The potential for AI to solve climate change is vast. But that’s not the whole story. As we look deeper, we see that some risks may not be worth taking.
AI’s potential for good is undeniable, but claims that it will “solve the climate crisis” are misguided, say specialists. As a tool, it’s incredible, but utilizing it is not exactly “free.”
Sam Altman, OpenAI’s CEO, admitted,
“AI will require vastly more power than expected.” [10]
A new report indicates that the electricity demands of AI result in an 80% increase in planet-heating emissions. How? By doubling the number of data centres, even with measures to improve their energy efficiency.
Increasing Global AI Energy Needs
In the US, electricity from coal-fired power plants is being prolonged to meet the energy needs of AI. [7] Separate research found that AI servers could consume as much energy as the Netherlands does in just three years. [8]
“One ChatGPT answer might consume seven to ten times more energy than a Google request” [13] according to Benjamin C. Lee, Professor of Electrical and Systems Engineering and Computer and Information Science at the University of Pennsylvania.
In 2023, Microsoft and Google each consumed 24 TWh of electricity, totalling 48 TWh. According to an analysis by Michael Thomas, the author of Distilled, this surpasses the power consumption of over 100 nations, including Iceland and Tunisia. [17]
Yes, Microsoft and Google consume more electricity than entire countries!
“There is no basis to believe AI’s presence will reduce energy use. All the evidence indicates that energy use will massively increase due to all the new data centres. We know there will be small gains in efficiency in data centres, but the simple math is that carbon emissions will go up,” says Michael Khoo, climate disinformation program director at Friends of the Earth. [9]
AI Risks Go Beyond Electricity Use to Contaminating Freshwater and Carbon Dioxide Emissions
Just the process of training GPT-3, which powers ChatGPT, might have used up to 700,000 litres of freshwater. A non-peer-reviewed study led by Shaolei Ren, Associate Professor at the University of California, Riverside, and others examined the freshwater needs behind the creation of ChatGPT.
The problem? The water to cool data centres evaporates, so no usable freshwater is left afterwards.
Plus, the training process for a single AI model can release over 626,000 pounds of carbon dioxide, revealed a study by researchers at the University of Massachusetts, Amherst.
The problem? This amount of greenhouse gas emissions is comparable to 62.6 gasoline-powered cars driven for a year. [11]
Does AI Cause More Harm Than Good to the Planet?
The potential for AI to do good is as big as its potential to harm.
Read more about the potential for AI to mitigate climate change in our article “Can AI Solve Climate Change? Good News About the Potential.”[MW1]
To reduce the emissions, companies must invest in renewable energy and use load-shifting to strategically move electricity consumption from peak to off-peak hours). [14]
Another problem that needs solving is the decades-old Von Neumann architecture, which separates memory and the CPU. A “Von Neumann bottleneck” restricts data exchange between processors.
We need a new kind of processor for AI that works inside the memory to make things much faster and more efficient. This could be 100 times quicker than what we have now. [15] [16]
But the responsibility can’t fall on technology leaders alone. Collaboration among governments, researchers, industry leaders, and environmental organizations is a must if we are to steer AI in the right direction.
The only way for AI to become a net positive for the environment is to stop being such an energy hog.
References
https://environment.upenn.edu/events-insights/news/ai-and-environmental-challenges
https://news.bloomberglaw.com/esg/ai-needs-so-much-power-that-old-coal-plants-are-sticking-around
https://www.theguardian.com/technology/2024/mar/07/ai-climate-change-energy-disinformation-report
https://www.cbsnews.com/news/artificial-intelligence-carbon-footprint-climate-change/
https://www.fairplanet.org/story/ai-environmental-impact-water-usage-data-centers/
https://www.route-fifty.com/emerging-tech/2024/06/how-ai-can-help-and-hurt-environment/397651/
https://thefutureeconomy.ca/op-eds/ai-environmental-harm-arun-iyengar-untether/
Comments