The student portal of Université Grenoble Alpes

AI has experienced unprecedented growth since the advent of deep learning techniques in the 2010s, and particularly in the last five years with the development of generative AI. This development has only been possible thanks to the convergence of two factors: on the one hand, the development of considerable computing resources, and on the other hand, the availability of gigantic datasets without which it would be impossible to train current AI models.

As a result, the operation of these AI systems relies on a significant physical digital infrastructure, generally made largely invisible to users, and whose manufacture and use, among other things, have a considerable environmental impact.

In this presentation, we will provide some elements to help understand this impact. We will show how this impact manifests itself both at the global level (for example, in the case of greenhouse gas emissions), but also at a very local level in terms of conflicts over the use of water resources, for example.

Finally, we will outline some trends and discuss the question of whether AI compatible with planetary boundaries is possible.

This conference is part of the "Science infused" series, supported by the UGA university library network.