Energy and environmental impact

The amount of energy needed for generative AI is increasing with each new model and iteration as more energy is needed for training. More parameters means more power consumption.

This leads to increased carbon monoxide emissions and incredible strain on our existing electrical grids.

Every time someone asks ChatGPT to complete a task as simple as summarizing an email the hardware consumes energy. 5 times more energy than a simple Google search.

The demand and pace of building new data centers is unsustainable as the electricity required comes from fossil fuel-based power plants.

Microsoft recently reached a deal to reopen Three Mile Island, (the site of the worst nuclear accident in American history) to power Microsoft's data centers.

Water

Chilled water is needed to cool the hardware used for AI training absorbing the heat from the servers. It’s estimated that for each one hour a data center consumes it requires 2 liters of water for cooling. This puts an increased strain on local water supplies.

For example, Amazon has more than 100 data centers worldwide, each one containing about 50,000 servers.

What is needed?

Currently there is a lack of transparency and accountability regarding the environmental impact of AI. New laws are needed that ensure the creation of AI are in line with environmental concerns.

At minimum, regulations requiring data centers recycle water for cooling. As well as ethical practices for the disposal of electronic waste, including lead, mercury and cadmium.

This is not comprehensive. The study of AI and its impact on the environment and our resources is new. But so is AI. In just 2 years AI use has exploded. I am not an expert in environmental science or even generative AI. I am simply an empathetic human who sees the need to help make everyone aware of the environmental impact of AI.

Written by an actual human without any use of AI

Sources

MIT News
Earth.org
National Education Association
NPR
Alliance for Science