TLDR: Researchers at the MIT Climate and Sustainability Consortium are deeply investigating the environmental consequences of generative AI, revealing a substantial increase in energy and water consumption, and emphasizing the need for industry-led sustainability efforts.
Cambridge, MA – A new study by researchers at the MIT Climate and Sustainability Consortium is shedding light on the often-overlooked environmental toll of generative artificial intelligence. The findings highlight both the direct and indirect impacts of these rapidly evolving technologies on global energy grids and water resources.
Dr. Noman Bashir, the Computing & Climate Impact Fellow at the MIT Climate & Sustainability Consortium, emphasized the dramatic difference in energy consumption. ‘AI data centers are much more power intensive,’ stated Dr. Bashir. He elaborated that an AI data center can be up to ten times more power-intensive than a conventional data center, leading to significantly higher power demands. To illustrate this, Dr. Bashir provided stark comparisons: a simple web search consumes one unit of energy, while the same search performed using an AI model requires ten units. Generating an image with AI escalates this demand even further, consuming up to 1,000 units of energy.
Beyond electricity, the research also points to considerable water usage. ‘There are also impacts of water usage, because these chips that run these models get very hot. And you need water to cool them down,’ Dr. Bashir explained, indicating a strain on water systems necessary for cooling the high-performance hardware.
The consortium’s work underscores an ‘environmental tax’ associated with the widespread adoption of generative AI, manifesting as added stress on power grids and increased consumption of natural resources. While individual users might feel a sense of responsibility, Dr. Bashir clarified that the onus for long-term sustainability primarily lies with the companies developing and training these powerful AI models. ‘Putting everything on the end user and making them feel guilty is not the right approach,’ he said, adding, ‘So, I don’t think the end user should feel completely responsible to do these and to manage this problem. However, there is a case to be made for judicious use of resources.’
Also Read:
- The Environmental Footprint of AI: Unpacking ChatGPT’s Energy Consumption
- MIT Study Reveals How ChatGPT Use Alters Brain Activity and Learning
For the average user, Dr. Bashir suggests a balanced approach. While limiting the use of more powerful AI tools and employing generative AI only when absolutely necessary can help alleviate stress on the power grid and environment, he also advises against excessive worry if AI is used responsibly. Ultimately, the study advocates for a concerted effort from the industry to integrate sustainability into the core of AI development and deployment.
