Missions

Supervision : Aurélie Bugeau (Université de Bordeaux, LaBRI), Anne-Laure Ligozat (ENSIIE, LISN)

Digital technologies have for long been seen as an asset for ecological transition, as it would allow us to bypass the material reality of industrialization. The concepts of ecological and digital transitions are thus considered to be closely linked, as "twin" transitions. Artificial Intelligence (AI) began to follow the same trajectory from the late 2010s, with its potential for environmental applications being highlighted, including in academic scientific articles [1].

At the same time, the environmental issues associated with the development of digital technologies have been highlighted by scientific research. These studies emphasize the direct environmental impacts resulting from the lifecycle of each equipment, as well as the indirect effects coming from the acceleration and profound transformation of societies [2].
For AI, the direct carbon footprint of several AI models has been estimated [3], some works even including life cycle assessments of hardware [4]. Like any environmental evaluation, these estimates are subject to high levels of variations. However, the direct impacts of digital technology in general, and AI in particular, are still a matter of debate [5].
For indirect impacts, AI research generally focuses on positive impacts, without considering the negative ones [6]. The responses to the call for consideration of negative impacts can result in scientific disagreement: the authors of [5] refer to "faulty estimates" referring to previous work]. The debate also goes beyond experimental protocols and estimates, with some authors [7] incriminating the involvement of major tech companies in the scientific community, questioning the very possibility of discussion in such opaque conditions.

Thus, we face a situation of uncertainty regarding the environmental impacts of AI, in which the different actors involved in the discussion do not succeed in imposing a consensus, invoking different ways of knowing and anticipating. In other words, we face a scientific controversy regarding the environmental impacts of AI.

Studying controversies reveals the power dynamics affecting the development of science, and leads to a more objective view of the situation. As the debates around the environmental impacts of AI are growing and researchers question the validity of certain proposals in terms of conflicts of interest [7] or scientific validity [5], it is time to have a clearer understanding of the ongoing controversy. Depending on the applicant's interests, a controversy involving AI in an application area (e.g. digital agriculture or healthcare) or concerning the installation of hardware infrastructures dedicated to AI, could be studied specifically.

The ecosystem in which AI takes place is thus subject to controversy and power struggles that need to be documented, since to our knowledge there is currently no analysis of these controversies. The aim of this postdoctoral research will be to provide a documented perspective on the various viewpoints to inform public policy and scientific decisions.


References
[1] Rolnick, D., Donti, P. L., Kaack, L. H., Kochanski, K., Lacoste, A., Sankaran, K., ... & Bengio, Y. (2022). Tackling climate change with machine learning. ACM Computing Surveys (CSUR), 55(2), 1-96.
[2] Freitag, C., Berners-Lee, M., Widdicks, K., Knowles, B., Blair, G. S., & Friday, A. (2021). The real climate and transformative impact of ICT: A critique of estimates, trends, and regulations. Patterns, 2(9).
[3] Luccioni, A. S., Viguier, S., & Ligozat, A. L. (2023). Estimating the carbon footprint of BLOOM, a 176b parameter language model. Journal of Machine Learning Research, 24(253), 1-15.
[4] Berthelot, A., Caron, E., Jay, M., & Lefèvre, L. (2024). Estimating the environmental impact of Generative-AI services using an LCA-based methodology. Procedia CIRP, 122, 707-712.
[5] Patterson, D., Gonzalez, J., Hölzle, U., Le, Q., Liang, C., Munguia, L. M., Rothchild, D., So, D. R., Texier, M. & Dean, J. (2022). The carbon footprint of machine learning training will plateau, then shrink. Computer, 55(7), 18-28.
[6] Ligozat, A. L., Lefevre, J., Bugeau, A., & Combaz, J. (2022). Unraveling the hidden environmental impacts of AI solutions for environment life cycle assessment of AI solutions. Sustainability, 14(9), 5172.
[7] Abdalla M, Wahle JF, Ruas T, Névéol A, Ducel F, Mohammad SM, Fort K. The Elephant in the Room: Analyzing the Presence of Big Tech in Natural Language Processing Research. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics - ACL 2023. 2023:13141-13160

Activités

- Collection and analysis of the corpus on AI environmental impacts and AI compute infrastructures.
- Identification of the main actors and positions (what their claims are) and of the different epistemic communities
- Interviews of stakeholders and new iteration of analysis in the light of the qualitative data collected through the interviews.

Compétences

- PhD in sociology of science and technology /sociology, economics or philosophy
or
PhD in computer science, applied mathematics, mathematics or frontier computing with experience in sociology
- Knowledge of qualitative survey methods and discourse analysis
- Knowledge on artificial intelligence algorithms would be a plus.

Contexte de travail

The candidate will join LaBRI's NeS (Numérique et Soutenabilité) research team. He or she will travel regularly between LABRI and LISN.

Le poste se situe dans un secteur relevant de la protection du potentiel scientifique et technique (PPST), et nécessite donc, conformément à la réglementation, que votre arrivée soit autorisée par l'autorité compétente du MESR.