Artificial Intelligence, the most powerful technological transformation underway worldwide, presents enormous opportunities as well as significant risks. However, it has a significant environmental impact, due to two factors: energy consumption and CO2 emissions.
The former, in particular, are necessary for training and running models. Here's an example: training the GPT-3 model (now superseded by other, even more energy-intensive models). This translates into a consumption of 1.287 megawatt hours (MVh), equal to the average energy consumption of 120 Italian homes.
To this energy consumption we must add the one recorded during the use of Artificial Intelligence, with a single request that can translate into an energy expenditure of 0,01 to 1 watt-hour (Wh). This may seem very little, but if you multiply this figure by the millions of requests per second addressed to Artificial Intelligence around the world, do you have any idea what level of energy consumption, all growing exponentially, can be achieved.
As for CO02 emissions, there are several scientific studies that quantify quite precisely the pollution produced by Artificial Intelligence. A recent study published in the magazineNature Machine Intelligences it shows that Training a single AI model can emit up to 284 tons of CO02, the equivalent of 125 transatlantic flights. Another study reports that a single training session produces the same amount of CO2 as 123 gasoline-powered cars for a year.
The multiplier effect of AI pollution could soon become a global environmental problem. According to the International Energy Agency (IEA), data centers currently consume an estimated approximately 180 million tons of emissions, less than 1 percent of the total. This number, according to some forecasts, will rise to 500 million by 2035, weighing on the overall budget. The IEA also estimates that electricity demand could double by 2030, also due to consumption related to Artificial Intelligence.
CO2 production from AI research is significantly higher than from other digital activities: a Google search is worth about 0,25 grams of CO2; a single prompt a ChaptGPT translates to 2-4 grams of CO2.
Faced with such disturbing prospects, which cannot fail to have consequences also on the climate crisis, There are, however, some positive aspects to keep in mind. All the major companies committed to dominating the sector, and the thriving market, of Artificial Intelligence (starting with OpenAi, Google, and Microsoft), are investing to improve energy efficiency both in the training phase and during research. And they are aiming to power data centers with energy from renewables.
Furthermore, and this is also a positive element, Artificial Intelligence, applied to the energy sector, will certainly make an important contribution to optimizing the entire system and orienting it increasingly towards renewable sources. But the fact remains that the pollution produced by Artificial Intelligence is only one chapter in the long book on the relationship between innovation and sustainability, a problem that science, politics, and humanity in general, with its lifestyles, have a duty to address.
How is energy consumption changing around the world? And how far are we really moving toward sustainability? Why are greenhouse gas emissions continuing to rise? You'll find the answers. in this book.
Read also:
- How much energy does artificial intelligence consume?
- How to make decisions better than artificial intelligence
- How to Get a Personal Assistant with Artificial Intelligence
Want to see a selection of our news?
- Sign up to our newsletter clicking here;
- We are also up Google News , activate the star to add us to your favorite sources;
- Follow us on Facebook, Social media coordinator e Pinterest.