The swift automation enabled by artificial intelligence (AI) products such as OpenAIs ChatGPT has received accolades for its effectiveness in a variety of business sectors. However, there are increasing concerns about the environmental effects of artificial intelligence breakthroughs.Since its public testing period in November, ChatGPT has been frequently used for jobs that include creative writing to, coding and exam responses, but its water footprint has mostly gone unnoticed. The upkeep of data centres for AI technologies like ChatGPT necessitates a substantial amount of water. Researchers from the University of California, Riverside, pointed out that while a single 500ml bottle of water may appear insignificant when multiplied by the billions of ChatGPT users, the total water footprint for inference becomes disturbingly big.In view of the global shortage of water problems, the scientists emphasised the importance of artificially intelligent systems demonstrating social responsibility and setting an example by minimising their own water footprint. For example, training the GPT-3 model alone needed 700,000 litres of fresh water, which is similar to the amount of water used to produce around 370 BMW cars or 320 Tesla electric vehicles. The researchers projected that the recently released GPT-4 AI systems water consumption would likely increase dramatically due to its bigger model size.Due to the absence of publicly accessible information, calculating the water footprint of GPT-4 is difficult. Despite the fact that online activities using ChatGPT take place digitally, the actual data is kept in large data centres that generate a lot of heat. To prevent equipment failure, cooling systems frequently use water-intensive evaporative cooling towers. Furthermore, the water used in such structures must be clean, which means it needs fresh water to prevent corrosion and microbial growth. Data centres also use a significant amount of water for generating electricity.What is a possible solutionFortunately, AI training allows for flexibility in scheduling. Unlike web searches or streaming services, which require immediate processing, AI training may be done at any time of day. Shoalei Ren, the studys corresponding author and an associate professor of electrical and computer engineering, argues that training AI models during colder hours when evaporation-related water loss is lower might be a simple and effective way to reduce excessive water usage.Image: PixabayAIs water consumption is an important issue due to its quick growth in technological demands. Addressing this issue is critical, especially given the escalating freshwater scarcity situation, extended droughts, and ageing water infrastructure. Global warming and ongoing drought conditions around the globe have heightened worries about water usage. Drought afflicted about 300 million people in Africa, Europe, North America, and Asia in 2022, with East Africa suffering its worst drought in 40 years, even certain major parts of the United States experiencing dry conditions, and countries such as France and Portugal experiencing record-breaking droughts.
