• btc = $70 061.00 - 490.50 (-0.70 %)

  • eth = $3 505.83 -32.94 (-0.93 %)

  • ton = $7.26 0.49 (7.29 %)

  • btc = $70 061.00 - 490.50 (-0.70 %)

  • eth = $3 505.83 -32.94 (-0.93 %)

  • ton = $7.26 0.49 (7.29 %)

13 Apr, 2023
2 min time to read

A recently published paper reveals that the process of training artificial intelligence algorithms not only demands enormous amounts of energy, but also an exorbitant quantity of water.

Researchers from the University of Colorado Riverside and the University of Texas Arlington have released a yet-to-be-peer-reviewed paper titled "Making AI Less Thirsty". The paper examines the environmental impact of artificial intelligence (AI) training, which requires substantial amounts of electricity and water to cool the data centers used for training.

The researchers conducted an investigation into the amount of water needed to cool the data processing centers employed by companies like OpenAI and Google. They found that Microsoft, which is partnered with OpenAI, consumed an astounding 185,000 gallons of water solely to train the GPT-3 model. The researchers calculated that this amount of water is equivalent to the quantity needed to cool a nuclear reactor.

The paper also revealed that the water Microsoft used to cool its US-based data centers while training GPT-3 was enough to produce "370 BMW cars or 320 Tesla electric vehicles." The researchers further noted that if they had trained the model in the company's data centers in Asia, which are even larger, "these numbers would have been tripled."

Moreover, the researchers highlighted that ChatGPT, a conversational AI model, requires the equivalent of a 500ml bottle of water for a simple conversation of about 20-50 questions and answers. While a single bottle of water might not seem like much, the paper points out that ChatGPT has billions of users, and the total combined water footprint for inference is still extremely large.

The paper suggests that companies such as Google and OpenAI should take social responsibility and lead by example by addressing their own water footprint, given the significant water usage associated with AI training. The researchers note that this would be a first step in quenching AI's unslakable "thirst."

Despite this recommendation, the researchers do not provide any clear solutions to address the issue of water usage in AI training. However, the paper sheds light on the significant environmental impact of AI and highlights the need for more sustainable approaches to AI development.