AI continues to consume more and more energy. Where will it end?

In the midst of the hype surrounding artificial intelligence (AI) and its impressive capabilities, one important aspect has been overlooked – its impact on the environment. Analysts are now warning that AI’s carbon footprint could be as damaging, if not more so, than that of bitcoin mining, which currently produces more greenhouse gases than entire countries. With record-breaking heat waves and the urgent need to protect our planet’s fragile ecosystems, it is critical to pay attention to the environmental impacts of AI.

Current state of affairs

Currently, the entire IT industry is responsible for approximately 2% of global CO2 emissions. However, consulting firm Gartner predicts that if the AI industry continues on its current trajectory, it will consume 3.5% of the world’s electricity by 2030. These alarming statistics underscore the urgent need to consider AI’s ecological footprint in our pursuit of technological advancement.

Sasha Luccioni, ethics researcher at open source machine learning platform Hugging Face, emphasizes the importance of considering the environmental impact of AI. Luccioni states, “Essentially, if you really want to save the planet with AI, you have to consider the ecological footprint as well. It doesn’t make sense to burn down a forest and then use AI to track the deforestation.”

The rise of ChatGPT and its implications

One of the most notable artificial intelligence systems is ChatGPT, developed by OpenAI with support from Microsoft. With more than 100 million users worldwide, OpenAI spends approximately $700,000 daily on computing costs alone to power its chatbot. This popularity has sparked an arms race among technology giants such as Google and Amazon, who are quickly devoting resources to developing their own natural language processing systems.

Similar to cryptocurrency mining, AI relies heavily on powerful GPUs to process massive amounts of data. For example, ChatGPT operates through massive data centers equipped with tens of thousands of power-hungry computer chips. Calculating the environmental impact of ChatGPT and other AI systems is difficult due to limited access to the necessary information.

Complex calculation of environmental impact

Computer scientist Roy Schwartz of the Hebrew University of Jerusalem emphasizes the lack of transparency in the disclosure of carbon dioxide emissions by artificial intelligence models. “Obviously, these companies don’t like to disclose which model they are using and how much carbon it emits,” Schwartz says. This lack of transparency makes it difficult for researchers to accurately assess the overall environmental impact of AI systems.

Nevertheless, estimates have been made of the energy consumption and carbon emissions associated with training AI models. For example, training GPT-3, the predecessor to ChatGPT, on a database of more than 500 billion words would require 1,287 megawatt-hours of electricity and 10,000 computer chips. This amount of energy would be enough to power about 121 homes in the U.S. for a year and would emit about 550 tons of carbon dioxide, the equivalent of 33 flights from Australia to the United Kingdom.

The future of artificial intelligence and energy consumption

As artificial intelligence develops, it is difficult to predict its future scale and energy efficiency. However, there is reason to believe that AI energy consumption will increase over time, similar to the trajectory of cryptocurrencies. The GPT-4 model, released in July, was trained on 570 parameters, 570 times those of GPT-3, meaning it could consume even more energy than its predecessors.

Another language model called BLOOM consumed 433 megawatt-hours of electricity while being trained on 1.6 Tbytes of data. These examples underscore the growing concern about the energy intensity of AI developments and their possible environmental consequences.

The need for environmentally friendly solutions

Addressing the environmental impact of AI is critical to the sustainable development of this technology. It is imperative that or companies must prioritize energy efficiency and transparency in disclosing carbon emissions associated with AI models. In addition, investing in renewable energy to power data centers and exploring innovative cooling technologies can help reduce the environmental footprint of AI.

According to computer scientist Roy Schwartz, “We need to think about how to make these models more energy efficient, but also more carbon efficient.” A balance between technological advancement and environmental responsibility is necessary to ensure a sustainable future for AI and our planet.

Notify of

Inline Feedbacks
View all comments
Would love your thoughts, please comment.x