In his recent blog post titled “The Gentle Singularity,” Sam Altman, CEO of OpenAI, spoke about ChatGPT’s water consumption. He estimates that a text query uses about 0.000085 gallons of water, or barely a fifteenth of a teaspoon.
On the energy side, each interaction consumes 0.34 watt-hours, equivalent to an LED bulb being on for a few minutes. These figures aim to reassure about AI’s ecological footprint while promoting progress toward a digital super-intelligence. But… is it true?
ChatGPT consumption: Figures disputed by experts
Does a ChatGPT query amount to consuming just a few drops of water? Sam Altman’s data is sparking debate…
A study published by the Washington Post in 2023 reveals that a 100-word email generated by GPT-4 consumes as much energy as a bottle of water. This disparity is mainly explained by data center infrastructure.
In Washington State, the abundance of water resources allows water to be used to cool servers, while in Texas, water scarcity encourages the use of electric air conditioning systems.
The energy consumed also depends on the nature of the query. For example, a simple “hello” to ChatGPT will require less energy than writing a 5,000-word essay. One thing is certain: AI has an impact on water consumption.
AI, a growing energy-hungry giant
Artificial intelligence already accounts for 20% of the electricity consumed by data centers, a figure that could climb to 70% in the coming years.
This explosion in energy and water needs, essential for cooling servers, is raising concerns. Sam Altman promises a reduction in AI costs through automation, but the necessary resources remain a major challenge.
AI therefore questions our resource management. A simple question asked to ChatGPT could have a higher environmental cost than we think. At a time when artificial intelligence is redefining our daily lives, rethinking its ecological footprint is becoming a priority.
Also read on this topic: How to use AI without (too much) polluting the planet? 5 tips for responsible use
Image and video generation: worse than text generation?
Text queries are not the only ones that have an impact on the environment.
A study conducted by Hugging Face and Carnegie Mellon University in 2023 reveals that creating an image using OpenAI models is equivalent to the energy consumption of a full smartphone charge. This finding highlights the growing impact of various artificial intelligence applications.
As for the environmental impact of videos, although significant implications can be envisioned, no concrete study has yet been conducted on this subject.
Source:
