The development of artificial intelligence or artificial intelligence (AI) is getting faster. However, behind the services that feel practical, there is a huge burden on electricity, water, and land.
As reported by Anadolu Agency, quoted Sunday, July 5, a report from the United Nations University on Friday warned that data centers supporting AI are projected to consume 945 terawatt-hours of electricity by 2030. A terawatt-hour or TWh is a unit for calculating large-scale electricity consumption.
The report, compiled by the UN University Institute for Water, Environment and Health or UNU-INWEH, said the environmental impact of AI has been "systematically mismeasured". Because, many assessments focus more on carbon emissions, but do not take into account the water and land footprint.
Water footprint refers to the amount of water used directly or indirectly to support an activity. In the report, the water footprint is used to show the AI pressure on water resources.
By 2030, the water footprint of AI-related data centers is estimated to reach 9.3 trillion liters. This figure is equivalent to the annual basic domestic water needs of 1.3 billion people in Sub-Saharan Africa.
The land footprint describes the area needed to support AI-related infrastructure. The report estimates that the land footprint of AI-related data centers could exceed 14,500 square kilometers, almost twice the area of the Jakarta metropolitan area or Jabodetabek.
"This report is not an attack on artificial intelligence," said UNU-INWEH Director Kaveh Madani.
He called for the responsible use of AI and faster steps to address the unintended impacts of the technology.
Anadolu reported that global data centers consume around 448 TWh of electricity by 2025. If considered as a country, data centers would rank 11th as the largest electricity consumer in the world.
The report also reminds that low carbon does not necessarily mean low water or low land. Some energy transitions can indeed reduce emissions, but at the same time can add pressure to water and land resources.
The report's highlights are not only directed at the training process of large AI models. According to the report, public debate has discussed too much energy to train AI models. In fact, the largest energy consumption comes from inference.
Inference is the process when an AI model that has been used answers a command or prompt from a user. This process is said to account for 80 percent to 90 percent of the total energy consumption of AI.
ChatGPT alone is estimated to process around 2.5 billion prompts per day. Its electricity needs are estimated at around 383 gigawatt-hours or GWh per year.
The environmental burden of AI also varies depending on the task. Images created by AI generally require about 1,450 times more energy than basic text classification.
The short video created by AI can even consume electricity equivalent to 200,000 times the spam classification.Perlengkapan TV & Video
The report calls for a more responsible AI ecosystem. The way is through transparency, efficiency from the design stage, environmental fairness, responsibility throughout the life cycle, global cooperation, and sustainable use.
comments