Séminaire économique - Artificial Intelligence and Statistics

Le 21 décembre 2023 a eu lieu le séminaire économique Artificial Intelligence and Statistics: navigating potentials, pitfalls, and use, organisé en collaboration avec la Société Luxembourgeoise de Statistiques.
A cette occasion, Florian Felice (Amazon Web Services & Université du Luxembourg) et Guillaume Osier (STATEC) ont présenté les sujets suivants :

Artificial Intelligence and Statistics: navigating potentials, pitfalls, and use
In recent months, Artificial Intelligence gained popularity and interest with the advances in (large) language models. In particular, the release of ChatGPT caught the attention from the public as well as industries. Large Language Models unlock new capabilities encompassing tasks ranging from the generation of text, code and images to the challenge of simplifying complex content through text summarization.
This talk aims to provide an insightful journey into the history and evolution of Artificial Intelligence. While these advancements offer unprecedented possibilities, they also bring forth challenges, including the potential for generating misleading information and perpetuating societal biases. Delving into these complexities, we will explore strategies to address some of the drawbacks. By understanding the roots of these issues, we aim to empower users to navigate responsibly through the landscape of sophisticated tools powered by AI. Our discussion will culminate in actionable suggestions for promoting the responsible and ethical use of generative AI, ensuring its utility without compromising integrity.
PowerPoint (Pdf, 2,04 Mo)

Concrete applications of ChatGPT and AI Language Models in social statistics
This talk will present the utility of AI Language Models, such as ChatGPT, in the field of social statistics. Some concrete examples of this are the coding of open-ended questions, the translation of questionnaires and other fieldwork-related documents, the writing of publications, or computer programming.
PowerPoint (Pdf, 1,02 Mo)