The amount of water used in cooling systems to prevent equipment from overheating during the development and operation of artificial intelligence models was investigated. With increasing AI applications, it is estimated that 4.2-6.6 billion cubic meters of water could be consumed by 2027.
Noting that recent research shows models like GPT-3 and BLOOM can lead to significant water and energy consumption, Assist. Prof. Dr. İnci Karakaş said, “This situation is quite concerning because difficulty in accessing potable and usable water is one of the most urgent problems shared by humanity.”
Assist. Prof. Dr. İnci Karakaş, Deputy Director of the Health Sciences Institute at Üsküdar Üniversitesi's Vocational School of Health Services (SHMYO), evaluated the issue of water resources used by artificial intelligence applications and the environment.
'Their applications offer many environmental advantages'
Stating that artificial intelligence applications offer many environmental advantages, Assist. Prof. Dr. İnci Karakaş said, “With artificial intelligence applications, energy consumption can be analyzed, potential savings areas can be identified, and energy can be utilized optimally. In terms of water resources, algorithms that determine water consumption can minimize the need for water, thus protecting limited usable water resources, and smart irrigation systems can be developed for specific crop types.”
'Models like GPT-3 and BLOOM are said to lead to significant water and energy consumption'
On the other hand, examining recent research on artificial intelligence, Dr. Karakaş noted that various models such as GPT-3 and BLOOM are stated to lead to significant water and energy consumption, and continued:
“A study conducted at the University of California estimates that with increasing artificial intelligence applications, 4.2-6.6 billion cubic meters of water could be consumed by 2027. This situation is quite concerning because difficulty in accessing potable and usable water, due to rapidly increasing population and depleting natural water resources, is one of the most urgent problems shared by humanity. Clearly stating the amount of water used in cooling systems to prevent equipment from overheating during the development and operation of various artificial intelligence models will be important from an environmental perspective. Along with the water footprint, which indicates the amount of fresh water used to produce a product or service, the carbon footprint must also be revealed to create environmentally sustainable artificial intelligence models.”
'Potable and usable water resources are limited'
Stating that potable and usable water resources are limited and unevenly distributed globally, Assist. Prof. Dr. İnci Karakaş said, “Water drawn from underground or surface water sources for agricultural or urban purposes, as well as for artificial intelligence applications, can reduce available water resources. After water is drawn from underground and surface water sources, changes in water quality contribute to water stress levels for subsequent uses. Global water resources become insufficient in terms of both quantity and quality, leading to the risk of water scarcity.”
'Water scarcity currently affects 4 billion people'
Assist. Prof. Dr. İnci Karakaş also explained that severe water scarcity currently affects 4 billion people, approximately two-thirds of the global population, for at least one month each year, stating, “Studies emphasize that unless integrated and inclusive approaches are adopted for the global water problem, almost half of the world's population could face severe water scarcity by 2030, and by 2040, approximately one in four children worldwide will live in regions experiencing severe water scarcity. Furthermore, the location and timing where artificial intelligence applications are trained and operated can also be important for water consumption.”
'Insufficient awareness has been created'
Also stating that there is insufficient awareness regarding the water resources used by artificial intelligence applications, Assist. Prof. Dr. İnci Karakaş said, “Since the amounts of water consumed during the development and operation of these applications are not clearly specified, ensuring water sustainability can be very difficult. The lack of a water footprint indication could also pose a potential obstacle for environmentally sustainable artificial intelligence applications in the future. It is important to evaluate the water footprint and carbon footprint together for sustainable artificial intelligence applications.”
'Environmental risks posed by artificial intelligence applications must be revealed'
Assist. Prof. Dr. İnci Karakaş, stating that artificial intelligence applications can be used in areas such as environmental science, health, education and research, scientific publishing, digital assistants, transportation and logistics, finance, and translation, concluded her words as follows:
“The environmental risks that artificial intelligence applications, used in many areas of our lives, may pose must be revealed. Since AI programs are complex, they require more energy than other computing systems. Determining exactly how much energy AI tools consume is extremely difficult.
'The effects of global warming could be felt more severely'
In a study conducted in 2022, it was estimated that the training of the BLOOM model resulted in 24.7 tons of carbon dioxide emissions; while equipment production and model operation resulted in 50.5 tons of carbon dioxide emissions. Carbon dioxide is seen to have the largest share among greenhouse gas emissions. With increasing industrialization and population growth, and artificial intelligence applications also raising carbon dioxide emissions, atmospheric temperatures could increase, making the effects of global warming felt more severely.
Due to global warming, sea level rise, glacier melt, severe winds and heavy rains, deterioration of air quality, reduction of natural water resources, water scarcity, and an increase in epidemic diseases could occur.”