How much water is required by AI to engage in a brief interaction using OpenAI’s ChatGPT system, equivalent to the volume of water in a standard plastic bottle? Artificial intelligence systems exhibit a considerable thirst, consuming approximately 500 milliliters of water for each short conversation with GPT-3. This calculation includes the water utilized for cooling the servers in data centers and the water used by power plants to generate the necessary electricity.
The study that provided these estimates also highlighted the variability in water consumption by AI systems, depending on the location and operational timeframe of the computing system processing the queries. From my perspective as an academic librarian and education professor, comprehending AI extends beyond merely crafting prompts; it entails grasping the underlying infrastructure, trade-offs, and ethical considerations associated with AI.
While some individuals perceive AI as inherently detrimental due to concerns about its significant energy and water usage, it is essential to recognize that these impacts are only part of a broader narrative. By transitioning from viewing AI solely as a resource drain to understanding its true environmental footprint, origins of effects, variations, and potential mitigation strategies, individuals can make informed decisions that balance innovation with sustainability.
Behind every AI query lies a dual water usage scenario: on-site cooling of servers to dissipate heat and water consumption by power plants supplying electricity to data centers. Water utilization varies significantly based on geographical location and time of year, with certain regions relying on alternative cooling methods like immersion cooling using non-conductive fluids.
The type of AI model being employed also influences water consumption due to varying complexities and hardware requirements. Estimating the water footprint of AI can be simplified into three steps without necessitating advanced calculations: First, seek reliable research sources or official disclosures; secondly, approximate the water usage per unit of electricity for cooling and power purposes; and finally, integrate these components for an overall assessment.
Comparing the water consumption of AI queries to common daily activities such as lawn irrigation or domestic chores reveals that while generative AI does utilize water resources, its impact remains relatively modest compared to other routine practices. Nevertheless, advancements in technology such as specialized chips, efficient cooling mechanisms, and intelligent workload management can optimize water efficiency within AI systems.
Enhancing transparency through data disclosure by companies enables stakeholders such as the public, policymakers, and researchers to evaluate achievable benchmarks and make fair comparisons among service providers.