The efficiency of artificial intelligence (AI) comes with a significant environmental cost. A study from the University of Massachusetts exposes the energy-intensive nature of training large AI models, resulting in a substantial carbon footprint. The findings indicate that the training process alone can generate approximately 626,000 pounds of carbon dioxide, equivalent to nearly five times the lifetime emissions of an average car.
While AI is recognized as a transformative technology, its environmental impact, including development, maintenance, and disposal, poses a pressing concern. In this article, I would like to discuss the concealed environmental challenges associated with AI and the critical need for proactive measures. Join me to learn how we can mitigate the risks that AI introduces to our world.
Problem Definition
The comparison between the artificial intelligence (AI) industry and the oil industry extends beyond economic parallels to include environmental impacts. Analogous to oil’s economic significance, data in AI is viewed as a lucrative commodity. However, the parallel further emphasises the outsized environmental impact of deep learning, a fundamental process in AI.
Deep learning, involving intensive training of neural networks, mirrors the resource-intensive nature of the oil industry. As AI advancements demand more computational power, the environmental toll becomes a significant concern, prompting discussions on sustainable development practices, energy-efficient hardware, and eco-friendly alternatives to address the ecological consequences of AI progress.
Positive Environmental Impact
Prevent Deforestation
Organisations like WWF utilise AI, such as Forest Foresight, analysing satellite images to detect and prevent illegal deforestation early by identifying potential areas for illegal activities.
Fight Wildfires
Artificial Intelligence collects and analyses data from sources like camera footage and emergency calls. This enables accurate prediction of fire growth and direction and helps to fight fires quicker and more efficiently.
Detecting Wildlife Diseases
AI algorithms swiftly analyse digital images to detect wildlife diseases in their early stages, providing a faster and error-resistant approach to disease management.
Identifying Recyclables
AI-guided robots developed by companies like AMP Robotics and MachineX are more efficient than humans in identifying and recovering recyclables. This significantly aids in recycling processes and reduces greenhouse gas emissions.
Recycle and Reuse Water
AI monitors water quality, identifying contaminants and signs of pollution. Early detection allows for timely actions by water agencies to prevent pollution. AI also aids in the extraction of chemicals from already contaminated water.
Optimising Operations in Different Industries
AI’s ability to process vast amounts of data quickly enhances precision agriculture. By analysing weather, soil, and crop data, AI tools increase yield, reduce environmental harm, and promote sustainable farming practices. It optimises transportation routes, reducing gas emissions by generating efficient routes and addressing traffic flow problems. Predictive analysis helps avoid or prevent traffic congestion. Artificial intelligence contributes to dynamic pricing on nearly expired food, improves inventory management, and predicts customer behaviour, thus reducing waste. Predictive maintenance by AI prevents costly repairs and enhances sustainability. With predictive algorithms and efficient production and transportation management, AI aids organisations in reducing their carbon footprint and building sustainable supply chains.
Negative Environmental Impact
Carbon Footprint
The carbon footprint associated with the training process of natural language processing (NLP) models in artificial intelligence (AI) has been explored in the above-mentioned research. The paper focused on prominent NLP models like Transformer, ELMo, BERT, and GPT-2, and it was found that as model size increased, computational and environmental costs grew proportionally. Additional tuning steps, such as neural architecture search, significantly raised costs for minimal performance gains. The most costly model, BERT, had a carbon equivalent of around 1,400 pounds, comparable to a round-trip trans-America flight. The research emphasises the need for greater efficiency awareness in AI development due to the substantial environmental impact, particularly as large neural networks become more prevalent.
Impact of AI on Ecosystems
AI applications, including driverless vehicles and drones, pose threats to natural ecosystems, leading to increased waste and potential environmental damage. The use of AI in agriculture may contribute to soil and water contamination through pesticide overuse, potentially leading to biodiversity loss. Ethical concerns arise in AI-based environmental management, where biassed decisions may prioritise economic growth over sustainability.
Electronic Waste Challenge
The disposal of electronic waste (e-waste) generated by AI technology presents a significant environmental challenge due to hazardous chemicals such as lead and mercury. By 2050, the World Economic Forum estimates e-waste will exceed 120 million metric tonnes, necessitating stringent laws and ethical disposal practices for secure processing and recycling.
Attempts to Reduce Negative Effects
To significantly reduce energy consumption in AI, investing in energy-efficient hardware and algorithms is crucial. Optimising both hardware and algorithms can achieve comparable efficiency with lower energy usage. Additionally, promoting ethical AI design practices, such as avoiding unnecessary data collection and addressing end-of-life considerations, is essential for responsible and sustainable development.
Transition to Clean Energy
Some data farms, like those in Iceland, have transitioned to clean energy sources, using the nation’s abundant hydroelectric and geothermal resources. Iceland’s cold climate eliminates the need for energy-intensive cooling methods which immensely helps to save both financial and natural resources,
Alternatives to Lithium
To address environmental concerns associated with lithium batteries, physical batteries made of concrete offer a promising alternative. This concept is not a distant vision — currently, in a Swiss valley, a 246-foot tower suspends two 35-ton concrete blocks. These serve as an initial prototype for a physical battery, collectively possessing sufficient energy to supply power to two thousand homes (equivalent to two megawatts).
Reuse of Water
Water reuse poses a significant concern in addition to the energy consumption challenges associated with data centres. Emerging technologies now offer the possibility of repurposing the high-temperature water generated by data centres. Ramboll suggests that these data centres have the capacity to supply surplus heat to a heating network. The elevated-temperature water, in turn, can be employed for heating purposes in buildings, greenhouses, and swimming pools.
Free Cooling
Implementing free cooling techniques and elevating operational temperatures in the data centre could result in substantial energy savings, reducing cooling expenses by up to 67%. Free cooling is one of the most eco-friendly cooling methods which involves the use of colder outdoor air to refresh the server premises. For a 10 MW data centre, the energy savings will also translate to annual savings of approximately 2.87 million dollars.
Transparency and Accountability
The lack of transparency and legal regulation of how AI companies operate makes it hard to define the exact impact on the environment. We need clear rules, laws, and responsibility to make sure AI is developed in a way that cares about the environment. It’s important to find a balance between the benefits AI can bring and taking care of the environment for a sustainable future.
Ethical Practices
Additionally, promoting ethical AI design practices, such as avoiding unnecessary data collection and addressing end-of-life considerations, is essential for responsible and sustainable development.
To Recap
The role of artificial intelligence (AI) in modern technology presents both positive and negative environmental impacts. AI’s efficiency, particularly in the training of large models, contributes to a substantial carbon footprint, equivalent to the lifetime emissions of an average car. Despite its transformative potential, AI development, maintenance, and disposal raise environmental concerns, paralleling the impact of the oil industry.
Positive environmental impacts include AI applications preventing deforestation, fighting wildfires, detecting wildlife diseases, identifying recyclables, and optimising operations in various industries. However, the negative environmental consequences involve a significant carbon footprint, threats to ecosystems, electronic waste challenges, and potential soil and water contamination in agriculture.
To mitigate these issues, initiatives such as investing in energy-efficient hardware, transitioning to clean energy sources, exploring alternative battery technologies, and implementing water reuse and free cooling techniques are being considered. Additionally, ethical AI practices, transparent regulations, and responsible development are essential for ensuring a sustainable and balanced future.
