Who will not like to see how his face turned into a character of Ghibli or Pixar studio? Unfortunately, not everyone has a skill to make this drawing or money enough to Pay the illustratorThus, the new tools that allow us to generate such images through AI became fashionable in social networks in a few hours. The result is wonderful, but accompanied by several problems. Firstly, we will steal their work to illustrators. And not only for not hiring their services. The theft is literally because artificial intelligence is trained in images stored on the Internet. As for the second problem, this is not so much, but we must pay more attention: AI consumes a lot of energy and water And the planet is not for dosing.
For the chat bots and other generative algorithms to which we have easy access, they can perform their function, you need large data processing centers with many computers that process the huge information necessary to obtain answers in seconds or minutes. We do not see them, but they are where it is far from our house, spending a large amount of energy. As well as water, especially because they overheat a lot, so they need it to proceed through powerful Cooling systems.
Each question that we ask a GPT chat or each creation of images through AI is the costs that accumulates, because it is usually not just a question or image. Sometimes we turn to these algorithms from the need, how and when we spend gasoline, because we need to or get into the car. However, all these attractive applications of artificial intelligence force us to use it Almost as if it were a gameThis leads to many interactions for already in a row and Huge energy and water costs.
How much energy do we spend when creating images through AI?
There are several studies that analyze the consumption of energy and water, which means creating images with AI. One of them was held in 2023 by a team of scientists from The company about the arms of the face and the University of Carnegie -Mellon. The study did not undergo review, so the results should be carefully read. Despite this, they are amazing enough to think about it a little more before creating images with IA, as if it were a game.
Logically, the costs depend on the complexity of the action or the algorithm in question. Therefore, these scientists analyzed 88 different models, Performing various tasks that we usually perform from the house. They came to the conclusion that the greatest expenses were taken with a big difference Generation of images through artificial artThe field on average, according to your analysis, the creation of one of these images is an energy consumption equivalent to Download your mobile phone.
In addition, as a specific case, the creation of 1000 images with a powerful model, such as Stable diffusion XLwill be the release from the atmosphere of the amount of carbon dioxide, equivalent to driving 6.5 km with a gasoline car.
The latter is important. Energy costs are not only economical. This also implies the issuance of a large number atmosphere Greenhouse gasesWith everything that entails a deterioration in global warming.
What about water?
In this study, so much attention was not paid to water consumption. However, some calculations have already been made. It is enough to take into account two simple data: the number of liters of water that requires KW -energy of energy, and the amount of kWh that is spent when generating images from AI.
Again, it all depends on the algorithm, so the known data is measured by fairly wide forks. For example, it is estimated that it generates one image – this is expenses between 0.01 to 0.29 kWhThe field with regard to water, the costs necessary for cooling all these processes from 1.8 to 12 liters per kWh.
This means that water costs for creating images using AI Between 0.018 and 3.45 litersThe field every time we turn our photo of the GHIBLI studio, we pull More than 17 glasses of water. And this is just an image. The problem is that we simultaneously generate many images. In fact, the average data center spends from 1.7 to 2.2 million liters of water per day.
Currently, it works in such a way that in the future AI can be more stable, but today is data. We can seem a little etching when they are calculated (never better), but the information is necessary. With her in the hand that everyone makes his own decisions.
Source: Hiper Textual
