Artificial intelligence chatbots that humanity uses to create fakes for social networks, generated images and other entertainment, threatens the environment. After all, AI consumes not only a lot of energy (for which Microsoft, for example, even restarts a closed nuclear power plant), but also needs water to cool computers.
Residents of the American Memphis have long been suffering from polluted environment. Mask's AI data center made it even worse. And his business – is not the only example of such behavior among tech giants who also fail to fulfill their environmental promises.
About how artificial intelligence and its data centers cost humanity – read in the material of Channel 24.
Water consumption – like the whole state
More and more people are using AI chatbots. And while it seems like a simple thing to do, they need a huge amount of electricity to answer users' questions. And even cooling servers in data centers harms the environment. This is stated in a study by The Washington Post, in which the publication involved scientists from the University of California at Riverside.
The study examined how much water and electricity the ChatGPT chatbot from OpenAI with the language model GPT-4, released in March 2023, consumes to write an answer of 100 words.
How much water is consumed by a chatbot in the following terms:
- at one time – 519 milliliters of water;
- Once a week for a year – 27 liters of water;
- The 16 million Americans who work and use chatbots spend 435,235,476 liters of water per week generating 100-word answers.
Last the indicator is impressive – that much water is consumed by all the households of the state of Rhode Island with a population of over a million people in a day and a half. This can be compared to Odessa or the Dnipro.
Why does AI need so much water and electricity
Every request to ChatGPT goes through a server that performs thousands of calculations to determine the best words to use in the response. Servers, usually housed in data centers, generate heat as they perform these calculations. Water is often used to cool equipment and maintain its performance. It transfers the heat produced in the data centers to the cooling tower. Where electricity is cheaper or water is scarce, air conditioners are used to cool servers.
Before chatbots can learn to fulfill requests, their training also consumes a huge amount of electricity. The large language models that allow chatbots like ChatGPT to generate realistic responses require servers to analyze millions of pieces of data.
How much electricity needed for AI and how Musk is polluting the city
A 100-word answer generated by an AI chatbot using GPT-4 requires this much electricity:
- 0.14 kWh or the operation of 14 LED bulbs in one hour;
- once a week for a year – 7.5 kWh, as much as 9.3 households consume in one hour;
- once a week during the year, 10 percent of working Americans require 121,517 MWh, equal to the electricity consumed by all households of the District of Columbia for 20 days (about 700,000 people, which is comparable to Lviv).
AI data centers also require huge amounts of electricity for other things, including cloud computing. And when these centers use water cooling in arid regions, it can deplete a precious natural resource.
Interesting!At the beginning of September, Elon Musk's new supercomputer called Colossus was launched. It runs on 100,000 H100 Nvidia GPUs. According to Musk, this is the most powerful system for training artificial intelligence in the world. In a few months, the capacity of the system is planned to be doubled. Nvidia claims that the new version of the chips – H200, with which Musk will complete the system, have twice the memory capacity and 40% more bandwidth than H100.
The main goal of Colossus – ensuring the operation of the Grok AI chatbot.
200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000In the state of Northern Virginia, where the world's largest concentration of data centers is located, local residents organized protests against the construction of data centers. According to the protesters, the centers not only do not provide enough jobs, but also “kill home values”.
In West Des Moines, Iowa, a new hotbed of data centers, water department reports showed that facilities operated by companies such as Microsoft used about 6 percent of the area's total water. After a lengthy legal battle, the Oregonian forced Google to release information about the use of its data centers in Dall, near Portland, Oregon. It turns out that data centers consume almost a quarter of all the water available in the city.
What can AI costs be compared to:
- data center Microsoft uses 700,000 liters of water during GPT-3 training;
- it takes about 7,000 liters of water to produce 0.45 kilograms of beef, which is mainly used for growing feed, cleaning the farm and for cows;
- li>
- training GPT-3 requires the same amount of water as producing 45 kilograms of beef, almost twice what the average American eats in a year;
- Meta used 22 million liters of water to prepare its open source LLaMA- 3;
- about 4,956 liters of water are needed to produce 0.45 kilograms of rice, which is grown almost entirely under water;
- the cost of water for LLaMA-3 is about 2 tons of rice – 164 Americans consume that much per year.
For example, Musk's supercomputer at full power will consume 150 MW of electricity (that much is consumed by 80,000 households) and 3.8 million liters of water per day.
Colossus' electricity is provided by 18 gas piston turbines, installed without permission of the authorities. Of course, this affects the region with an already bad ecological situation – Cancer rates in southwest Memphis, Boxton, where the data center and turbines are located, are historically high, four times the US average.
And life expectancy – 10 years less even than other parts of the city. In general, the object was built without the knowledge of the community, and the city authorities signed a non-disclosure agreement. Local activists are already calling to pay attention to the consequences of Musk's actions in the outskirts of the city.
It is interesting that Boxton – the black community that suffers the most from pollution. There are also city sewage treatment plants in the area. In addition, the data center creates a load on the Memphis water system, which is over a hundred years old.
How tech giants aren't keeping their green promises
Big tech companies have repeatedly promised to make their data centers more environmentally friendly with the help of new cooling methods. But these promises are often not fulfilled.
In July, Google published an environmental report that said carbon emissions had increased by 48%, mainly due to AI and data centers. The company also recovered only 18% of the water used – this is very far from the 120% that is the company's goal by 2030.
“Artificial intelligence can be energy-intensive, so we're constantly working to improve efficiency,” – said Kayla Wood, spokeswoman for OpenAI. Ashley Settle, a spokeswoman for Meta, said the company prioritized “the stable and efficient operation of data centers while ensuring that people can rely on the company's services.” Craig Cincotta, Microsoft's general manager, said the company “remains committed to reducing the intensity of resource extraction” and added that Microsoft is working to establish data center cooling methods that “completely eliminate water consumption”.
At the same time chip companies like Nvidia will continue to build chips that generate more kilowatts of power per server to do more computing, experts cited by The Washington Post say. According to the University of California, artificial intelligence is creating unprecedented demands on data centers that tech companies have not yet faced.
The rapid development of artificial intelligence is not only changing the behavior of people in social networks and posing challenges to the media. It is accelerating climate change, adding one more factor to a long chain stretching back to the Industrial Revolution. But unlike the era of steam locomotives and burning coal, humanity must behave more responsibly. If only because the point of no return in the climate has probably already been passed.