Today I read two thought-provoking articles about the environmental impact of Artificial Intelligence (AI). These articles can be found at the following links:
https://tinyurl.com/samuelforrestblog-article1
https://tinyurl.com/samuelforrestblog-article2
Any other sources used are referenced as appropriate.
Last Edited Date: 10th November 2024
Review Scheduled Date: 10th November 2025
Author: Samuel Forrest
My Thoughts on why AI is problematic for the environment:
AI has certainly been revolutionary for many industries around the world, and is now being used daily by individuals around the world, improving productivity and boosting profits. However, there is a great environmental toll. After reading these two articles, I have identified that there are 4 clear ways in which AI is problematic for the environment, including:
-1) Microchip Production / Sourcing
-3) Water Usage
-4) Power Usage
Of these 4 main areas, I personally think that the most concerning issue is the huge energy consumption of AI technology. If the current trends in AI usage and data centre inefficiencies continue to increase, we could expect AI data centres to represent between 4-8% of global energy usage by 2030 (Goldman Sachs). However, other projections suggest that this figure could actually account for up to 20% of global energy usage. (Article 1).
This is a staggering figure, to put it in perspective, as of 2024, the agriculture and transportation sectors currently account for around 5% of total global electricity consumption. As we can see from the pie chart below, the majority of this electricity used for AI driven data centres is predominantly from non-renewable sources (such as oil, gas and coal). This therefore means that AI has a great carbon footprint.
A single OpenAI's ChatGPT 3.5 query consumes 10x the energy of a Google Search (International Energy Agency & Article 2). The average ChatGPT query consumes around 0.01kWh of electricity. After some simple calculations, I found out that you only have to ask ChatGPT 15 queries to boil the average Kettle, which uses 0.15kwH of energy to boil 5 litres of water. This could be even lower for more modern models, such as ChatGPT 4o/4plus. This really shows just how unsustainable the usage of AI is, and shows that data centers and large AI companies are using just obscene amounts of energy. It also shows that the AI we use all the time daily, such as ChatGPT, other generative software, CoPilot in most Bing searches, Chatbots on many website we visit, are using a huge amount of energy. They also contribute to your personal carbon footprint.
Driven mostly by AI, the number of data centers has surged to 8 million from 500,000 in 2012 (Article 2). These centers house the microchips powering AI systems, many of which require rare earth materials such as Silicon and Nickel, which are mostly extracted in unsustainable ways that eventually lead to deforestation, soil erosion and habitat destruction.
Another environmental challenge is the water consumption of data centers, which is used to cool the AI microchips and other electronic components. This may be considered as a larger problem that chip production and sourcing, considering the fact that 20% of the world's population lacks access to clean water. The fact that AI technologies worsen this issue is alarming.
Possible Solutions to AI's environmental problem:
From my initial research, it looks like there are still relatively few large-scale solutions addressing AI's environmental impact. However, there are some possible strategies:
Optimising AI model efficiency: We can reduce the energy required to run AI models by making them more efficient. We can use 2 main strategies: Model pruning and Knowledge distillation. Model pruning focuses on cutting down on the unnecessary components of a model, by removing redundant or unimportant parameters during the training process. Knowledge distillation is another method where smaller AI models learn from larger AI models, creating a smaller AI model which is more efficient and faster, but only has a small performance toll (Sources: Knowledge Distillation (IEEE). A great example of Knowledge distillation would be ChatGPT 4.0 and ChatGPT-4 mini, which is a smaller version of the original GPT-4 model. It uses knowledge distillation to retain the capabilities of the bigger GPT-4 model, but at a lower cost and higher efficiency. This means it is more suitable to less complex tasks, but can boost amazing efficiency.
Personal Reflection:
Personally, I believe that this issue is serious and overlooked. Having had a conversation with a friend about the environmental cost of sending just a few prompts to ChatGPT, I was inspired to write this article. I now think twice before sending ChatGPT a useless prompt, and will instead use Google search.
In 2023, Microsoft reported that its greenhouse emissions were 30% higher due to AI activities, which made it more challenging for the company to meet its strict environmental and sustainability goals. I believe that as AI continues to grow, striking a balance between AI advancements and sustainability will become more difficult. On the other hand, I personally believe that Apple's approach to Apple Intelligence, and allowing the iPhone processers to use on-device processing will reduce the environmental impact of sending data to Apple or OpenAI's data centers. Of course, this may be more of a financial decision, but I believe it will further help Apple to reach it's strict sustainability goals.
Conclusion:
As a whole, I believe that AI's environmental impact is a pressing issue that requires attention. I hope that we soon will see improvements in AI's training process, energy consumption, cooling process and meetings will be held about AI's environmental impact. Furthermore, I believe that all companies utilising any type of AI should integrate the environmental and carbon footprint impact of AI into their business models. For companies, finding the balance between sustainability and the demand for AI will be a significant challenge.
Did you like my article?
Email me here: samueljforrest@outlook.com
Connect with me here: linkedin.com/in/samueljforrest
Website: samuelforrest.me