Integrating environmental impact of ai in a data center

Integrating environmental impact of ai in a data center

Séquence du séminaire les Rencontres de Statistique Appliquée 

Brainpower Meets Machine Power: Chat-GPT synergies with social scientists

by Paul Gay (Pau University and Paris-Saclay University)

This presentation is an introduction to the environmental impact of machine learning. It will describe the different impacts from the digital sector: carbon and water footprint, metal consumption, ecotoxity, etc. The main findings and orders of magnitude from the state of the art will be provided as well the current limits and uncertainty in the estimation of these impacts. It will be shown that these impacts can be analysed at different scopes: a program, a computing device, an data science project in a data center, the use of an open source model by a community... Long term impacts will be also covered through the concepts of the Jevons effects and other indirect effects. The last part will give entry points to available tools and methods that data scientists can use to measure and reduce the impact of their projects. These include measuring the energy consumption of a piece of software, and considering larger and more general impact thanks to life cycle analysis methodologies.