ChatGPT has gigantic water consumption – according to study
The hype surrounding artificial intelligence is not abating. The focus is primarily on the AI language model ChatGPT. However, a new study now demonstrates the gigantic water consumption of the software.
Since the beginning of the year, people around the world have been discussing a specific tool: ChatGPT. OpenAI’s artificial intelligence hit the headlines at record speed and received a lot of positive feedback as well as criticism. Italy, for example, banned its use due to data protection concerns. At the same time, researchers fear that development could get out of control.
But one aspect has been overlooked so far: The ecological footprint of tools like ChatGPT. This question scientists now asked themselves and found interesting results. To do this, they examined the energy requirements and thus also the water consumption of AI. The investigation included both the OpenAI algorithm and Google Bard.
ChatGPT and Bard water consumption increases with each server request
The results seem startling. The training of the GPT-3 algorithm (i.e. the predecessor of GPT-4, which ChatGPT currently uses) required around 700,000 liters of fresh water. Any conversation with the artificial intelligence is like pouring a large bottle of water down the drain.
And this at a time when we should rethink our consumption in the light of climate change. Because of the warming, the groundwater levels sink and there are more droughts.
If we now have an algorithm like GPT-3, which consumes water from the cooling tower of a nuclear power plant for training, this aggravates the global situation in the field of drinking water supply. In addition to the population, the artificial intelligence diligently “drinks” with every interaction.
Artificial intelligence: Water consumption could be significantly higher
The researchers assume that companies train AI systems in relatively efficient data centers. On the other hand, if locations are used that are not resource-friendly, consumption could even be three times higher. In addition, the focus was primarily on consumption, i.e. the water that no longer flows back into the cycle.
With a view to the challenges of climate change, it is becoming increasingly clear that stricter rules are needed. Google alone needs about 8.7 billion liters of water for data centers in just three states. It’s definitely not sustainable.
Also interesting:
