Also Read :

---

• Carbon Capture and Storage (CCS) on Distinc Options in Indonesia
• Is Methane Potential Cause Greenhouse Effect after Carbon?

---

Summary :

• Approximately 2,200 gallons of ultrapure water (UPW) are required to manufacture a microprocessor. Setting that aside, training a massive language model like GPT-3 can require millions of gallons of fresh water, and executing a GPT-3 inference for 10-50 queries consumes 500 milliliters of water, depending on the time and location of the model’s hosting. GPT-4, the model that ChatGPT currently utilizes, is reportedly larger in size, and as a result, it is expected to use more water compared to GPT-3
• From 2018 to 2023, the scale of reference models increased by 3,000 times (from BERT’s 340 million parameters to an estimated 1.5 trillion in GPT-4). This expansion results in bigger data centres, which demand substantial electricity to energize servers and water for their cooling.
• The main challenge appears to be finding a compromise between the advantages of AI technology for users and reducing its adverse effects on the environment.

Water Usage for AI

In discussions regarding emissions, it’s unavoidable to consider ChatGPT’s water footprint. A study conducted by the University of California, Riverside, reveals that a dialogue consisting of approximately 20 questions consumes half a liter of water, which is comparable to a small plastic bottle. Training ChatGPT can consume as much water as manufacturing 370 BMWs or 320 Teslas.

Shaolei Ren, an associate professor in electrical and computer engineering at the University of California, Riverside, noted in his 2023 research:

Approximately 2,200 gallons of ultrapure water (UPW) are required to manufacture a microprocessor. Setting that aside, training a massive language model like GPT-3 can require millions of gallons of fresh water, and executing a GPT-3 inference for 10-50 queries consumes 500 milliliters of water, depending on the time and location of the model’s hosting. GPT-4, the model that ChatGPT currently utilizes, is reportedly larger in size, and as a result, it is expected to use more water compared to GPT-3.

What effect will AI have on the climate?

Although cloud computing options have allowed AI models to reach everyone with a smartphone, this ease fosters a false sense of immateriality and insignificance. In truth, the recent advancements in AI performance are founded on progressively larger models trained on continually growing datasets.

From 2018 to 2023, the scale of reference models increased by 3,000 times (from BERT’s 340 million parameters to an estimated 1.5 trillion in GPT-4). This expansion results in bigger data centres, which demand substantial electricity to energize servers and water for their cooling.

To provide context, the last training phase of GPT-3 is believed to have released 500 tonnes of CO2. For GPT-4, according to leaked details regarding its training period, estimates vary from 6,000 to 12,000 tonnes of CO2. In line with the pattern of updates to OpenAI’s models, we can anticipate that usage might rise tenfold for the fifth iteration.

It’s important to mention that model training is merely the beginning, as various intermediate experiments could use just as much or even more energy. The absence of clarity about the exact effects of these processes is troubling.

Furthermore, employing these models after training carries certain implications. One query to ChatGPT (GPT-3) uses multiple times more energy than a Google search, a discrepancy that is expected to increase with newer, larger models.

Ways to utilize artificial intelligence for the benefit of the planet?

Artificial intelligence holds significant promise in aiding environmental protection and fostering a more sustainable future for our Earth. Nonetheless, it is crucial to handle this technology appropriately to guarantee that its effects are beneficial for both the environment and society overall.

AI enables the analysis of extensive data collections, like satellite imagery, assisting scientists in tracking climate change. It can also aid in developing improved climate models, recognizing trends, and forecasting future changes, which enhances understanding and more efficient mitigation of climate change. Additional uses encompass water preservation, fire suppression, as well as the detection and retrieval of recyclable resources.

Experts from the National Oceanic and Atmospheric Administration are utilizing AI to enhance climate models, weather predictions, and various Earth-related applications and systems. The United Nations Environment Programme employs AI to assess atmospheric carbon dioxide levels, shifts in glacier mass, and rising sea levels.

How AI are able to be Greener?

Indeed, there are methods to make artificial intelligence (AI) more environmentally friendly:

• Enhance algorithms: Enhancing algorithms can decrease the resources and energy required to train and operate AI models.
• Eco-friendly energy solutions: Transitioning data centers to eco-friendly energy solutions, like solar or wind energy, can lessen the ecological footprint of AI.
• Recycling and reuse: Encouraging the practices of recycling and reusing hardware utilized in AI processing can diminish the harmful environmental effects.
• Enhanced cooling: Utilizing advanced cooling technologies in data centers can decrease the energy required to sustain suitable temperatures. Microsoft is evaluating a submerged data center system.
• Utilizing AI in conservation: Employing AI to enhance environmental actions, like tracking climate change, overseeing natural resources, or safeguarding biodiversity, can contribute to developing more sustainable methods.

What EBT Heroes Know about the Relationship between Carbon and AI ?

AI requires substantial natural resources, and its carbon footprint can be considerable, particularly during the training of large language models. Nevertheless, it is equally essential to acknowledge the beneficial effect that AI can exert on arranging and safeguarding the environment. Artificial intelligence aids in tracking climate change, examining ecosystems, enhancing energy use, and various other actions that contribute to safeguarding the Earth.

The main challenge appears to be finding a compromise between the advantages of AI technology for users and reducing its adverse effects on the environment. By creating more effective algorithms, utilizing energy resources more responsibly, and encouraging sustainable practices in the tech industry, we can advance in both artificial intelligence and environmental conservation.

Also Read :

---

• China Plans to Cut Out Carbon Dioxide (CO2) on 2035
• Indonesian Carbon Capture & Sequestration Regulation Overview

---

Editor: Savira Oktavia

References:

[1] AI’s carbon footprint – how does the popularity of artificial intelligence affect the climate?

[2] Will AI be helpful or harmful in the fight against climate change?