The Environmental Impact of Digital Technologies and Data
Author Javier Canales Luna
Published at https://www.datacamp.com/
In light of the urgent climate crisis, industries must evaluate and mitigate their environmental impact, including the digital sector. While digital technologies are pivotal for achieving global sustainability goals, recent studies highlight their significant contribution to carbon emissions, resource depletion, and e-waste generation. This article aims to shed light on the environmental footprint of digital technologies, covering all stages from manufacturing to disposal, and emphasizing the role of infrastructure like data centers and telecommunication cables. By drawing on recent research, it serves as an introductory overview to raise awareness about the environmental impact of digital activities and data usage.
1. Manufacturing
Digital devices, such as desktops, laptops, tablets, and smartphones, are the gates for the internet. The rapid digitalization of society is consistent with the increase in digital devices. Estimates suggest that there will be 29.3 billion networked devices by 2023, up from 18.4 billion in 2018. This equals 3.6 networked devices per capita by 2023. Moreover, following the development of the so-called Internet of Things (IoT), IDC predicts that there will be 55.7 billion connected IoT devices by 2025.
A growing number of studies raise concerns about the environmental footprint associated with the massive manufacturing of digital devices. In particular, two areas deserve special consideration: mining activities, and the production process.
Mining
Digital devices– as well as green technologies, such as solar panels, wind turbines, and electric vehicles– are highly dependent on a considerable number of critical materials to function, including minerals, metals, and, most particularly, rare earth elements. For example, most smartphones can carry roughly 80% of the stable elements on the periodic table, including most of the 17 rare-earth elements.
It takes an enormous amount of energy to pull critical material out of the earth.
Production
The manufacturing of digital devices is an energy-demanding process, significantly higher than the use phase. Among the different activities, the production of semiconductors, sometimes referred to as microchips, requires vast amounts of resources.
Moreover, manufacturing activities create wastewater that is released into surrounding waterways. In total, the water required to make a single smartphone, including mining activities, is estimated at 3,190 gallons (12,760 liters).
2. Distribution
The distribution of digital devices rely on a highly globalized and fragmented supply chain. Whether ships, airlines, railroads, or trucks, the transportation of digital devices around the world comes with an environmental footprint, namely in the form of carbon emissions from the burning of fossil fuels.
3. Usage
Besides the energy required to keep all of our digital devices charged and running, three other aspects are important to consider when assessing the environmental footprint of our online activity: data centers, large scale machine learning, and network infrastructure.
Data centers
The digitalization of society will lead to not only an increasing number of digital devices but also an increase in worldwide data traffic. The total volume of data is expected to reach 175 Zettabytes by 2025, with cloud computing applications driving the majority of this growth.
Source: IDC Datasphere whitepaper
Despite the immaterial taste that the word ‘cloud’ has, the truth is that the cloud is heavily rooted in physical hardware.
Several studies suggest that, despite efficiency gains in data center design and operations, the energy bill of data centers is likely to increase as a result of the rapid demand for cloud services from data-intensive technologies, such as video streaming, cloud gaming, social networks, autonomous cars, cryptocurrencies, IoT devices, virtual realities, and artificial intelligence systems.
Besides the carbon footprint of data centers, there are also growing concerns about the water required by cooling systems to keep heat under control inside data centers.
Large-Scale Machine Learning
The release of ChatGPT in late 2022 by OpenAI is having far-reaching implications for nearly every industry. ChatGPT is a highly advanced AI system that can produce highly accurate responses in a conversational way.
Researchers have found that the computational and environmental costs of training LLMs grow proportionally to model size and considerably increase when additional tuning steps are adopted to improve the model’s accuracy.
For example, in 2019 it was estimated that training an LLM with 213 million parameters can emit more than 626,000 pounds of carbon dioxide equivalent, which is nearly five times the lifetime emissions of the average American car, including those associated with manufacturing.
The carbon footprint of ChatGPT –based on the GPT-3 (Generative Pre-trained Transformer) model, consisting of 175 billion parameters– is hence likely to be much higher.
Chart: MIT Technology Review. Source: Strubell et al.
This is only the cost of training the model. But once these models are trained, they still require energy to generate responses based on the information provided by users. The total amount of energy and associated carbon emissions will vary depending on the number of requests and the strategy employed to handle requests. In the case of ChatGPT, it can be millions per day, as it is repeatedly crashing due to exceptionally high demands. For a preliminary analysis of the carbon footprint of ChatGPT, check out this article.
Ultimately, the total emissions from these AI systems will depend not only on the amount of electricity they require but also on the way that electricity is produced. For example, using solar or wind energy to run the systems will result in fewer emissions. However, it’s important to note that the manufacturing and deployment of green technologies are still costly to the environment.
In sum, it’s vital to include environmental considerations in the development of large-scale machine learning models and, more broadly, any kind of AI system.
4. Conclusion
In conclusion, while digital technologies have revolutionized our lives, they also pose significant environmental challenges. From mining for critical materials to the energy-intensive operations of data centers and machine learning models, every aspect of the digital lifecycle contributes to environmental degradation. It’s crucial for industries to prioritize mitigating their environmental impact, especially in the digital realm. This article provides an overview of these environmental implications, aiming to raise awareness and encourage discussions on sustainable practices. Moving forward, stakeholders must integrate environmental sustainability into digital innovation, whether through green data practices, promoting repairability, or investing in renewable energy. By doing so, we can ensure that technological progress aligns with ecological stewardship and work towards a more sustainable future where the benefits of digitalization are not outweighed by its environmental costs.