Topics In Demand
Notification
New

No notification found.

1136

1

Generative AI has made tremendous progress in recent years, revolutionizing industries, and transforming our lives in ways that were once unimaginable. However, this progress has come at a cost, with the increased use of deep learning algorithms leading to a significant contribution to the global carbon footprint. In this blog post, we will examine the ways in which generative AI is contributing to the global carbon footprint and the urgent need to address this issue.

 

The Contribution of Generative AI to the Global Carbon Footprint

 

According to a 2020 report by Massachusetts Institute of Technology, training a large deep-learning model can emit over 626,000 pounds of carbon dioxide equivalent emissions. This is equivalent to the lifetime emissions of five cars. In a 2022 report by Stanford University, CO2 equivalent emissions by some selected machine learning models were up to 10 times the lifetime emissions of an average car.

Bar chart

Description automatically generated with medium confidence

Source: Artificial Intelligence Index Report 2023, Stanford University

In this report, a metric called Power Usage Effectiveness (PUE) is used to assess the energy efficiency of data centers. PUE is the ratio of the total amount of energy consumed by a computer data center facility (this includes the energy consumption by support systems like air conditioning) to the energy delivered to computing equipment. This is how four LLMs considered in the report scored in terms of data center PUE and other factors.

Table

Description automatically generated

Source: Artificial Intelligence Index Report 2023, Stanford University

Data centers, which are critical to running deep learning algorithms, account for approximately 1% of global electricity consumption, according to a report by the International Energy Agency. In 2020, the energy consumption of data centers in the United States alone was estimated to be 200 terawatt-hours (TWh), or roughly 2% of the country's total electricity consumption. Despite efforts to improve energy efficiency and adopt renewable energy sources, the energy consumption of data centers is projected to continue to rise. By 2030, the energy consumption of data centers worldwide could reach 1,200 TWh, which is equivalent to the annual electricity consumption of the entire United Kingdom.

Despite continuous improvements in chips specialized for neural-network processing, such as GPUs (graphics processing units) and TPUs (tensor processing units), the demand for aggressive computing is still very high. Other risks associated with Generative AI include:

  • Waste Generation: As technology evolves rapidly, the hardware used to train deep learning algorithms quickly becomes obsolete, leading to the disposal of large amounts of electronic waste. According to a report by the United Nations, the world generated 53.6 million metric tons of e-waste in 2019, a figure that is projected to rise to 74.7 million metric tons by 2030.
  • Resource Depletion: The production of electronic devices used in training deep learning algorithms requires the extraction and consumption of natural resources such as minerals, metals, and rare earth elements. As the demand for these resources continues to rise, their depletion can have significant environmental impacts, including deforestation, pollution, and habitat destruction.
  • Water Consumption: Data centers, which are critical to running deep learning algorithms, require large amounts of water for cooling purposes. According to a report by the Natural Resources Defense Council, data centers in the United States consume an estimated 626 billion liters of water annually, equivalent to the annual water consumption of 7.2 million Americans. The developments in the generative AI space are expected to drive with them the consumption of water and pose a risk to the environment.

 

The Need to Address the Issue

 

The contribution of generative AI to the global carbon footprint is a significant issue that needs to be addressed urgently. One possible solution is to improve the energy efficiency of data centers. This can be achieved by implementing energy-saving technologies, such as liquid cooling and renewable energy sources, and by using more efficient hardware. Another solution is to develop "green AI" algorithms that are designed to be more energy efficient. This could involve using smaller models, reducing the number of training iterations, and optimizing the architecture of the neural network. According to a study published on the arXiv, reported Green AI energy savings can go up to 115%, with more than 50% energy savings being rather common.

Adopting a circular economy approach to hardware production and disposal, including the reuse and recycling of electronic components can also be one of the ways to minimize environmental risks attributed to generative AI. Collaboration between the technology industry, governments, and research institutions will remain critical to address the issue. Finally, raising awareness of the issue among AI practitioners and consumers can help drive change. By taking a sustainable approach to AI development and deployment, we can ensure that the benefits of generative AI are realized without causing undue harm to the environment.


That the contents of third-party articles/blogs published here on the website, and the interpretation of all information in the article/blogs such as data, maps, numbers, opinions etc. displayed in the article/blogs and views or the opinions expressed within the content are solely of the author's; and do not reflect the opinions and beliefs of NASSCOM or its affiliates in any manner. NASSCOM does not take any liability w.r.t. content in any manner and will not be liable in any manner whatsoever for any kind of liability arising out of any act, error or omission. The contents of third-party article/blogs published, are provided solely as convenience; and the presence of these articles/blogs should not, under any circumstances, be considered as an endorsement of the contents by NASSCOM in any manner; and if you chose to access these articles/blogs , you do so at your own risk.


images
Dhiraj Sharma
Principal Analyst

© Copyright nasscom. All Rights Reserved.