Research Areas

We use the AI Supply Chain proposed in “Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model“ by Sasha Luccioni, Sylvain Viguier, and Anne-Laure Ligozat as a starting reference point to map the planetary justice impacts of AI and organize our research projects and campaigns. Each stage corresponds to a research area.

Building on this framework, we will map specific substages and stakeholders for each main stage of the AI supply chain. This includes identifying key actors, infrastructures, labor conditions, environmental costs, and socio-political dynamics within each phase—ranging from resource extraction and hardware manufacturing to model training, deployment, and waste management. By doing so, we aim to trace the asymmetries of power, labor, and environmental responsibility across the AI lifecycle, ensuring that issues of justice, accountability, and sustainability are foregrounded in AI governance and advocacy efforts.

Illustration of a pile of ore.
Illustration of a landfill.
Illustrated arrow pointing right.
Illustrated arrow pointing left.
Sketch of a factory with smokestacks emitting smoke
Abstract illustration of people crossing a zebra crosswalk
Illustrated arrow pointing right
Illustrated arrow pointing left
Illustration of a microchip with multiple pins
Illustrated arrow pointing down.
Illustration of a data centre with servers along a long corridor.
Illustration of extracted ore.
Illustration of factory smokestacks emitting smoke
Illustration of a microchip with numerous connected pins
Illustration of a data centre
Sketch of stick figures walking on a zebra crosswalk
Illustration of a landfill with wavy smell signs.

Active Projects

Model Training

White square

This project, situated within the Model Training research area of the AI + Planetary Justice Alliance, investigates how AI training datasets in agriculture embed epistemic hierarchies that privilege Global North ways of knowing. Focusing on India, it critically examines how datasets used to train agricultural models often reflect standardized, techno-scientific approaches—while sidelining local, traditional, and Indigenous knowledge systems.