The General Impacts of Environmental AI

AI has become necessary to many aspects of our lives, but its growing popularity has large environmental costs. Training and running artificial intelligence systems need considerable amounts of computing power and electricity leading to the emission of large amounts of carbon dioxide and largely contributing to climate change.

The impact of AI systems is going to be from the applications they are built for, not really the cost of training.

Different AI systems, such as the ones used in research, need different amounts of computing power, therefore producing different amounts of emissions. For example, training OpenAI's GPT-3 model alone generated around 500 tons of carbon dioxide. In contrast, simpler models result in significantly lower emissions.

AI is bringing new and exciting opportunities which range from improving worker productivity to advancing scientific research. While this is hard to ignore.AI requires a staggering amount of electricity and water.AI also releases loads of carbon dioxide. This is because most of the energy used by AI is from non-renewable resources. Because AI requires a lot of water to cool the systems it is putting a few of the water supplies at risk. Furthermore, there is the environmental cost of transporting the hardware to AI training centers releasing even more carbon dioxide.

To put this into perspective, a calculation revealed that one single question to ChatGPT produces about 4.32 grams of CO2.While this does not seem huge individually, considering the amount of questions ChatGPT receives every day it adds up to huge amounts of CO2.For example 15 questions to ChatGPT adds up to one hour of video streaming,139 questions is equal to one load of laundry washed and dried on a clothesline and 92,593 questions is equal to one round trip flight from San Francisco to Seattle.