AI technology is radically transforming industries around the globe, and the environmental impacts of its widespread adoption are just now starting to come to light. Recent discussions chaired by professors such as Jon Whittle have highlighted a growing demand for transparency on AI’s environmental impact. While data centers are growing increasingly vital, they’re some of the biggest consumers of water and energy in the country. This increasing reliance creates significant and dangerous strains on our natural resources.
A recent report commissioned by Google reveals a particularly surprising one. It shows that the average text prompt created with its Gemini AI uses an estimated five drops of water. Furthermore, the energy used for that same response is roughly the energy of nine seconds of TV. This little nugget of wisdom makes one wonder about the sustainability of this AI technology, especially given the skyrocketing demand.
A new 2023 research paper further complicates this picture. It takes a swing at estimating ChatGPT-3’s water consumption, calculating that approximately 500 milliliters of water goes into generating between 10 and 50 medium/long-length responses. As AI models continue to increase in sophistication, we need to keep a close eye on their environmental impacts.
Water Consumption in Data Centers
In addition, data centers are the unsung heroes of enabling AI applications. For all their advances, they pose significant new threats to our water resources. In Sydney alone, data centers now use the same amount of drinking water that it would take to fill 1,400 Olympic swimming pools each year. Recent projections are chilling. Within the next ten years, this number could rise to one fourth of Sydney’s entire yearly drinking water budget.
The Pawsey Supercomputing Research Centre, led by CEO Mark Stickells, is making strides toward sustainability with its geothermal cooling system. This new, innovative approach cools the systems much more effectively. It conserves five billion liters of water annually. Still, Stickells admits that these developments are only scratching the surface. A highly coordinated and significantly larger effort is needed to train the AI models we’re seeing today with effectiveness.
“But this is only a fraction of the size of the systems that are training the AI models of today.” – Mark Stickells
Stickells elaborates on this to highlight the need to unpack the infrastructure behind AI technology. Each engagement with AI has an unseen cost as it consumes energy and water.
“So every time we pick up our phone to do something very cool with AI, I think we should be thinking about the infrastructure, the power, the water, the service that’s needed to provide that to you.” – Mark Stickells
Alternative Water Solutions
As cities continue to face pressure from the influx of data centers, novel approaches to ensuring a water supply are being sought. To help achieve this, Sydney Water is exploring advanced wastewater treatment technologies as an indirect source for delivering high-quality water to these facilities. Through this initiative, EPA can do much to relieve the pressure on our drinking water supplies. Simultaneously, it makes sure that data centers have what they need to operate.
Danielle Francis from the Water Services Association of Australia on her great opportunity She envisions Australia as a global leader in resource efficient data centers. With more than 260 data centers across the country—about a third situated in Sydney—the potential for economic growth is substantial.
“The growth of data centres creates a unique opportunity to help grow our national economy and productivity, but it will create extra demand for valuable water resources.” – Danielle Francis
As data centers continue to grow, it’s important to be intentional about where they’re sited and how much resources they consume. As Wayne Rylands told us, the location of these resources is key to the most effective administration.
“It’s about the location of the data centres.” – Wayne Rylands
The Need for Comprehensive Assessment
They argue that to truly grasp AI’s environmental cost, we need a full life cycle evaluation. Professor Jon Whittle implores stakeholders to look past what the technology might do right now, and focus on wider factors. His argument is that it takes lots of energy and resources to train AI models.
“When you look at the environmental impact of AI, you have to look at the full life cycle,” – Professor Jon Whittle
Whittle’s grievances with the state run deeper than the construction of data centers. These mega-facilities can cover up to 25 football pitches, often causing highly significant environmental impacts.
“There’s building the data centres which can have a big environmental impact because they can be up to the size of 25 football pitches.” – Professor Jon Whittle
Further, e-waste produced from obsolete tech is an added problem to be solved as the field of AI continues to grow.
“And there’s also the … e-waste.” – Professor Jon Whittle
The discussion around AI’s environmental impact is rapidly becoming a concern with the advancement of technology and increase in demand. Stakeholders need to focus on transparency and take sustainable measures to reduce negative impacts on water resources and energy use.
