Open Science and AI: What MeasureDev 2026 Means for Official Statistics
by Lorenz Noe
20 May 2026
Development measurement has come a long way from simply measuring a country’s GDP. Today we ask data systems to capture sustainability, inclusion, and the breadth of human and planetary wellbeing. This is most visible through the Sustainable Development Goals: 17 goals, 169 targets, and 234 unique indicators and its emphasis on Leaving No One Behind.
But the work is unfinished. Concepts like equity, resilience, and institutional capacity remain hard to operationalize. The challenge is twofold: defining what matters and measuring it well. Do we have the right data? Where are the gaps? How uneven is statistical capacity to plug the gaps? Where are the investments to build more statistical capacity?
AI is now forcing fresh reflection on every one of these “how” questions through AI for data and data for AI. This year’s Measuring Development Conference, “Open Science in the Age of AI: Balancing Privacy and Transparency,” asked how a decade of open science practices needs to evolve as AI reshapes what’s possible. Four takeaways from the conference stood out for data for development and resonate with Open Data Watch’s work in measuring development.
1. AI-readiness begins with open data
A common question is what AI can do with official statistics. But a deeper question is whether our official statistics are ready for AI. Usually, the answer is “Not yet”.
What makes data AI-ready is much the same as what makes data useful to humans: clear metadata, consistent definitions, machine-readable formats, stable APIs, documented provenance.
How this resonates: The Open Data Inventory (ODIN) by Open Data Watch assesses exactly these dimensions across over 195 countries. As an independent NGO assessment, freely available to use and explore, ODIN gives statistical agencies, policymakers, donors, journalists, and citizens a common, trustworthy benchmark for where data systems stand and where they need to go to be ready for the next stage in the AI revolution.
2. Engagement drives transparency; culture beats tools
Ted Miguel’s keynote made an important point that carries over from academic research to the day-to-day work of official statistics: quality and disclosure both rise when researchers feel others are engaging with their work. Demand and use are the gravity that pulls transparency along.
The same logic applies to official statistics. NSOs publishing into a vacuum have weaker incentives to invest in metadata and disaggregation than those whose data are actively used. LLMs can help with this discovery process by lowering the cost of engagement and signaling to researchers that their work is being found and interpreted. But Miguel brought in a necessary qualification to the discussion around the potential of AI: tools are only tools. An open data culture inside research institutions and statistical agencies will outperform any portal, dashboard, or chatbot.
How this resonates: Our work on the enabling environment for gender data use and on intersectional data capacity analysis for development impact is grounded in this premise of cultural change and resilience. Through these publications, our technical assistance for ODIN and our work in networks such as the Collaborative on Citizen Data and others, we emphasize the importance of engaging with users and thinking of demand first in order to put the open data and feedback mechanisms in place to avoid the hard work of official statistics going unused.
3. MCP servers to better serve development data
As LLMs become an important interface for everyday questions, statistical agencies face a problem: their data are being summarized, paraphrased, and sometimes invented by models that have no live connection to the source. Model Context Protocol (MCP) servers seek to solve this problem by offering a standard way to let AI systems query official statistics directly, so answers come from the authoritative agency rather than the model’s memory.
Joao Azevedo, UNICEF’s Chief Statistician, shared a striking number during the conference: more than 30 MCP servers for official statistics have launched in the last 18 months, with about 30% from official governments or agencies. Instead of waiting passively to see how AI handles their data, some statistical systems are actively shaping the pipeline of bringing data to users.
How this resonates: This discussion echoes the flurry of open data portals that countries have set up in the post-SDG era in order to publish development data more openly. As with MCP servers, this was in response to new demands and technical capacities but needed strong standards around accessibility and metadata to function effectively as a means of communicating with users. ODW and PARIS21 have written about the challenges that particularly low- and middle-income countries face in implementing these pages and many of these challenges will also face NSOs when considering how to implement AI tools.
4. Coordination challenges
The level of engagement that the conference identified with AI tools to produce and share data is to be encouraged, particularly in low- and middle-income countries. But, as Azevedo and others noted, many of these efforts are unfolding without international guidance, and coverage and quality vary widely. We risk reproducing in developing AI tools and standards the fragmentation that open data has spent a decade addressing.
How this resonates: We find that coordination challenges do not stop at AI tools. Development finance flows still need better reporting to capture the investments in statistical capacity, such as those highlighted on the Clearinghouse for Financing Development Data. Forthcoming work by ODW on the Inclusive Data Compass will shine a light on the availability and openness of data on population groups like people with disabilities, migrants, and LGBT populations to make sure they are reflected in policy discussions. Throughout, the Data Value Chain serves as the organizing principle for connecting production to use so tools serve users and goals.
What’s next for ODW
AI raises the stakes for getting the foundations right, but the goal has not changed: better evidence for better policies and decisions. ODW is increasingly working at the link between data and policy by expanding from producers and disseminators to users and policymakers and sharpening our understanding of what makes data impactful.
That also means keeping AI advancement on its toes. Accuracy, completeness, reliability, relevance, and timeliness are necessary but no longer the ceiling. The real measure is whether data are improving lives.
