For the past few months, almost every activity and job I’ve done has involved the help of Artificial Intelligence (AI), especially ‘ChatGPT.’ It made me wonder: if by 2050 the world’s population truly reaches nine billion, as experts predict, and AI becomes a part of everyone’s daily life, how much energy will AI require?

I posed this question to my new companion, ‘ChatGPT,’ and here is the response:

“Energy consumption by AI varies greatly, especially depending on whether it involves model training or inference. Training large models like GPT requires an extraordinary amount of computational power, with energy usage reaching hundreds of thousands of kilowatt-hours (kWh), equivalent to decades of household energy consumption. This training process occurs in data centers that utilize thousands of GPU or TPU units, leading to a significant carbon footprint. Although inference is more energy-efficient, its large-scale application still carries notable impacts. By 2050, with the global population expected to reach 9 billion, if everyone uses AI in their daily lives, energy demands could reach around 328.5 terawatt-hours (TWh) per year — approximately 1–2% of today’s global energy consumption. However, this prediction greatly depends on future technological advancements, such as improved hardware efficiency, optimized algorithms, and the increased use of renewable energy, which could significantly reduce AI’s carbon footprint.”

AI is often hailed as a key technology to help humanity combat climate change. With its ability to process vast amounts of data, predict extreme weather patterns, and assist in the transition to renewable energy, AI certainly offers unprecedented solutions. However, behind these benefits lies a paradox that raises an important question:

“Is AI truly the solution, or is it exacerbating the problem?”

Behind the Scenes: AI’s Growing Energy Consumption

We live in an era where data is the “new oil” — and AI is the giant machine extracting it. Training large AI models like GPT-2 and GPT-4 requires immense computational power. A study from MIT revealed that training large AI models can produce carbon emissions equivalent to the lifetime usage of five cars.

These are not just statistics. Upon closer inspection, this AI trend brings consequences that cannot be ignored. We are talking about tens to hundreds of thousands of kilograms of CO₂ generated just to make an AI model intelligent. Ironically, the technology we hope will save the environment is placing a considerable burden on the planet.

Let’s talk about the carbon footprint. Another well-known study showed that training a single large AI model can emit as much carbon as 125 round-trip flights between New York and Beijing. This raises the question: how can AI, which aims to help us combat climate change, be ethically acceptable if its environmental footprint is so large?

AI Infrastructure: Data Centers and the Energy Consumption Dilemma

The data centers supporting AI are a crucial part of this problem. Currently, the data center sector accounts for around 1% of the world’s total energy consumption, and this figure is expected to rise as AI continues to develop. As we move towards a digital, data-driven economy, the ethical challenge lies in the dependence on energy-hungry infrastructure.

Developed countries, which host the majority of global data centers, tend to benefit more from this technology, while developing nations often get left behind. This disparity raises another ethical question: is this technology truly inclusive, or does it merely deepen existing economic and social divides?

Green Solutions for AI: Keeping the Carbon Footprint Low

However, it’s not all bad news. There is a growing movement toward “green AI,” a concept that emphasizes reducing energy consumption during AI model training. Google, one of the major players in AI, has started adopting more energy-efficient learning systems. Additionally, they are transitioning to data centers powered by renewable energy.

While still limited, this step demonstrates that there is a way forward. By continuing to develop more energy-efficient technologies and smarter resource management, AI can remain an environmentally friendly solution. Moreover, greater efforts are being made to leverage AI in optimizing renewable energy grids. AI helps balance the supply and demand of energy from fluctuating sources like wind and solar power, reducing reliance on fossil fuels.

AI and Climate Change: Paradox or Opportunity?

We are at a crossroads. AI clearly holds enormous potential to help us mitigate the impacts of climate change. However, the use of this technology does not escape the ethical questions surrounding its own environmental impact. If not managed properly, AI could become part of the very problem it aims to solve.

It’s important to recognize that AI is not a silver bullet that can solve everything. Like any technology, AI needs careful attention to its environmental impact. Perhaps the solution lies in developing more efficient AI models and utilizing renewable energy to reduce the carbon footprint of this technology.

Conclusion: AI for Good or Harm?

Ultimately, the ethical challenge of AI in the context of climate change is complex. On one hand, this technology offers concrete solutions to urgent problems, such as optimizing renewable energy, predicting extreme weather, and adapting agriculture. However, the high energy consumption needed to train large AI models remains a significant hurdle.

The future of AI in addressing climate change depends on how quickly we can resolve this paradox. Developing greener technologies, efficiently managing infrastructure, and using renewable energy are key to ensuring AI truly becomes a solution — and not part of the problem.

As a final thought, here’s the challenge we face: how can we harness artificial intelligence to build a greener future without sacrificing the planet we are trying to save? Share your thoughts in the comments, and let’s discuss solutions together!

Referensi:

1. Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and Policy Considerations for Deep Learning in NLP. Massachusetts Institute of Technology.
2. Patterson, D., et al. (2021). Carbon Emissions and Large Neural Network Training. Google Research.
3. Bender, E. M., et al. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
4. Shehabi, A., et al. (2016). United States Data Center Energy Usage Report. Lawrence Berkeley National Laboratory.
5. Google AI Blog. (2020). Our Commitment to Building a Carbon-Free AI Future.
6. Amodei, D., & Hernandez, D. (2018). AI and Compute. OpenAI Blog.
7. Schwartz, R., et al. (2020). Green AI.
8. Siemens. (2020). AI for Renewable Energy Optimization.
9. WEF. (2020). How AI Can Revolutionize Agriculture and Help Solve Climate Crisis.

Contributor:

Eka B. Panuntun