Statistics and Computational Methods Seminar Series - 2024/25

Speaker: Cristiano Tessitore (Eurostat)

Title: Generative AI for Official Statistics

Abstract

Artificial Intelligence (AI) and in particular Generative AI presents both opportunities and challenges for official statistics, reshaping all the phases of the statistical production (from data collection to dissemination). The speech will begin by framing the context of AI’s historical development, outlining how machine learning and deep learning have led to the rise of generative models. We will then explore basic AI concepts, including model training, fine-tuning, and inference, providing a foundational understanding of how these systems operate. A key focus will be on prompting techniques for Large Language Models. Copyright and intellectual property (IP) issues may emerge, raising concerns about data ownership and content attribution. Additionally, we will address ethical challenges and bias, emphasizing the risks in AI-generated contents. The regulatory landscape is also evolving, with frameworks such as the AI Act setting new standards in the European Union. Finally, we will discuss use cases, showcasing how AI can enhance the statistical process and the statistical products.