The development of AI not just brings us closer to a sci-fi fictional future but also comes with a hefty cost – its water footprint. A new study, titled “Making AI Less Thirsty,” has found that a single conversation between 20-50 questions and responses using OpenAI’s ChatGPT will require up to 500 ml of water. This research was conducted by researchers from the University of Colorado Riverside and Texas at Arlington.
Microsoft is one of the main perpetrators of ever-rising water consumption, as the company utilized 185,000 gallons of water to train GPT-3. This figure is equal to the quantity of water used to cool a nuclear reactor correspondingly. Similarly, had the training taken place in any of Microsoft’s massive data centers located in Asia, the figure would have tripled.
The water used in data centers whether to cool the servers or generate electricity must be clean and fresh. This requires regular water intake, which is drawn from freshwater sources such as rivers or lakes. What’s more, the water usage to cooling the system is usually greater than that of air-cooled systems.
The researchers suggest that to help reduce the water consumption, data centers need to be more efficient in their water usage. Also, the locations in which they are hosted should be taken into consideration since the climate of the area also affects the required cooling rate.
Furthermore, OpenAI as a research & technology organization striving to ensure that artificial general intelligence (AGI) benefits all of humanity has made strides in finding ways to reduce their water consumption. The company has even gone as far as partnering with Google to create new cooling infrastructure that decreases their water usage by 97%.
The paradox of the development in AI is the grim realization that more power leads to more water usage and heightened environmental costs. Fortunately, companies such as Microsoft, OpenAI, and Google have taken steps towards making AI ‘thirsty’ free and this new research provides valuable insights into how further steps can be taken towards making AI greener for our environment.
