Is AI making climate change threat more severe, or helping find a solution
Climate change is a colossal question the scientists, engineers, and industry experts are grappling with. Wrathful weather has pushed up the temperature by more than a degree in the last century and any further rise can turn into a devastating threat to our ecosystem. It will push up the sea levels, inundating great cities built on the coastline, negatively impact crop production which in turn will result in inflation in prices, taking many eatables out of the purchasing ability of a large section of population, and severely hit the way of life humans have been following since centuries.
AI can serve as a powerful tool to fight climate change in an array of disciplines. The process of climate change is enormously complex with a huge number of variables. People tend to focus on the physical aspects when discussing climate, such as the precipitation level, damage to ozone layer, wind patterns, and level of carbon dioxide in the atmosphere. However, these are just the outcome of the changes happening to the planet earth.
With AI, humans can better understand what is happening behind the obvious. For instance, AI can help understand elements that are constantly evolving with climate change, enabling us to make more informed predictions regarding environmental changes and choose the right set of mitigation efforts. This is important as climate data sets are humongous and one ends up taking plenty of time in collecting and analyzing data.
However, AI is not all hunky dory for climate. There are not just ethical concerns, but also worsening of problem due to AI operations.
Surprised?
Training complex deep learning models consumes enormous amount of computational power, resulting in a significant carbon and water footprint. AI industry requires an unsustainable demand for hardware and the raw materials, negatively influencing air and soil quality in the planet. The industry’s ecological consequences have to be taken into account before deploying AI for controlling damage to the climate.
Carbon footprint of AI
Scholar Kate Crawford noted in her book Atlas of AI that training of AI models requires rare earth minerals such as mercury and zinc in large quantities. Operating data centers requires power that consumes thousands of tons of coal and oil. Computational power and resources needed for running AI algorithms comes hard on the environment.
Sophistication of Amazon Web Services, Microsoft’s Azure and other AI systems comes at an environmental cost. AI bigwig Hugging Face disclosed in a 2022 research paper that developing its language model BLOOM culminated in more than 50 metric tons of carbon dioxide emissions, which amounts to emissions during 60 flights from London to New York. Statistically, BLOOM launch consumed significantly less than other language models around.
A University of Massachusetts paper, published in 2019, revealed that training a single AI model could emit more than 626,000 pounds of carbon dioxide, which is approximately five times the lifetime emissions of the average American car and this includes the manufacturing process. Large language models (LLMs) such as GPT-3, GPT-4, and PaLM2 (powering Bard) are more energy-intensive.
Environmental degradation doesn’t end at training and actual use of AI system could emit much more CO2. A UC Berkeley and Google research discovered that training GPT-3 took 1,287 megawatt hours of electricity. This energy was enough to supply an average American household for about 120 years. Around March of 2023, millions of daily queries hit the chatbox, consuming about 4,000 to 6,000 megawatt hours of electricity, which was three to five times energy spent in training the system.
GPT is just one of the models on the market. As AI adoption increases, it is a given that environmental costs will go up as well.
Water footprint of AI
When training AI, voluminous amount of water is thrust through data centers to prevent overheating of servers. According to a 2023 study, 700,000 liters of fresh water a day was pumped through Microsoft’s US data centers training GPT-3. US data centers of Google needed 12.7 billion liters of clean water for on-site cooling in a single year.
Water footprint of AI cannot be measured without factoring in the water consumed while producing semiconductors. Graphics processing units (GPUs) are an integral part of the AI training process. Shaolei Ren, an associate professor of electrical and computer engineering at University of California Riverside, mentioned in a 2023 paper, that an average semiconductor factory would require millions of gallons of water a day for cleaning.
As water isn’t a resource even distributed across the Earth, arranging clean water for AI’s massive consumption becomes a problem. To ensure super-quick responses from ChatGPT, OpenAI needs to set up its servers across the globe. Many of these regions have paucity of water and this creates a conflict with the local users. Water consumption of AI is a topic that cannot be left unattended.
But AI is a part of solution as well
Carbon and water footprint of AI notwithstanding, AI can be used efficiently for the good of the environment. As a 2020 research paper suggests, of the 169 targets set by the United Nations in its Agenda for Sustainable Development, AI could assist significantly in 79 percent.
A report published by the nonprofit Climate Change AI in 2022 discusses how artificial intelligence could be used to positively influence climate change. It could serve as a robust tool for bringing in an array of advantages such as facilitating precision agriculture, optimizing power grids, enabling smart buildings, and boosting transportation systems. AI could be used in several ways for promoting energy efficiency and reducing greenhouse gas emissions.
Predictive abilities of AI can come handy from climate modeling to the honing of mitigation policies. Thanks to AI, policy makers get a tool to weigh the pros and cons of a certain approach. AI could play a positive role in engineering, social sciences and policy. The technology is particularly effective when an execution needs to be done in lesser time or at a larger scale.
Specific applications of AI related to climate change
AI can help us cultivate insights into uncertainties related to climate change. This would help us improve the existing economic models and carve out better observation programs. AI will help understand elements of climate change that are themselves undergoing a process of evolution, enabling us deploy better organized mitigation efforts.
Let us have a look into few specific applications of AI that can assist in the endeavor against climate change.
Oceans have a role to play in the climate change as they transfer and absorb heat. However, oceans are among the least studied parts of the planet. We still lack any clear idea how these vast water bodies respond to various environmental changes or how the climate change influences the oceans themselves. AI can help understand the exact relationship between the oceans and climate change.
AI can also assist satellites orbiting the Earth to make observations how various environment-related factors influence climate change. Satellites have been effective in monitoring forest fires and how they affect the volume of carbon dioxide in the atmosphere. They can help keep a watch on any object or phenomenon on the planet with a relatively neutral approach because of the distance. AI can also help in ensuring safe placement of the satellites in the orbit and ensure they disseminate correct information.
The Arctic can serve as a good example how various factors affect climate change. It is changing in a dramatic way and the temperatures are increasing steadily. Ships collect data required by scientists and policymakers in the Arctic during spring, summer, and fall. However, ice conditions in the winter force the ships to depart, creating a significant gap in data collection. AI-powered robots can prove quite useful in scenario, facilitating collection of information regardless of the season, and help in understanding trends and patterns.
The way ahead
Ideal path ahead will be to reduce the carbon and water footprint of AI, while roping it for the stuff that it can do better than humans. Used efficiently, AI can make the endeavor against climate change more sustainable.
Thankfully, tech conglomerates such as Google and Microsoft are aware of the need to make their operations environment-friendly. Within the next decade or so, they plan power all their operations with renewable energy. They are also working to run training of AI systems at optimal times. For instance, they might combine the regular power source with solar panels, and train the systems when the sun is highest in the sky to reduce carbon footprint.
Another strategy could be to run water-guzzling operations in the area that are water-efficient. For example, a data center in Canada is likely to be more water-efficient compared to a data center in Arizona, America. For data center operations, companies could choose places where water is in abundance. If training is being imparted with conventional power sources, it could be done at night to bring down water consumption.
Technology behind AI is also developing fast. Models like transformers are capable of processing more data in less time, which slashes energy consumption. For example, transformers model take less time in processing more data. Settings of the cloud service can be changed to ensure that the training consumes less energy.
Frugal AI is an approach that focuses on designing more robust models with less data. These models require less data and fewer features for predicting. Lesser data restricts the amount of computational power of memory which reduces the resources used for training.
A game-changer application of AI is to provide industries and businesses with tools for accurate measurements of their environment footprints. Machine Learning Emissions Calculator enables you to find the carbon footprints of your AI models and take remedial measures. Such tools make AI fundamentally aligned with efforts related to climate change.
