Artificial intelligence is one of the most debated topics in technology in the second half of January 2023. Much has been written about how artificial intelligence can improve our lives and streamline our everyday lives, for better or worse. As with all digital technologies, there are also environmental consequences of using artificial intelligence. This applies to both the training of the models, especially the massive use of the significant language models such as ChatGPT we are seeing now.
The CO2 footprint from training and the use of artificial intelligence is a steadily growing problem, as it requires a lot of energy to operate and cool the data centers. The spread of artificial intelligence is exploding, where all the major players, such as Google, Microsoft, OpenAI, Meta, and others, will have their share of the cake. But it's not necessarily all negative: AI can also help reduce our carbon footprint by optimizing industrial processes and helping us make more sustainable decisions, for example, in new transport solutions. It can be challenging to accept that we need to influence the environment in this way to enjoy the benefits of AI.
In this article, we will examine the environmental factors of artificial intelligence, including training and use, and how it affects our CO2 footprint. We will also study how artificial intelligence can help reduce our carbon footprint and contribute to a more sustainable future.
The article is intended as a basis for debate in education and contains links to sources dealing with CO2 footprint and artificial intelligence. Let's start with a quote from the EU magazine "Horizon - the EU Research & Innovation Magazine":
Artificial intelligence (AI) technology can help us fight climate change – but it also comes at a cost to the planet. To truly benefit from the technology's climate solutions, we also need a better understanding of AI's growing carbon footprint, say researchers.
The first consideration is the environmental impact of operating artificial intelligence such as ChatGPT. Several people have tried to estimate what this means for the climate:
1 million users with 10 questions each = 29,167 h. of A100 GPU time per day 29,167 hours * 407W = 11,870kWh per day 0.000322167 * 11.870 = 3.82 tCO₂e per day (California emmision used) That’s about 3 months of an average American’s footprint of approximately 15 tCO₂e per year. Or put another way the same CO₂ emission rate as 93 Americans.
Another suggestion for a calculation of energy consumption for training is this:
The latest language models include billions and even trillions of weights. One popular model, GPT-3, has 175 billion machine learning parameters. It was trained on NVIDIA V100, but researchers have calculated that using A100s would have taken 1,024 GPUs, 34 days and $4.6million to train the model. While energy usage has not been disclosed, it’s estimated that GPT-3 consumed 936 MWh.
All these calculations are estimates, as OpenAI has not provided concrete information showing what hardware has been used. (There are also no estimates of time and power consumption.) However, the forecast mentioned are reasonable and perhaps even a little conservative. However, they tell us that using artificial intelligence regarding climate is not free. Notice that the calculations are made with one million daily users. These were the status after just one week with ChatGPT!
On February 2, several media outlets wrote that it is estimated that ChatGPT will reach 100 million active users by January 2023! It took TikTok 9 months and Spotify 4 1/2 years to get 100 million users. ChatGPT is supposed to be the fastest-growing IT product ever and on par with Pokemon Go.
Kasper Groes Albin Ludvigsen has written a post on LinkedIn in which there is a calculation of ChatGPT's power consumption: ChatGPT may have consumed as much electricity as 17,526 Danes in January.TikTok
The second consideration is that many researchers believe we must use artificial intelligence to combat or reduce CO2 emissions.
Artificial intelligence models can help analyze large amounts of data and identify trends and patterns that can improve the effectiveness of climate action in the long term. For example, artificial intelligence can monitor and model climate change, predict the weather, and analyze data on CO2 emissions to help reduce emission levels. Artificial intelligence can also be used to optimize the use of clean energy and help design more sustainable homes and infrastructure.
Below, we have made a small collection of links that are obvious to use in subjects such as technology, history of ideas, social studies, or similar for debate on artificial intelligence concerning the environment and climate.
With this article, we hope that the debates around educational institutions will get going.
Link collection:
CO2 footprint of training and operating artificial intelligence (e.g. ChatGPT)
Using ChatGPT in Education At Viden.AI, we have not decided regarding the General Data Protection Regulation (GDPR) and the use of ChatGPT. Therefore, exercise caution when incorporating the program into education or storing sensitive personal information.
Kasper Groes Albin Ludvigsen
Payal Dhar
Kasper Groes Albin Ludvigsen
Melissa Heikkilä
Solve Sæbø
MAIEI
Prateek Majumder
Tim Cross
Karen Hao
Public Interest Entrepreneur
Will Buchanan
Gert Madsen
Owen Mulhern
Louise Vogdrup-Schmidt
Mark Minevich
admin
