A recent study about artificial intelligence (AI) has raised environmental concerns as the AI ChatGPT requires 500 ml of water to answer 20-50 questions. The study emphasizes the need for sustainable practices and uncovers the hard truth of the environmental cost of training and deploying large AI models.
AI Models- Thirsty Models
GPT-3 and GPT-4 are two large models that have secret hidden water footprints. Training GPT-3 at US’s Microsoft data center consumes almost 700,000 liters of fresh water which is equivalent to producing hundreds of cars. For a simple conversation ChatGPT, requires a 500ml of water, considering its billions of users.
Water Consumption of Data Centers
Data centers are all water-thirsty and energy-intensive, data center is where AI models reside. 12.7 billion liters of freshwater alone is consumed by Google’s US data to keep it cool in 2021. A significant amount of water is required for power generation and they only rely on water-intensive cooling systems.
Data centers produce heat which needs water for cooling it down. Pure water is required for power generation at the data centers and for water-intensive evaporation for cooling towers to prevent corrosion and microbial growth.
A search for solutions
There are suggestions to train the AI models during the cooler hours of the day to prevent water loss through evaporation. Addressing and uncovering water footprint AI models is crucial as there have already been freshwater scarcity and aging water infrastructure, the very reasons that have led to the development of environment-friendly AI models.
AI, an Environmental Concern
With freshwater scarcity and droughts increasing every day around the world, sustainable practice is essential to bring them into practice. Vital mindfulness is required as we explore the potential of AI models like ChatGPT which is and can impact our environment.
In publishing and graphic design, Lorem ipsum is a placeholder text commonly used to demonstrate the visual form of a document or a typeface without relying on meaningful content.