We all know artificial intelligence – ChatGPT, Google Gemini, Grammarly, Photomath – the magical programs that can generate essays, study guides, and resources. But this amazing tool has some serious environmental repercussions. You may have seen people online discussing how to use AI responsibly and ethically, but that conversation has not included how to use AI in an environmentally sustainable way.
Every question asked to AI is routed to a data center and then fed into the AI system, which is a longer and more energy intensive process than normal internet use. Due to the increased energy consumption, AI servers heat up more than normal servers, so they require more freshwater to cool. It is estimated by a researcher at UC Riverside that each string of AI prompts uses around 16 ounces of water.
Because AI has both direct and indirect impacts on the environment, its net impact is hard to quantify. The direct impacts are usually negative, like greenhouse gas emissions, water consumption, and pollution, but the indirect effects can be positive: AI’s prediction and pattern recognition abilities have the potential to mitigate 5-10% of global emissions over the next 5 years.
“Google says its total greenhouse gas emissions climbed nearly 50% over five years, mostly due to electricity that powers AI data centers,” states NPR.
AI results use ten times the electricity of a regular Google search, and generative AI can use up to 33 times more energy to complete a task than a task specific software. Additionally, since AI models are never off, they are always using energy even when there are no queries.
Most of our electricity still comes from fossil fuels, so this high consumption of electricity is very concerning. A study by UMass researchers found that training large AI models can use 626,000 lbs of carbon dioxide, which is roughly five times the lifetime emissions of the average car.
The number of data centers has doubled in recent years, and the increase in AI combined with the surge in electric cars and industrial production threatens US electrical grids.
According to a Forbes article, the increased usage of AI means we “need to look toward abundant natural gas and the potential of nuclear energy to meet high energy demand, in addition to constructing solar and wind farms.” However, these solutions will increase land consumption, and tapping into nuclear energy and natural gasses will increase water and air pollution.
AI models at the moment seem to have the potential to grow past our threshold for renewable electricity and our current finite computer resources, which means that emissions will only continue to rise.
This does not mean that we should stop using AI entirely, as it is an amazing resource, but we should consider the environmental costs of AI when we use it, and try to limit queries to tasks too complicated for Google.