AI Against Climate Change: Challenges and Solutions

In 2022 scientists from NASA’s Goddard Institute for Space Studies reported about a 2 degrees Fahrenheit increase in the global temperature since the late 19th century. The reason for this warming trend that has been steadily increasing in is human activities. Human-driven greenhouse gas emissions not only did not stop or reduced but actually reached the highest record in 2022. Data that has been collected from weather stations, Antarctic research stations, ships, and ocean buoys show a consistent warming trend, but the question that remains open is what we can do to change it?

The promise of AI to help combat climate change

In recent years, scientists and technology leaders have started to explore using data science techniques and AI/machine learning tools to reduce greenhouse emissions, i.e., the release of gases that trap the heat in the Earth’s atmosphere caused by burning fossil fuels. For example, Peter Clutton-Brock, co-founder of the Centre for AI & Climate, a Britain-based think tank, said to the Thomson Reuters Foundation that the technology was "pushing back boundaries" for climate modeling by analyzing a huge amount of unstructured data, such as images, maps, or graphs that otherwise would not be feasible. The automated technology is currently being used for monitoring deforestation in the Amazon, exploring and understanding the dynamics of sea levels and ice sheets, sending alerts in case of natural disasters, and designing greener smart cities. On a smaller scale, AI could also help monitor a single household, for instance, by automatically switching off lights.

Finding ways to incorporate AI to advance the climate efforts

The goal articulated by the global community is to reach net-zero emissions by 2050. While AI can help spur the climate efforts, as shown by the survey conducted in 2022 by the Boston Consulting Group of public and private-sector leaders, many respondents believe there are significant barriers in using AI in the organizations for their own climate change efforts. Specifically, while 87% of respondents believed that AI is a helpful tool in the fight against climate change, 78% of respondents indicate insufficient access to AI expertise in their organizations. To tap into the AI’s potential to enable data-driven approaches to combat carbon emissions, Boston Consulting Group proposed a framework. The framework has three separate parts where AI can be used: (a) mitigation, (b) adaptation and resilience, and (c) fundamentals. Specifically, mitigation consists of estimating the carbon footprint, forecasting and improving energy efficiency, and monitoring the main carbon-capture storage sites. In short, AI could help gather and analyze complex data that would further help to inform policymakers on how to better make decisions related to climate change. The adaptation and resilience field consists of projecting long-term trends and building early warning systems for extreme weather events. In addition, AI could be used to help manage crises, strengthen infrastructure, and preserve biodiversity by identifying and monitoring the crisis epicenters or species that are in danger. Finally, fundamentals include the research and modeling of the social and economical changes in relation to climate change. For example, forecasting carbon prices or giving recommendations for climate-friendly consumption for individuals as well as companies or entire countries.

The figure shows the framework proposed by the Boston Consulting Group to use AI to advance climate efforts.

Better predictions may not equal a rapid progress

A difficult truth to acknowledge, as argued by some scientists, for instance, Vaclav Smil is that rapid abandoment of burning fossil fuels is unlikely. In his book How the World Really Works: The Science Behind How We Got Here and Where We're Going, Smil gives various examples why this is the case, including food production. Food prouduction is dependent on burning fossil fuels, from producing meat to harvesting vegetables. For example, making 1 kilogram of bread requires around 210 – 250 mL of diesel, harvesting 1 kilogram of tomatoes – 650 mL of diesel (i.e., for fertilizer, plastic greenhouses, etc.), and for a kilogram of roasted chicken, it is around 300 – 350 mL of diesel.

Another challenge altogether is whether there is sufficient political will to follow the predictions derived from machine learning models. This issue is well demonstrated by a situation that occurred in the city of Flint, Michigan, in the United States, where drinking water contamination resulted from lead pipes in some houses. A group of computer scientists developed a model that could predict which pipes required replacement (i.e., the locations to be dug up and the lead pipes to be replaced with the copper ones), with an 80% accuracy rate. The model helped prioritize areas where lead pipes were most likely to be present, and four out of five areas examined did indeed contain lead pipes, saving significant amounts of time and money. However, despite the model's proven accuracy and success, city officials began to ignore its recommendations after complaints from citizens started to appear. This resulted in a decrease in success rates to around 20%, with hundreds of copper pipes being dug up needlessly. Eventually, a court mandate enforced the use of the data-driven model's recommendations. This story illustrates that while machine learning models may offer accurate predictions, prediction and human judgment are not equivalent. Ultimately, the success of data-driven predictions will depend on the willingness of individuals to take into account the model's recommendations.

The dual role of AI in climate change efforts

AI can be highly beneficial in helping to solve the climate crisis, however, AI itself is a significant emitter of carbon whose environmental impacts need to be measured. Researchers from the University of Massachusetts Amherst in 2019 estimated that training a single big language model (think of ChatGPT) is equal to around 300,000 kg of carbon dioxide emissions. To give you an estimate, it is equivalent to 125 round-trip flights between New York and Beijing. Furthermore, a single model is usually not sufficient to achieve complex tasks and the computing power drastically increases with the task complexity. When OpenAI developed a one-versus-one bot in 2017 for playing the video game Dota-2, it required around 60,000 CPU cores on Microsoft Azure, however, when OpenAI Five has been developed as a general-purpose reinforcement learning system, the computational requirements increased to 128,000 pre-emptible CPU cores on the Google Cloud Platform. Thus, often when using big and complex AI models researchers require having access to data centres that could account for 10% of the total electricity used in the world. Currently, only storing data itself from online activities like emails or videos already account for about 1% of global electricity use according to the International Energy Agency.

The Power of Individuals to Fight Climate Change

Helping to fight climate change is not solely in the hands of large organizations, as individuals, we can each contribute to this cause. This contribution can include lifestyle choices to help reduce carbon emissions. Another way is to use one’s computational and data science skills to support the efforts of such communities as the Collaborative Earth. Through the Open Science framework Collaborative Earth aims to bring together individuals with different expertise to advance ecological restoration work. For instance, one of the projects undertaken by the Collaborative Earth involves using remote sensing data and convolutional neural networks to detect beaver dams. This information can then be used to monitor the impact of beaver damns on ecosystems. Beavers have many beneficial impacts on soil and vegetation, as well as helping to control water flow. However, in areas where beavers come into contact with human settlement, they often face threats of being killed. Having detailed information about the climate benefits provided by the beavers would be relevant for informing the beaver management policy. If you are curious about how you can contribute to making a difference, you can visit the Collaborative Earth's website to learn more about their different projects. You may find a project that speaks to your skills and interests and be part of the solution in fighting climate change.