Photo by Ella Ivanescu on Unsplash
The use of AI is killing Earth. When people think about artificial intelligence, they typically consider its advancements in the professional world, ChatGPT, the automation of their daily tasks, or even self-driving cars. What most people don’t realize is the immense energy required to power even the simplest of models and the drastic impact it has on the planet. Each question asked of an AI model consumes real energy and adds continuous harm to our planet.
Before discussing the environmental consequences of generative AI, it is essential to consider how it operates and the energy requirements that underpin its development. Each query sent to the AI model goes through computation, during which the model breaks down the input and uses its parameters to perform large-scale calculations to determine the best response. Popular models such as GPT-3, Llama 3, and GPT-4o have billions–if not trillions–of parameters, which demand a large amount of energy to power, as stated by Adam Zewe from MIT. Unfortunately, the size of these models is growing, and they require more electricity than before.
Growth in Generative Model Parameters Graph by Tejas Raja on Research Gate
Once we start asking where these models are hosted, the reason behind the environmental deterioration caused by AI becomes clear.
Data centers.
They’re large-scale warehouses that contain hundreds of servers, GPUs, graphics processing units, and TPUs, tensor processing units, all of which require immense cooling. These data centers power the world, providing cloud computing services, network equipment, and supporting the training and deployment of AI models. According to data scientists from Google and the University of California, Berkeley, the training of GPT-3 used approximately 1,200 megawatt-hours of electricity, which is enough to power 120 homes in the United States for a year. All while generating approximately 552 metric tons of carbon dioxide, enough to power 120 typical passenger vehicles in the USA each year.
However, considering the electricity use of these large data centers isn’t going to give us the complete environmental picture. Another potential culprit should also be considered: water. Because these large data centers must be kept at a cool enough temperature to run at maximum efficiency, cooling them requires a significant amount of water. Researchers at Lenovo concluded that, on average, data centers use approximately 300,000 gallons of water daily to maintain their cooling systems. Many of these centers are located in already water-scarce regions, such as Silicon Valley and other tech areas. Here, the water usage by data centers exacerbates water scarcity, putting pressure on residents and local ecosystems.
Artificial intelligence is marketed to solve the world’s most significant problems, helping to combat some of the most detrimental diseases, such as cancer, Parkinson’s, and Alzheimer’s, while making life easier. However, behind that promise lies a disturbing reality: the brighter these models become, the more energy they require, and hence, the more CO₂ they emit. People celebrate ChatGPT for writing essays, coding apps, and assisting with daily tasks, but fail to ask whether all of this is worth it. Is it okay to emit hundreds, if not thousands, of metric tons of CO₂ and drain millions of gallons of water just to generate poems or build apps faster? If “intelligence” comes at this price, can it even be called progress?
AI isn’t evil, and few are saying it should be eliminated. What is needed is accountability from both the companies and the users. Companies should start disclosing the emissions from training their models more clearly and adopt cleaner forms of cooling and training. Users, on the other hand, should be aware that each query they send has a hidden cost. AI usage requires rethinking. Just because we possess power does not mean we should overuse it. If AI development and usage keep at their current pace without much consideration for the environment, artificial intelligence won’t be our salvation, but our downfall.
Be First to Comment