Towards Socially and Environmentally Responsible AI

Large AI models create unequal environmental harm: The paper finds that existing systems for distributing AI computing tasks disproportionately concentrate energy consumption and environmental damage in certain regions, often those that are already marginalized, creating what researchers call "environmental inequity."

AI performance varies by geography due to resource inequality: Users in disadvantaged regions with limited computing resources consistently experience worse AI model performance compared to users in resource-rich areas, representing a form of social inequity that has been largely overlooked in responsible AI discussions.

A new solution can redistribute AI's costs fairly: The researchers propose "equitable geographical load balancing" that uses mathematical optimization to deliberately spread AI computing tasks across regions in ways that balance social and environmental costs, rather than concentrating them in a few locations.

Real-world data proves the problem and solution work: Testing on actual AI inference data showed that current systems create "disproportionately large social and environmental costs in certain regions," while the proposed equitable approach successfully "fairly balance[s] AI's negative social and environmental costs across all regions."

Responsible AI requires addressing both social and environmental justice: The paper highlights that truly responsible AI development must simultaneously consider how computing resources are distributed globally and how environmental harms are shared, rather than treating these as separate concerns.