By – Shreya Parameshwaran

Abstract

While artificial intelligence (AI) is changing economies around the world, its environmental effects are largely unseen. AI requires large data centres with a lot of power and resources that use up tremendous amounts of energy in the form of continuous training and computation, creating significant impacts through CO2 emissions. Just training a single large-scale AI model can produce the equivalent CO2 emissions of several cars, and as of today; data centres consume approximately 1 – 1.5% of the world’s total electricity. This article looks at three specific aspects of the energy utilisation of AI, their infrastructure requirements, as well as how the AI economy has impacted the technology industry, including the role of Big Tech. The implementation of sustainable practices in developing AI may avoid worsening of environmental impacts instead of solving them.

Introduction: Intelligence with an Energy Cost 

We often hear that Artificial Intelligence is a piece of the puzzle, this means that it is only a small part of a much larger solution that helps people to achieve different advancements in areas such as climate change prediction, disaster response, and resource optimisation. Despite these claims, what remains unrecognised is that AI systems require massive amounts of energy to operate. Advanced model training and deployment require powerful computing platforms to handle the computational requirements of AI, resulting in significant use of electricity and carbon emission due to their fossil fuel-based produced energy sources.

According to the International Energy Agency (IEA), global data centres consumed about approximately 460 terawatt-hours (TWh) of electricity in the recent years, with demand expected to increase significantly due to the rise of AI technologies. The consumption of data is not an accident – it is central to the way that AI functions. A study from the University of Massachusetts Amherst also found that one big deep-learning model can release about 284,000 kilograms (626,000 pounds) of carbon dioxide, nearly five times a car emits over its whole lifetime — and this is solely from the one-time process of training the model, before a single user interaction even takes place. Hence, these results show that digital technology is neither clean, nor harmless to the environment.

Core Mechanisms and Energy Infrastructure

There are multiple mechanisms by which AI has an ecological footprint, all of which create an ecosystem for its energy consumption. Model training represents the basic mechanism of energy consumption within AI. To build AI algorithms, you must keep running high volumes of data through fast computers, often Graphics Processing Units (GPUs) or Application-Specific Integrated Circuits (ASICs), repeatedly for long stretches of time. This approach uses a lot of electricity, and it will keep doing so for years since it gathers and processes extensive amounts of data.

In addition to the energy required for training AI systems, the implementation stage — commonly referred to as inference, creates a significant additional demand for energy consumption. Due to the enormous number of A.I. interactions by end-users such as chatbots, recommendation engines, and searches engine queries; the constant computational resources required to support millions or even billions of users worldwide create sustained and continuously increasing energy demands on a global scale. 

The data centre’s influence on the other hand cannot be overemphasized in relation to AI. Data centres serve the dual purpose of powering the computing that is necessary for various operations and, therefore, require sophisticated refrigeration systems to manage the heat generated during that process. The refrigeration component accounts for a significant amount of total energy consumption; thus, increasing the energy consumption associated with AI expansion is unavoidable. Companies like Google recognize that improving efficiency in the operation of data centres is necessary as AI workloads are anticipated to grow exponentially. However, offsetting the benefits from operational efficiencies with the rapid growth associated with deploying AI solutions continues to be an ongoing challenge.

Energy Inequality and Global Impact

AI is harmful to the environment but is experienced very differently in various areas. The majority of AI infrastructure (e.g., data centres, large-scale computational facilities) is found in more technologically advanced countries; however, the huge energy that is consumed from these data centres also increases global greenhouse gas emissions, contributing to climate change on a global basis. Developing nations—despite hosting fewer AI facilities themselves, tend to be impacted the most from the effects of climate change, since they are already much more susceptible to these impacts.

Furthermore, the growth of AI infrastructure may emphasize current disparities in access to energy. Regions with unreliable and unavailable sources of electricity have already expressed fears of unequal distribution of resources as large-scale digital infrastructures are prioritized for energy usage. Reports indicate that if Digital Technology’s increasing demand for energy, including AI, continues beyond sustainable limits, it could place considerable strain on the global energy grid. This presents a paradox for AI, while it is frequently cited as a driver of development and progress, its consideration could ultimately support and worsen environmental inequities instead of improving them.

Corporate Expansion and Environmental Accountability

Several big tech firms, including Microsoft, Amazon, and Meta, have spurred forward the rapid advancement of artificial intelligence (AI). Each of these companies have made public commitments to achieving sustainability through carbon neutrality and minimizing their overall impact on the environment. At the same time, the continued expansion of the infrastructure necessary for developing AI poses a clear challenge to these companies’ sustainability-oriented goals.

For instance, Microsoft has released sustainability reports showing an increased level of emissions over the past couple of years due in large part to their rapidly growing cloud computing business and their increased demand for artificial intelligence services. Therefore, a conflict emerges between the need for technological innovation and environmental stewardship; although businesses are making substantial investments in renewable sources of energy, optimizing energy use from existing resources, and using other means of reducing their carbon footprint by increasing their use of renewable resources, the large number of AI projects continuing to be developed are increasing overall energy usage.

Policy Failures and Technological Trade-offs

There are two major systemic issues contributing to the environmental consequences of AI systems. The first is the absence of an established regulatory framework for the energy consumption associated with these AI technologies. In contrast to conventional industries, which are subject to emissions and resource regulations, AI is still fundamentally unregulatable because the technology is evolving faster than governments and legal systems can create effective regulatory standards; hence, companies can expand their energy consumption without facing any enforcement actions.

Moreover, there exists an inverse relationship between performance and efficiency. Advanced AI models typically require an exponentially larger amount of compute resources and therefore, consume more energy. As a result, there is a structural incentive to produce increasingly capable systems with regards to both performance and efficiency, despite incurring significant environmental costs. Unless corrective measures are taken, this trajectory poses the risk of rationalizing unsustainable practices in AI development.

Policy Solutions

Mitigating the ecological footprint of artificial intelligence will take both technological innovation (green tech) and legislatorial measures. One approach to accomplish this goal is with “Green AI.” Green AI refers to using technologies and policies which promote sustainability, with respect to both energy and materiel, while providing performance (efficiency). Research on this subject is focused on improving algorithms, so that performance can be maximized with as little processing power consumed.

Additionally, relocating data centres to renewable power is also necessary to reduce the carbon pollution arising from AI. Having greater transparency in energy use and emissions reporting will hold AI developers more accountable and help people make more informed choices. If we have global standards and governance of AI’s environmental outcomes, we will be able to harness the goodness of AI today, while protecting future generations.

Conclusion: Progress with Consequences

Artificial Intelligence (AI) is changing everything, and its potential is increasing day by day. However, it has become increasingly difficult to ignore the negative harm it continuously causes to the environment. Rate of energy usage is alarming and continuing to rise in these systems. Consequently, it’s not so much a technology issue but the responsibility of everyone who uses AI. Individuals must be aware of how they communicate with these systems, especially if we ever want to make AI genuinely sustainable. Though AI can help tackle problems, they will become part of the problem in terms of resource usage for running AI systems.

AI’s development in the future should therefore be undertaken with a balance between innovation versus care for the environment. Unless conscious efforts are made to reduce the energy footprint, AI could become a part of the problem it is trying to solve. It is not just a technical challenge; it’s a moral obligation. It must be taken seriously by all levels of government (local / national / international), companies all over the world and the global community.

About the Author

Shreya Parameshwaran is a law student with a strong interest in environmental justice, human dignity, and land sovereignty. Through legal research and writing, she seeks to explore how law can support communities affected by ecological conflict and challenge structures that threaten both people and the environment.

Image Source : https://medium.com/@emperorbrains/the-impact-of-ai-on-environmental-sustainability-d7c621b18379