The capabilities of Artificial Intelligence (AI) have made a significant leap forward in the last few years with the major advances of generative AI. Based on the extensive training on vast data sets with billions of parameters, generative AI, notably through Large Language models, is capable of generating texts, images and videos at a level almost indistinguishable from those made by humans. Despite its recent emergence, it has rapidly permeated various facets of modern life, becoming an integral part of daily routines for a large part of society.

There is little doubt that generative AI is going to play an important role in statistical organizations. It has the potential to greatly improve efficiency and productivity of tasks  across the whole statistical production chain ranging from data and metadata collection, editing and validation to methodological and coding assistants. It can also help deliver better services to society, for example, by providing more user-friendly access to data dissemination platforms and tailor communication of official statistics to the users’ needs. These opportunities are not just theoretical, but very much real.

The High-Level Group for the Modernisation of Official Statistics (HLG-MOS) has served as an international platform for practitioners in national and international statistical organizations to exchange experiences and lessons learned to navigate this fast-evolving landscape together. Building on its white paper “Large Language Models for Official Statistics”, HLG-MOS launched several streams of works, notably catalyzed  by its project on Generative AI for official statistics (https://unece.org/sites/default/files/2024-03/HLG2023%20ProjectProposal2024%20GAI_0.pdf), aimed at taking stock and exploring applications of generative AI in statistical organizations.

This session aims to showcase the concrete examples where generative AI can add values and how various offices have implemented them in practice. The session will focus on sharing best practices for the implementation (e.g., prompt engineering, project management), associated risks, mitigation of emerging risks to quality as well as practical approach/tools (e.g., software, architecture) that are instrumental for facilitating the successful integration of this new technology in statistical organizations.