- Google launched a study that describes how much energy and water uses its AI
- Experts say that figures lack important context
- Google insists that these figures represent the average user experience
A new Google study states that its Gemini AI model only uses very minimal water and energy for each notice, with the average use of around 5 drops (0.26 milliliters), equivalent electricity used for 9 seconds of television observation (approximately 0.24 watts), resulting in around 0.003 grams of CO2 emissions.
However, experts have been quick to dispute statements with The edge Claiming that Google omitted the key data points in its study, drastically underlining the model’s environmental impacts.
While models and data centers have become more efficient, it seems that there is more in the history of what Google is leaving.
The tip of the iceberg
One of the authors of an article cited in the study, said Shaolei Ren, an associate professor of electrical engineering at the University of California to the publication; “They are hiding critical information. This really spreads the wrong message.”
The models of AI as Gemini are backed by data centers: huge warehouses full of servers that consume intense amounts of water and energy, striving local resources.
Governments around the world have been sanctioning the construction of these data centers, despite the destruction they could lead to the local countryside, and it is likely that consumers are those who pay for the extra energy used.
One of the biggest concerns with Google’s study is that it omits the indirect use of water in estimates, which form the majority of the use of AI. While the figures are technically correct, the missing context of the use of extreme energy paints a misleading image.
The study only analyzes the water used by the data centers to cool their servers, but outstanding is the electricity that these data centers demand, which in turn leads to new gas and nuclear plants, which also cool their water systems or use steam to convert the turbines.
However, water is not the only metric that Google misrepresented, and the document only describes a measure of carbon emissions ‘market -based’, which compensates for the figure using Google promises to use renewable energy to support energy networks. Savannah Goodman, director of advanced energy laboratories Techradar pro,
“We hope to share environmental metrics that are representative of the behavior of a typical user, and reasonably comparable to time. However, with the rapidly evolving landscape of AI model architectures and the behavior of the user of the AI assistant, there are atypical values of small subset subsections of indications served by models with low use or with high token counts.”
“To share metrics that represent the experience of a typical user and are solid for this field in rapid evolution, we choose to measure the metrics for the median notice, which is robust for extreme values and provides a more precise reflection of the energy impact of a typical warning.”
