June 5th is the World Environment Day. Established in 1972 by the United Nations General Assembly, the World Environment Day serves as an occasion for us to stop and reflect on pressing environmental problems, to thing about actions that can promote ecological sustainability, and to encourage and promote global action. The theme for the 2024 World Environment Day is land restoration, desertification, and drought resilience, with focus on the restoration of healthy land, prevention of desert expansion, and the management of water shortages [1].
While trees, healthy soil, and clean water are essential for a happy planet, in this article, we will explore a different, and at first glance, unlikely source of environmental impact: artificial intelligence (AI). AI has rapidly integrated into our daily lives and in various ways, is shaping the future of humanity. The impact that AI has on the environment is a complex issue and while it can offer huge benefits, for example, through the improvement of supply chain, urban traffic management, and enhanced weather modelling among others, the cost of training the large AI models are not insignificant. The concern is that even if AI is applied to climate-conscious technology, the costs of building and training the models for these technologies could leave society in a worse environmental situation than before [2].
Energy Consumption in AI Training
Training AI models consumes a significant amount of power and time. For instance, training a model like GPT-3 requires approximately 1,287 MWh of energy. But that's not it. When AI models like ChatGPT are answering our questions, they are also using energy - and at good quantities. OpenAI's ChatGPT can consume around 564 MWh every day to remain available and ready to answer any question at any time [4].
Training a single AI model can emit as much carbon as five cars in their lifetimes
Why is the carbon footprint so high?
If we continue to consider OpenAI's GPT-3 as an example, this large language model consists of around 175 billion parameters. Training such a model requires colossal amount of computational resources to determine the ideal values for these parameters. Besides the real-estate size of the data centres, where AI training takes place, require constant power supply and constant cooling, hence the high energy consumption and carbon footprint. Data centres can in fact account for 3% of global electricity supply and 2% of total greenhouse gas emissions [5].
The artificial intelligence industry could consume as much energy as a country the size of the Netherlands by 2027
Water Usage
Huge power consumption has one huge side effect: heat. Keeping data centres cool, requires substantial water usage. Google is reported to have consumed 4.3 billion gallons of water in 2021 to cool the servers in their data centres [7]. Understandably, this water usage places a strain on water resources, particularly when one considers the drier, hotter climates that we are facing and the increased risk of drought.
Hardware and E-Waste
Unfortunately, the environmental impact does not stop with energy and water consumption. The hardware used for AI training, including GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units), has a limited lifespan. To make matters worse, the rapid advancement in AI technologies means that these components often become obsolete quickly, leading to increased electronic waste (e-waste). Disposing of e-waste improperly can lead to soil and water contamination, further exacerbating environmental issues [8].
The other side of the coin
Although AI models do have a significant environmental price-tag, perhaps ironically, these technologies can also be used to minimise environmental costs in various sectors. This can be obtained by optimising the systems used in these various sectors to reduce their environmental impact. Let's take a look at some examples.
Urban Traffic Management
Picture your morning commute. Chances are that this involves time stuck in traffic. The good news is that AI can have a significant impact on improving urban traffic, reducing traffic congestion and helping in better planning and managing traffic flows. This has been explored within the Department of Systems and Control Engineering where AI technologies have been applied to optimize traffic management by analysing real-time data from cameras or sensors, and adjusting traffic signals, rerouting vehicles, and improving overall traffic efficiency. Our work has also focused on the application of AI for predicting traffic patterns, accidents, and congestion hotspots, allowing authorities to proactively address issues before they escalate. This helps in better planning and managing traffic flows.
Weather Prediction and Climate Modelling
The ability to accurately predict the weather has impacts that go beyond determining whether or not you should wash your car! Accurate weather forecasts lead to better preparedness for extreme weather conditions, and so, improved resource management. The fact that AI algorithms can process vast amounts of meteorological data, means that they can create precise and reliable models of weather forecasts. Moreover, the speed with which these algorithms can process this data means that these models can be refined with newer incoming data far quicker than what could be handled by traditional methods. In addition to the day-to-day weather forecasting, AI can also be used to model the impact of climate change effects such as the effects of global warming, ice melt and other phenomena. Such information can, in turn, provide insights for policymakers.
Renewable Energy Optimization
The concepts of modelling and prediction used for weather prediction can also be applied to model, predict and therefore optimise energy generation. Predicting the energy consumption has an important role ensuring optimal use of energy. Knowing the demand as well as what can be generated through renewable energy sources can help in balancing the generation between renewable energy sources and other non-renewable sources such that the required energy is made available to consumers while minimising waste, creating truly smart grids. Moreover, by tapping into weather monitoring, AI-driven solutions can be implemented which allow for optimisation of the positioning of wind turbines and solar panels to optimise the energy generated through renewable sources.
Balancing the Costs and Benefits
The examples discussed above are but few of the application areas of AI but these clearly demonstrate that despite the negative effects that AI usage has on the environment, there are notable benefits of the use of AI to the environment. By acknowledging the negative impacts of AI helps in identifying mitigating approaches that help shift the balance of AI impact on the environment. Indeed efforts are being made to reduce the negative impact of training. Companies like Google and Microsoft are investing in so called green data centres, that is data centres where the energy consumed is generated through renewable sources, with innovative cooling systems and where the water consumed is reduced, as well as implementing water reuse practices, recycling water for non-potable reuse such as irrigation or toilet-flushing, thereby further reducing the water consumption [9].
It is also possible to make efforts to design algorithms that require less computational power by being more efficient, using techniques such as model pruning and quantisation to reduce the size and complexity of AI models, or by being more creative and translate the ways with which human brains manage to train to their AI counterparts [10].
Policy and Regulation
Relying on the goodwill societal-consciousness of tech companies in curbing the negative impact of AI on the environment is not sufficient. Governments and regulatory bodies need to have an active role in ensuring that AI development and deployment is environmentally sustainable. These would include establishing policies that promote energy efficiency, support renewable energy adoption, and regulate the recycling and disposal of electronic waste to reduce soil and water contamination.
Collaboration between academia, industry, and government is essential for driving innovation in sustainable AI. Collaborative research initiatives can result in breakthroughs in energy-efficient AI technologies. Government agencies that work closely with academia and tech industries can lead to science-based policies and guidelines that help to promote the continued development of sustainable AI. Establishing industry standards for energy efficiency and sustainability in AI will further establish widespread adoption of best practices, creating benchmarks and certification for sustainable AI practices.
Conclusion
The impact of AI on the environment is a multifaceted issue that requires careful consideration and action. While the environmental costs of training large AI models are significant, the potential benefits of AI in areas such as urban traffic management, weather prediction, and renewable energy optimization cannot be overlooked. By adopting sustainable practices, developing more efficient technologies, and implementing supportive policies, it is possible to harness the power of AI for environmental good while minimizing its negative impacts.
References
Sonika Nitin Nimje (2024) World Environment Day 2024: Date, Theme, History, Importance, Quotes, Buisness Standrard.
Mike Thomas, Matthew Urwin (2024) The Future of AI: How Artificial Intelligence Will Change the World, Built In.
Alex de Vries (2023), The growing energy footprint of artificial intelligence, Joule, Volume 7, Issue 10, 2023, Pages 2191-2194, ISSN 2542-4351.
Karen Hao (2019) Training a single AI model can emit as much carbon as five cars in their lifetimes, Artificial Intelligence, MIT Technology Review.
Thangam, Dhanabalan, et al. (2024) "Impact of Data Centres on Power Consumption, Climate Change, and Sustainability." Computational Intelligence for Green Cloud Computing and Digital Waste Management, edited by K. Dinesh Kumar, et al., IGI Global, 2024, pp. 60-83.
Zoe Kleinman and Chris Vallance (2023), Warning AI industry could use as much energy as the Netherlands, BBC News.
Heather Clancy (2022) Sip or guzzle? Here's how Google's data centres use water, GreenBiz
Forti, V., et al. (2020). The Global E-waste Monitor 2020: Quantities, Flows, and the Circular Economy Potential. United Nations University (UNU).
Google (2021) Google Water Stewardship - Accelerating positive change at Goolge and beyond.
Charmaine Lai (2023) AI is harming our planet: addressing AI's staggering energy cost,Numenta
Comments