The most senior statisticians of countries across and beyond the UNECE region meet in Geneva this week for the 72nd annual Plenary Session of the Conference of European Statisticians. Kicking off their deliberations, a shared session with UN-GGIM Europe, the United Nations’ European geospatial body, will take an in-depth look at the role of Artificial Intelligence (AI) and Large Language Models (LLMs) in the future of statistics and geospatial data. In a high-level seminar convened by Statistics Canada, the German Federal Statistical Office, and UK Ordnance Survey, the representatives of national statistical offices and national cadastre and mapping agencies will share practical experiences, look at what we’ve learned so far, and reflect on how we can safely harness the potential of AI and LLMs.
Transforming how official statistics are made
Many people are now familiar with some of the possibilities presented by LLMs thanks to the rise of ChatGPT and other tools. These powerful tools are now being employed to reshape statistical production, making the day-to-day tasks of statisticians more efficient and accurate.
Statistical offices, like any other organization, can benefit from improved operational efficiency and workflow optimization of mundane office tasks, such as reading emails, writing reports, and attending meetings. Beyond this, the potential is great in the field of statistics, as the tools can be used to help generate analyses from large datasets, produce and read data tables, and translate complex statistical facts into easily-understandable language for accessible dissemination.
For example, Statistics Norway is enhancing efficiency with internal chatbots, while the Turkish Statistical Institute, TurkStat, leverages AI for better data collection, processing, and analysis. In Mexico, advanced image segmentation methods are guiding the integration of geospatial and demographic data for urban classification. UNCTAD has developed an open-source LLM application to retrieve information from a vast volume of documents efficiently.
Changing how people obtain and use statistics
The possibilities for revolutionary change don’t stop at statistical production. We may soon witness radical shifts in how people look for information and what they expect a national statistical office to be able to provide. With natural-language prompts, users may want to ask a question and receive an answer based on statistics, instead of having to find a dataset and perform their own calculations or identify the correct figures for themselves. Official statistics must rise to the challenge of ensuring that their authoritative figures, produced in accordance with internationally-agreed principles, inform the results of these prompts and searches.
Navigating the future
The HLG-MOS white paper on LLMs, produced by UNECE’s high-Level Group for the Modernisation of Official Statistics, offers a glimpse into the future, detailing current and potential applications as well as the risks associated with such transformative technology.
Important questions remain, and will be debated in the seminar. What obstacles are there to adopting AI and machine learning in statistical and mapping offices? How can we ensure the quality and transparency of AI-driven statistics? How can statistical offices upskill their workforce and attract new talent in a competitive market? What guidelines should they follow to ensure responsible AI use?
The seminar is part of the annual plenary session of the Conference of European Statisticians, which meets to discuss emerging issues, agree on priorities, and adopt new statistical standards and guidelines. This year’s session on 20-21 June includes discussions of linking up data across topics and sources to enhance the insights they offer; integrating statistical and geospatial data; and much more. The Conference is expected to endorse new statistical standards on the role of national statistical offices in achieving national climate objectives; measuring subjective poverty; and a revised Generic Statistical Information Model.