustainable outcomes at this scale and urgency that we need to move are only possible with data-powered solutions. What we have built— as, Sant'Anno, you alluded to our system— is we call it Data 360. Only 4 years ago, the World Bank Group had open data spread across more than 80 separate portals, which came, of course, with its challenges in terms of interoperability, but also in terms of consistency of documentation sometimes. You heard me oftentimes in New York talking about how the World Bank Group is a knowledge institution as much as we are a financing institution, but to further strengthen the knowledge part, we needed to strengthen further the data bank part of the institution. So what we have decided and we have done is to consolidate, harmonize, and build for the age of artificial intelligence our data work. And the result was Data360. This is our flagship open data platform. It's publicly available. It integrates more than 300 million data points, over 10,000 indicators, and more than 50 years of time series disaggregated by sex, age, employment, income, location, and education, all quality assured and tied to source documentation, which is publicly available. That is 40 times more development data than we had previously published in one coherent AI-ready ecosystem, which is drawing over 100 million page views and roughly 1 billion application programming interface calls per year. But Data360 is not just a portal, just a website. It is built to make data legible to artificial intelligence. We have launched the Data360 Model Context Protocol, a mechanism that steers leading AI tools like ChatGPT, Gemini, Copilot, Perplexity, and others to our data, helps them read our metadata correctly, and ensures that when AI generates a development answer, it cites a trustworthy source. This is how we move from AI that sounds authoritative to AI that actually is. I share, Ambassador Kafour, your concern. As soon as I came to New York last September, I felt that how important it is to support your efforts with the data work of the Bank in multilateral collaboration. So Data360 was actually the first series that we launched in New York where we, let's say, trained colleagues from across our partners and missions on how to use Data360 2060 and accompany your great efforts here. Now, I would like to focus on governance. That was a point that was raised across the board. None of these efforts work without governance. Inside the World Bank Group, we have restructured the World Bank Group data governance architecture around 3 dedicated councils covering development data, corporate and operations data, and enabling technology so that quality standards and accountability move together. And this model has become a reference point for peer MDBs. We have also built a capacity-building platform offering learning tracks in data collection, quality management, analytics and AI for data, and data for AI, taught to the same standards that the Data360 is built on, so that these capacity platforms and policy move as one. Now, partnerships Partnerships are core to this effort as well. It's no exception. Partnerships are core everywhere, here as well. The World Bank Group is the largest funder of data and statistics for low- and middle-income countries in the world, with an active portfolio, as we speak, of roughly $2.3 billion across 47 countries. But a lot more has to be done, so that's why partnerships are so important. Through the Global Data Facility, every dollar invested has mobilized roughly $950 in new World Bank lending for country data systems. So the leverage there— we're talking about financial leverage, but there is also this leverage. Through our Development Data Partnership, we have brought 39 incredibly powerful partners, 11 international organizations. Of course, our UN colleagues are always our partners in that. 28 private companies, including Google. Google, our colleague, is there. Microsoft and LinkedIn to share high-value proprietary data that are freely and securely available for the public good, with real use cases already— tracking development after disasters, measuring pandemic impacts, improving road safety in data-scarce environments. Now, to Singapore, Rwanda, and Jamaica, your presence here and your leadership matters. You are not just waiting for the data revolution, you are actually leading one, demonstrating that statistical capacity is buildable, that AI can serve official statistics. And all of you in the room are participating in this journey. So the World Bank Group and my presence here is not just to provide an update on information, but also to learn from all of you, to continuously learn the challenges— from the challenges you face, from the good practices you're pioneering, the systems you're building, the choices you're making, and we carry this knowledge together with our expertise and our own knowledge across client countries facing similar policy choices to support. So we will continue to support the data agenda with financing, with knowledge, with platforms like Data360, and with the conviction that well-informed decision-making is not a luxury, it is core in delivering the development outcomes. So let me leave you with one final thought: data infrastructure, data governance, data skills, and data partnerships are not 4 separate initiatives. They are not 4 separate discussions. They are one approach designed to work together. And the only way to achieve the data revolution is to do it together. So we would welcome you to join, and we will definitely join you, so that we can together build them at the time and with the urgency and at the scale that is required. Thank you.

Thank you, Maria, and especially also, I think, for emphasizing that just like with humans, with institutions too, we are all on a continuous learning curve, and it's never too late to start. So, I would now like to invite Dave Hotchkiss, who's engineering manager at Google, to offer his perspectives, particularly on how we can harness technology and build collaborative, scalable AI that can analyze massive datasets, maybe Data 360++, to drive more impact.

Thank you to our moderator, to the Excellencies, our esteemed hosts. Thank you for being here today. I'm delighted to be included And as we've heard, a consistent theme on the trustworthiness, the reliability, and the integration of data. With 30 years of experience in the industry, I couldn't agree more that this is a common thread. This is something that we see throughout and is endlessly becoming more important as we think about technology moving forward. As described in the brief, the goal of timely decision-making while maintaining credibility, accountability, and public trust. This remains the same throughout the integration of new technologies. As we think about the challenge today, when we're trying to bring together these different datasets that are coming online, one of the biggest challenges is harmonizing and making it interoperable. And so, when we take a complex topic like climate change, changing from a perspective of looking at small variables like the water temperature or air temperature To thinking about the full story and telling a narrative requires that we have interoperable data that also takes into account population density, population demographics, what are health outcomes, and then building an understanding of how they're interrelated. And so, to accomplish this today, we tend to work with a lot of data analysts, individuals who have been incredibly well-educated. It's a lot of human capital to bring these datasets together from various portals that are spread across the internet, information that's stored in different databases, attempt to map all of these different schemas against one another, rename column headers, establish where there is or is not harmonization across these datasets in order to build these comparisons. And this is incredibly expensive. This is cost-prohibitive for many scenarios. It's an ineffective use of human capital and ultimately results in what we've heard before, which is a lot of us end up flying blind when it comes to making well-informed decisions. And this is where the opportunity is with platforms like Data360, with work that Google is doing through Data Commons, to try to make it possible that individuals who bring a small dataset are able to normalize or harmonize that very quickly using tools that are open source and available freely in order to then be able to make more data-driven decisions. 6 or 7 years ago, Google approached some U.S. agencies to say, there's this opportunity called schema.org. There's a way to mark up information that's on the internet. And the— at the time, there were around 40 million domains already using this. It seemed to have pretty broad adoption. And the government's response that we heard was, here's a 400-page PDF, it's available online, it's already there, and so it's open. Not quite what we were hoping for. And so we set out to try to prove the value of having that structure and harmonize data. And so Data Commons is related very closely to Google's mission, where Google looks to organize the world's information and make it universally accessible and useful. We focus on the statistical component of that. And so we started, as we've just described, with a team of specialists to map schemas, to harmonize data, and to try to make some of that public data available where the licensing is appropriate for others to grow and interconnect their own data. And so as we've built out these thousands of tables, these hundreds of billions of rows, this is a graph that we want to encourage others to use or to take the concepts and figure out how they can extend that for themselves. And so for a large organization that's looking to host, analyze, or publish their own data, quite similar to Data360, we want to make a platform that's available so others can use and interoperate. The lessons that we've learned from a mindset internally, because we are also going through this AI revolution. We did not start this project during a world where AI was everywhere, and therefore we knew that we could rely on agents to think about harmonizing and adjusting schemas. We've been learning as well that it needs to be a conscious effort. Everyone is seeing that timelines are condensed. Development assistance has dramatically reduced in the last couple of years in particular. It's harder than ever to try to allocate people to focus on a task force or something new. And yet we're saying that everyone needs to dedicate a mindset or a portion of their mindset to thinking about AI readiness and AI-appropriate data. Within our own team, we've found that it's helpful to dedicate a window of time to say whether it's a hackathon, a competition to generate insights, to include everyone, everyone, not just engineers, but most importantly, to bring in people from other functions Take advantage of the fact that with AI Studio, with Codex, with Claude, anybody can write code now. That's no longer a walled garden. Anybody can use tools that are available and start to interoperate with whether it's portal data, whether it's data that's available through an MCP service. Start to use those types of coding tools and you're going to feel the friction points yourself. And then you're going to think about how can we resolve those friction points and how can we move forward? And you might think that, as I said, this would be natural at a technology company like Google, but no, it was very much a forced exercise for us to say we're going to take a week, we're going to stop our regular work streams, and we're going to focus on this. And now if you go back and look at my teams as a proxy, look at their token usage, you can see that clear inflection point last year where every product manager, every program manager, everybody that worked in marketing took a week to then come back and do a demonstration to the rest team of, here's how I've been able to leverage this type of technology. And so, we can see that inflection point clearly, and we see ourselves making better products available. And the lesson that we learned besides just, here's an opportunity to identify our own friction points, is that the technology, frustratingly or not, is constantly changing. Last year, the big thing was Model Context Protocol, and that drove this real interoperability for data and how can you start to bring more systems together. And now we see that adopting to agent skills, toolkits, and in particular, a lot more of an interoperability in a swarm perspective. So instead of one-on-one relationships, you're seeing these higher complexity systems. That has allowed us to start to think about how does AI apply not just to taking insights from data, but throughout the system. And so, as we've already heard here today, the data is only useful as part of an ecosystem, something that's living. And it's not just to be considered a kitchen sink that you throw more information into and then something will magically happen. The information is most valuable when it's consistently growing, when it's consistently being increased in harmony with the other data that's around it. And this leads to an environment like a true ecosystem where all of the living elements improve or adjust and accommodate to the environment that it lives within. And so, AI as an opportunity, it's available for thinking about the data generation and whether it's durable analysis and automation of how do you extract data from satellite imagery, geospatial information, whether it's weather stations. But we've also heard examples from the interoperability of payment and transactional data and being able to aggregate health outcomes. And so, governments are uniquely positioned to think about how can we use AI to protect privacy and to build well and robust-founded datasets to be used as input. When we think about data cleansing and normalization, there's a natural place for AI to maintain a much broader context. That 400-page PDF is impossible to keep in one person's mind. And yet, with the expanding context that's available to the models that we use today in the frontier models, it's possible to maintain far more of that context than any individual could and to understand when there's appropriate footnotes, when we should have attributes that are noted on an observation. So, not just does it appear on page 300 of this document, but it can also appear real-time when we're reviewing that statistic and know that this is the footnote that's relevant and important. It also can then help with proposing what are the code maps and how might you harmonize it for human use, for human review. And as we think about data sharing, you've already heard today about a number of open-source tools, open-source standards, how to enable model context protocol, and how to enable that type of data reuse that can lead to additional insights. You can imagine that with the UN, we have right now a document around 193 193 nations talking about the Sustainable Development Goals. Over time, imagine the power that it could be to have 193 books, each one representing the power of an individual country and the way that it can glean off of its own data where it exists within the ecosystem of the UN and how to work with other nations for development. And so with that, thank you for being here. I look forward to answering some questions.

Thank you so much, Dave, and sort of full disclosure, we are great users of the Data Commons. So— but thank you also for talking about the dynamic nature of the data ecosystem and its extensive growth complexity and interconnections within it, and so the need for harmonization, but also for pointing out that the human in the loop is not just to make AI better, but to make humans much more productive and more effective. So, dear audience, I— at this point, we've heard several very illuminating presentations, and there is much room for discussion. Unfortunately, there is limited time, but I would like to open the floor to questions, thoughts, reflections. If you would like to direct them to anyone in particular, please do so, or you can just ask the panel in general. My request is that please keep this really focused and short so the panel has a chance to respond. So, I see our first volunteer here, and then we'll go here. So, please introduce yourself. Over to you.

Thank you. Thank you so much, Mr. Moderator. My name is Hugo Garcia from Guatemala. In Guatemala, we use different data approaches for development, from building our social household registry a costly effort with uncertain long-term funding, to using satellite images and artificial intelligence to map informal settlements in Guatemala City, working together with the United Nations system. But these tools rely on data sources that underrepresent indigenous and rural populations, the very places with the largest development gaps. So, for the panel, before we scale up these tools and add more artificial intelligence to develop monitoring, what financing and governance models can make sure the data system behind them are lasting, not just a costly one-time effort, so that AI closes these inclusion gaps instead of making them worse?

Thank you.

Thank you for that question about how to get AI to help close inclusion gaps. And I will let the panel reflect on that, but let me take another question from here.

All right, great.

Thank you.

So, thank you to the panel for your insights. My name is Kevin Yuan, and I am a youth advocate from Hong Kong. So I think, as mentioned in the discussion, AI really only works when citizens trust that governments provide high-quality data. However, there is some research indicating that when AI runs into challenging or impossible prompts, it sometimes cheats and deceives users instead of honestly indicating a lack of knowledge or ability. So this would obviously act as a significant undermining of the trust that AI systems truly need, especially when created by governments.

So my question here is that how can governments and organizations mitigate this risk of deception to ensure that the information provided to users is accurate?

Thank you. And in the interest of fairness, let me go to the gentleman at the back, and then maybe we'll let the panel respond to these, and we have a second round. Please, sir.

Let me shout. Is it possible? This is my specific question goes to Dave, but let me introduce myself.

Attila Karaman, the Director of Training and Cooperation at CESPIC, subsidiary organization of Islamic Cooperation mandated with.

Statistics as well, and catering the needs of 57 developing member states. Um, Dave, as your current role at Google as the engineering manager, you are close to the technical realities of the shift toward agentic AI, highlighted also at the I/O 2026 event in May 2026. As Google leverages its data comes technology to build AI-ready public data infrastructure with NSOs. How can this ecosystem be structured into a truly sustainable support model? Uh, my specific question in that regard: how will Google actually design these agentic frameworks so that local staff in developing regions can independently maintain and adapt these workflows locally rather than being permanent dependent on the proprietary cloud infrastructure?

Thank you.

Sure.

If you don't mind, I can work to address these in order.

Yeah. So, why don't we have you take that, and then we just go down this way and carry on.

That works as well. We are very much committed to ensuring that the platforms that we build and that the interactions that we have with agents are open to whomever is interested in leveraging that and making that a non-proprietary availability. And so, in the case of Data Commons as a platform, it is developed on GitHub. It is open source today. It is possible for anybody to come and to use that. We are developing in a way that is leveraging, at the moment, some of our own best-in-class storage mechanisms, but we are also working to ensure that those are interoperable standards such that anybody could come along and adapt the interfaces that are there to work with other storage mechanisms, other agents, and to be able to use their own models if they should choose. And so, what we feel is a foundation priority is to make sure that there is transparency in how this is developed, that anybody can use that without cost, and that we continue to allow for interoperability so that if there are components that you would want to swap out and not to leverage our own models or our own infrastructure, that that's an option. And so I would invite you to go look at the source on GitHub, point out wherever we might have some element that defaults to a proprietary element, and let us know how we can best adopt that in order to make that more accessible. I'll also point out that what we've seen over the last couple of years is very much this rapid decline in the cost of the use of the models, especially on the inference side. The affordability of the models continues to improve, and that that's something that we don't expect to slow down. And so, even in a scenario where you are exceeding what's available from any of the frontier models with a free quota client, there is a continued drive towards making those more affordable.

Thank you. Anything you would like to share as a response to these questions? or something more general too.

Yeah, no, thank you. I'm also learning quite a lot from colleagues and my experts here beside me on this panel, and I actually thank our colleague from Guatemala for the question that he raised because I too am very interested in hearing how we can use AI to close the inclusion gap for the marginalized communities. We have a similar problem in Jamaica, and So that would be quite useful for us as well.

Maria.

Thank you. These were excellent comments and questions. Some point directly to the discussion of data, but some go beyond, you know, the application of AI as a technology more broadly. So, on data, I shared previously how we, in our path as World Bank Group— and this is reflected also in the support we provide to countries, it can be the application of the technology in water systems or in energy or tracking or disaster risk management, et cetera, there is an investment that needs to be made on how data becomes legible and operate and discuss, let's say, interact with AI. To the point that I was making on how AI doesn't only sound authoritative, but it really is. And you have the source of data, and for researchers, they can know, cross-check and they can validate and verify. And that's very, very important. More broadly on the use of AI, as part of our digital strategy and AI strategy, what we are trying to do is support countries with the connectivity, of course, which also links with energy access, etc., with building the foundations basically, with skills, with the use of small AI, which is very promising in the developing countries where they may not necessarily have access to expensive infrastructure and they can use everyday devices. But again, even in that, more importantly, it's very important to engage in these discussions as today, as we are speaking, actually, in Geneva, they were having the first UN dialogue on AI governance. Our team has been there. We are launching in September the World Development Report, which is the flagship publication for the World Bank Group, is going to be on AI. We are going to bring it in New York as well to discuss with colleagues, which talks about the issues of the opportunities that AI is creating for emerging developing economies, the safeguards that need to be there, implications on job creation, and how to basically make sure that it's not leaving anyone behind, doesn't create further inequalities, but helps lift everyone to grow within and among countries. So I think it's very important that you're here. I think it's very important that you continue to follow these discussions as they evolve, because my view is— and I think from the New York office you have seen that— this has to be a continuous discussion and not a one-off. We need to continue investing in our common understanding and collaboration in New York, on the ground, at the regional level, so that we can advance this agenda. And your contributions to our youth are of— I think are invaluable.

Thank you.

Thank you, Maria. I will now skip Eng Chuan for the moment, in the spirit of leaving the best for the last. And Vivek has kindly declined to answer any questions, so I guess, you know, if you're sponsoring the event, you do have the privilege. privilege of doing that. So, let me go to Raoul.

Thank you very much. I think, just to add to what they said from the question of Guatemala, I think this is work in progress. Data is a big issue, so especially in developing countries, data is a big issue, and that's why we are putting centers— AI is a tool, is a tool, but it works better when there's data. When there's lack of data, then it's not going to deliver the way it's supposed to be delivered. I think we need to work on that to address some of the questions to what the colleague from Guatemala mentioned. Now, to the young delegate, I think his question is, how do we use AI responsibly? I think this is the question. The responsible use of AI is a combination of a lot of things. Good data, quality data is going to determine on how responsibly you are going to apply the tool, which is AI. But I think more importantly, I believe that AI is helping us to enable work in decision-making but it should not take away human judgment. I think human judgment has to remain important even when you're going to use the AI. So it's data, it makes the work easier, it makes you to analyze a lot of data, but the human judgment has to remain at play going forward. Thank you.

Oh, all right.

Okay.

Go ahead.

Thank you. Thank you very much, Mr. Moderator. And maybe just to build off our representative from Rwanda, to the questions that were asked, really, one on the issue of closing the social inclusion gap, but then also, how do you address issues of what we've heard today on hallucination, you know, deceptive AI, as it is sometimes perceived? Really, the human capacity element is going to be very important, maybe tying that also to the jobs agenda. Part of human capacity development must go hand in hand in tandem with AI use. We've talked a lot about how AI should be developed, collected, disseminated, stored, interoperability, and all of that. I think the use factor also, not in just what platforms are available, what tools are available, but how we, and in particularly the future generations, will be equipped with with the skills to do that. In Singapore, if I may just share 2 very brief policy tools on the issue of hallucination and the issue of ensuring data accuracy and AI accuracy, we've built AI Verify. It's a toolkit. Now, what AI Verify essentially is, is a checklist of any developer, any user of an AI-enabled service to basically check against and corroborate against a set of crowdsourced and solicited input from around the world, from a variety of stakeholders, to basically check against that if their services are providing accurate, verifiable data. The other approach we've really taken to data interoperability and AI development is what we would call a sandbox approach. I mean, this is not new. We've seen many countries now entering the fray into the AI conversation really take a multi- stakeholder collaborative and consultative approach. The idea is that you vet whatever you have developed against the wider crowdsourced intelligence of various stakeholders, including civil society groups, indigenous groups, local populations, international organizations, and a lot of that is just how do we tap on the collective expertise of those who are interested, who are engaged, and who are developing developing and using AI. And today, that seems to be everybody, or almost everybody. So, I think that really has to be— to tap on what Maria was saying about how this has to be a continued conversation, but not just a conversation, but a continued consultation whereby everyone, as we develop more robust datasets and tools, it is also to ensure more robust human capacity and skills to have that absorptive capacity to use AI and data. Thank you.

Any closing thoughts?

Yeah, thanks very much. I must say that I'm not an expert, or I'm not the best. Everybody is an expert in their own areas. So I just want to use some more technical way to respond to the questions from Guatemala. I think in terms of data inclusion or AI inclusion, first of all, from our own experience, we need to have unique identifiers. And with unique identifiers, we can link up everybody in the system in your country to be able to capture— as long as there's a touchpoint with the government, you will be able to know who these individuals are. And that's really for the good of the society. You are able to track down even the more vulnerable groups to be able to reach them, give targeted support. Now, sometimes we do not have data. And that's the case when there's non-response in the census or in the survey. And when you have non-response, usually you do data imputation. And this is really where AI can help. If you have the method to do data imputation, I can assure you that AI learns very fast, and they will be able to replicate what the human has done in terms of data imputation, and they'll do it much better. So this is where I think there's really hope for a much more efficient way of data production. That's one. On the point on hallucination, I hope I got it correct. In the NSO, in our context, when we are using AI, we are also very fearful of letting the AI do whatever it wants in terms of data imputation, in terms of data creation, in data cleaning. Or we do have a bot, for example. This bot sits on our website, and the users, public users, are able to come to the bot to say, plot me a chart on CPI, plot me a chart on GDP. And the bot does all the scraping of the data, put together the data, and create the chart for the user. Now, it's quite easy for some public user to ask for something that we don't have, and the bot could easily hallucinate and give you something that is false. So what do we do? We actually did prompt engineering in a way to sandbox this AI to search for information that's available from our internal databases. That's it. If you cannot find the information, that's what we program into the bot. please tell the user that data is not available. And that's really possible because we have done it. So there is no way this bot could hallucinate for the user, the public, the public user. So that is one way to make sure that it really— the bot doesn't hallucinate. So that's the answer to your question. And I want to say that going forward into the future, AI bots and AI companies or Google, for example, they are very competitive. So if the bot keeps hallucinating, it will soon lose its job. So really, it's serious. I'm serious. So Claude, Gemini, ChatGPT, DeepSeek, play around with these bots. Please play around with them. And you find your answer through this bot. And somehow they will converge onto a correct answer. And you can discount those that are not correct. And increasingly, they give you the links to where they find the information from, and you do your own human check on whether the bot is hallucinating or not. I'm just telling you all these things from my own practical experience. Thanks very much.

Thank you, Eng Suan. And I'd like to thank our great audience for their engagement here. Wonderful questions. Super panel as well and just share 2 things with you. The first is Hugo from Guatemala raised, I think, a fundamental question, which is how do you safeguard and how do you encourage human uniqueness when we are living in a world where AI is generating probabilistic-based answers? This is actually a really deep question. And I think— I don't know the answer to it. And maybe some of it is in the dynamic richness of the ecosystem you're talking about, where unique identifiers will have a place to stand above the mass. And for my friend from Hong Kong, Kevin, is that right? Okay. So what you said reminded me of someone, of a Scottish philosopher, John Smith, from many years ago who had this interesting quote which was used to encourage me to study. And so I thought I'd share it with you. Now, he said, nothing that you will learn in the course of your studies will be of the slightest possible use to you, okay? Save only this, that if you work hard and intelligently, you should be able to detect when a man is talking rot. And that, in my view, is the main, if not sole, purpose of education. So I think with Anshu having already guided you on how you can detect this for AI systems, I think you're well on your way. So thank you all, and a big round of applause for our wonderful panel. Thank you.