AI: Climate Terminator Or Savior?

Image by Rootnot Creations.

The ravenous energy hunger of artificial intelligence is supercharging global electricity demand at a pace the renewable energy rollout simply can’t match.

The use of generative AI tools is booming, with ChatGPT traffic alone growing by 113% in a year;

• AI data centres may demand 10 times more power over the next five years, and these facilities require high levels of constant power 24/7, representing a particular challenge for intermittent renewable power sources;

• The growth of AI has undermined tech sector climate plans, including for Microsoft, which pledged to be “carbon negative” by 2030 only for its emissions to rise 23.4% since 2020 primarily due to AI data centres;

• About 60% of the extra data centre demand is expected to be met by burning fossil fuels, increasing global carbon emissions by about 220m tonnes;

• Keeping data centre supercomputers cool also has enormous water consumption demands, with a single conversation with ChatGPT equivalent to a standard plastic water bottle;

• AI is also being used to supercharge fossil fuel production, with one tool from Wood Mackenzie touted as capable of adding another trillion barrels of production from existing oil wells.

Reprogram The Machine

Just as AI can be used to ramp up fossil fuel extraction, it however can also be utilised to help optimise climate solutions.

• One study found AI improvements to power, transport and food consumption could reduce global emissions of greenhouse gases by 3.2 to 5.4bn tonnes by 2035;

• That includes improving the efficiency of renewable energy systems by greatly improving grid management and increasing the load factor of solar PV and wind by up to 20%;

• AI could also be used to make buildings more energy-efficient by automatically adjusting lighting, ventilation, heating and cooling based on weather data;

• An Extraordinary World Meteorological Congress recently called for AI to be deployed to improve early warning systems for extreme weather events linked to the climate crisis;

• In one project, meteorologists in Malawi are running AI-enabled forecasts locally to support timely early warnings for high-impact weather events.

What You Can Do

A raft of efforts are underway to reduce the emissions impact of generative AI itself.

• Users are being encouraged to choose the right AI model for the task - using simpler models for less complicated queries, and choosing more eco-friendly models;

• Instead of the large, general-purpose models, employing small models tailored to specific tasks can cut energy use by up to 90%;

• Leaderboards run by Hugging Face and ML.Energy rank AI models based on how much energy they use;

• But they cannot compare the large, closed AI models that dominate the sector, sparking a campaign to pressure AI companies to be more transparent with their data;

• Users could also run AI models on their own computer rather than the cloud, generating 400 times fewer carbon emissions;

• Language used also matters: AI anthropologist Cliff Jurkiewicz warns that the use of pleasantries like “please” and “thank you” in chat prompts could burn through about a gigawatt of energy in America alone;

• The biggest environmental benefits can come from putting pressure on AI companies rather than users, with the infrastructure used by data centres 10-15 times more significant than individual usage choices.