Environmental Impact

AI systems negatively impact the environment, both through direct and indirect causes.

AI directly impacts the environment through factors including the high energy cost required to train AI models, the high volume of water required to cool hardware used to produce and maintain an AI system, and the resources required to build this hardware in the first place. The energy needed for AI's computing power can put strain on existing electrical grids, often overloading them. This can lead to power failures or require more resources to construct new grid capacity. Moreover, data centers used to house AI operations can exhaust local water resources, especially in areas more prone to droughts. Coltan, mined largely in the Democratic Republic of the Congo, is a necessary material for AI hardware. This can cause great damage to the nearby land and wildlife, whilst also exacerbating labor exploitation.

Indirect impacts include increased emissions from other industries. For example, AI may allow oil or ore companies to extract more resources and identify potentially profitable locations. In areas chosen for resource extraction, the local area is damaged and the activity contributes to higher carbon emissions.

The negative environmental impact of AI is more greatly felt in the Global South as the Global North and West reap more benefit. AI thus preserves, maintains, and exacerbates preexisting environmental inequalities.

Negative Environmental Impacts Exacerbated by AI Guide: Training an AI model can produce roughly 626,000 lbs of carbon dioxide - nearly five times the lifetime emissions of the average car. Researchers estimated that creating GPT-3 consumed 1287 megawatt hours of electricity, generating 552 tons of carbon dioxide equivalent. Further research also estimates that 700,000 litres of water may have been consumed when training the model.