According to a new report from the United Nations, not using expressions such as “please” and “thank you” when talking to artificial intelligence chatbots could save enough energy to meet the annual electricity needs of 760 thousand people in Sub-Saharan Africa. The research revealed the often overlooked environmental impact of artificial intelligence.
According to the news of South China Morning Post; The report published by the Water, Environment and Health Institute of the United Nations University shows that the real cost of artificial intelligence is not limited to carbon emissions; He emphasized that it also includes water consumption, land use and electronic waste.
Huge energy consumption of data centers
According to the news in Gazete Oksijen, it was estimated that data centers, which constitute the physical infrastructure of artificial intelligence, consumed approximately 448 terawatt hours (TWh) of electricity last year. According to the report, if data centers were a country, they would rank 11th in the world in energy consumption, approximately at the same level as France.
It was stated that approximately 4.5 trillion liters of water were used to produce this amount of electricity. It was stated that this figure was sufficient to fill 1.8 million Olympic swimming pools.
It was also noted that artificial intelligence infrastructure requires an area of approximately 6,900 square kilometers, which is approximately 4.5 times the surface area of Greater London.
The environmental impact of artificial intelligence is growing
The report stated that this environmental impact is expected to grow further as the use of artificial intelligence increases rapidly. It is predicted that artificial intelligence-driven electricity consumption will reach 945 TWh by 2030 and will constitute approximately 3 percent of global electricity demand.
In addition, it was stated that artificial intelligence infrastructure could produce up to 2.5 million tons of electronic waste annually by 2030. It was reported that this amount is equivalent to approximately 250 Eiffel Towers being thrown away every year.
The load is not distributed evenly
According to the report, the environmental burden of artificial intelligence infrastructure is not shared equally across the world. It was stated that the USA and China host 90 percent of the world’s artificial intelligence-focused cloud infrastructure, while smaller countries have limited say over access, pricing and data management.
It was emphasized that some small countries provide the raw materials or energy required for artificial intelligence systems, but despite this, “strategic benefits flow to other countries while taking on the burden of mining activities and electronic waste.”
More costly than training processes
The report stated that training artificial intelligence models requires a large amount of energy. For example, it was estimated that approximately 100 GWh of electricity was consumed during training of OpenAI’s GPT-5 model.
However, according to the research, daily user operations lead to much more intensive resource consumption. It was stated that ChatGPT, which processes approximately 2.5 billion requests per day, consumes approximately 383 GWh of electricity annually.
Researchers stated that to reduce this consumption, users can adopt the “short and concise usage” method. It was noted that removing phrases such as “please” and “thank you” reduced the amount of words and data ChatGPT produced by 30 percent.
According to the report, this simple behavioral change could save between 87 and 98 GWh of energy per year. It was stated that this amount is equivalent to the amount of electricity used in residences by 760 thousand people in Sub-Saharan Africa.
“Artificial intelligence is not as abstract as it seems”
The director of the institute and the lead researcher of the report, Prof. Kaveh Madani said that artificial intelligence is not as “abstract” technology as it is thought.
“Artificial intelligence is often described as a weightless, virtual technology. However, in reality, behind every prompt, image or video lies a growing infrastructure consisting of energy systems, water use, land consumption, mineral extraction and electronic waste,” Madani said.
The report also emphasized that these “hidden environmental costs” need to be made visible before they become unmanageable.
Governments are also taking action
It was stated that governments around the world are taking steps to meet the energy needs of artificial intelligence. It was reminded that in May, China launched an action plan aimed at “world leadership” in the field of clean energy for artificial intelligence IT infrastructure by 2030. Source: Newspaper Oxygen
