With artificial intelligence, generating images and text has become common and accessible to everyone. However, recent studies reveal that this seemingly harmless activity comes at a significant environmental cost. They use up electrical power and gulp water – both in enormous amounts.

Water consumption

I recently learned that Chat GPT “drinks” water with every prompt we give it. I had never thought of it, but my friend mentioned it, and I looked it up. According to a study by researchers at the University of California, Riverside, training a large language model like ChatGPT consumes approximately 500 milliliters of water for every 5 to 50 user prompts or questions. As you can probably assume, this water cools the powerful supercomputers required to train these models.

In short, the growing demand for AI tools has led to an increased demand for data centers, which are also significant water consumers. Microsoft, for instance, reported a 34% increase in water consumption from 2021 to 2022, primarily attributed to its AI research. This surge in water usage has raised concerns about the sustainability of AI development, especially in water-scarce regions.

Electricity consumption

Researchers at AI startup Hugging Face conducted another study in collaboration with Carnegie Mellon University. It has shown that AI models also have a significant electricity footprint in addition to water consumption. Moreover, generating an image consumes more energy than generating a text.

The process of training and running AI models relies on powerful GPUs. Riverside found that generating an image using AI has a carbon footprint equivalent to charging a smartphone. Furthermore, the electricity consumption of AI models is expected to grow. A study by Alex de Vries of VU Amsterdam School of Business and Economics shows some concerning data. He predicts that, by 2027, the AI industry could consume as much energy as entire countries like Argentina, the Netherlands, or Sweden.

Addressing the environmental impact of AI

Addressing AI’s environmental impact is a must, considering the pace of its development. Some of the strategies include:

• Optimizing algorithms and utilizing specialized hardware to make AI models more energy-efficient.
• Powering AI infrastructure with renewable energy sources like solar and wind power to reduce its environmental footprint.
• Openly discussing the environmental implications of AI models

AI will certainly continue to affect various aspects of our lives, and it’s here to stay. So, it’s crucial to address its environmental impact immediately. By promoting efficiency, transparency, and sustainable practices, we can harness the power of AI while minimizing its strain on our planet.

[via AP, Engadget]