Abstract

In recent years, the IMF has used machine learning for research and experimental applications. Some examples include nowcasting GDP through machine learning models, developing proxy measures of trade using ship positions, and coding economic time series using textual analysis. With the increasing availability of generative artificial intelligence (AI) tools, the IMF is also looking at how these techniques can improve its data services, from data collection to data dissemination. However, when used in production these tools can increase the risks to provide inaccurate, misleading, or potentially harmful data to the users. The presentation will discuss opportunities and challenges for a responsible use of AI and machine learning for data and statistics production at the IMF.