The potential for Artificial Intelligence to address our most pressing global challenges — particularly climate change — is massive. Yet, as AI training exponentially increases the need for compute power and energy resources, its environmental footprint raises a crucial question: Could the cure be contributing to the disease?
This paradox dominated discussions at New York Climate Week and will likely take center stage at COP 29 next month. The computational demands of training and running large AI models have transformed data centers into the world's fastest-growing electricity consumers.
The reality forces us to confront an uncomfortable truth: the very technology we're banking on to fight climate change might be speeding it up. The debate has divided many tech leaders and luminaries. Google CEO Eric Schmidt has championed AI as the key to addressing climate change while others warn it could actually accelerate the climate crisis.
However, the AI landscape is more nuanced than headlines suggest. While large-scale generative AI models consume significant energy, other AI solutions — particularly those designed to use AI on the edge — are demonstrating that AI can address climate challenges while maintaining a sustainable footprint.
While some forecasting points to AI mitigating 5% to 10% of GHG emissions by 2030 thanks to its ability to scale proven technology, it is also triggering unprecedented increases in energy consumption. A single ChatGPT query requires nearly ten times the electricity of a standard Google search. Data centers powering these AI models are rapidly consuming a larger share of global electricity — projected to account for 6% of U.S. power demand by 2026, up from 4% in 2022.
Even more worrying, some forecasts predict AI could consume 15% of global electricity by 2030. Therefore, while Large Language Models (LLMs) promise to revolutionize renewable energy and optimize power grids, the massive energy requirements of some of these solutions threaten to undermine their environmental benefits.
In order to address this AI power consumption challenge in the future, new techniques will need to be developed to lower computational usage and become more efficient in training LLMs. In the future there may even be the need to regulate or certify AI models based on their energy consumption in a similar way we recognize a sustainable building as "LEED" (Leadership in Energy and Environmental Design).
However, not all AI is created equal, and not every AI solution is an energy hog. AI being used today on the edge is designed with energy efficiency in mind, and is proving that artificial intelligence can advance climate goals without additional environmental costs. These low-powered applications of AI require less training power but also operate more intelligently by processing data where it's generated. And they’re already demonstrating that AI can be part of the climate solution right now rather than adding to the problem.
The potential to deploy these solutions at scale exists, as nearly every IoT-connected device will soon have the capability to run AI models. Furthermore, Edge AI models utilize adaptable processing rates to align with task requirements. This enables lower-capacity operation over extended periods to reduced energy consumption and extends hardware longevity.
While we await future climate solutions powered by advanced language models like GPT-4 Omni, edge AI is already making a difference. My team has seen firsthand how AI on mounted cameras is helping recycling facilities gain 100% visibility into waste streams to divert 66k tons of recyclable material from landfills, incineration, and oceans over the last year - and avoid 20 tCO₂eq emissions.
We’re part of a growing ecosystem of AI solutions that are using practical, edge AI approaches to address the climate crisis. Pano AI, for instance, uses high-definition cameras that leverage deep learning to detect and prevent wildfires. Meanwhile, Kobold Metals, backed by tech titans Bill Gates and Jeff Bezos, utilizes computer vision and machine learning to identify where to find critical minerals like lithium, which are vital to the world's energy transition. These examples highlight AI’s strength on the edge in data gathering and analysis with machine learning and neural networks assisting in extracting data and information from images to make actional recommendations.
While concerns grow around the energy consumption of large-scale AI computing, edge or on-device AI solutions are already delivering positive climate outcomes. Although large language models dominate headlines in terms of their future climate solution potential, these edge AI applications are creating measurable environmental benefits today without adding to our carbon footprint. Thus proving there is a practical way forward to harness AI's potential while minimizing its environmental impact.
_________________________________________________________________
Gaspard Duthilleul is the COO of Greyparrot, a pioneer in waste intelligence. An engineer by training, Gaspard has steered high-stakes projects across Europe, Asia, and the US over the last decade. He leads Greyparrot’s operational strategy, scaling impact and utilizing waste data to drive a more sustainable approach to resources. _________________________________________________________________
