Artificial intelligence (AI) has emerged as a powerful tool across various industries, revolutionizing healthcare, astronomy, and more. While AI offers numerous advantages, including its role in addressing climate change, it also poses environmental challenges. This article explores the environmental impact of AI, focusing on energy consumption, carbon emissions, and water usage, as experts grapple with the complex balance of benefits and harm.
AI’s dual nature
Determining whether AI technologies bring more harm or good is a complex task. On one hand, AI enhances efficiency and productivity in sectors like healthcare and climate science. Conversely, certain AI systems consume substantial energy and amplify industries already harming the environment, posing ethical and environmental dilemmas.
The environmental impact of AI remains poorly understood, requiring further research to assess its full scope. Experts stress the need for a comprehensive examination of AI technologies, such as Google’s Bard and ChatGPT, which are recognized for their text-based capabilities.
Large language models (LLMs) like Bard and ChatGPT have garnered attention for their energy-intensive nature. These models demand massive computing energy during both training and utilization phases, raising concerns about their carbon footprint.
Carbon emissions in AI training
A recent report by the Canadian Institute for Advanced Research (CIFAR) sheds light on the carbon emissions associated with training LLMs. Training is an energy-intensive process, with specialized computer chips becoming more powerful in recent years. This boosts learning speed but also escalates resource consumption, including electricity from non-renewable sources, contributing to carbon emissions.
According to the research, Microsoft’s GPT-3 emitted 502 tonnes of CO2 during its training, equivalent to the annual emissions of 304 homes. DeepMind’s Gopher, a 2021 LLM, emitted 352 tonnes of CO2. Importantly, carbon emissions continue even during AI usage when responding to queries, as each interaction has a carbon impact.
Water depletion by AI
AI systems, particularly LLMs, generate heat during operation, necessitating cooling mechanisms. This results in the consumption of vast quantities of fresh water, further straining environmental resources. For example, Google’s data centers consumed 12.7 billion liters of fresh water for cooling in 2021, while Microsoft GPT-3’s training center used around 700,000 liters of fresh water.
Smaller algorithms’ impact
Smaller AI algorithms, like Bloom, also contribute to environmental concerns. While their individual impact may seem minor, their cumulative effect can be substantial when deployed in user-facing applications that receive millions of queries daily. These calculations do not account for additional environmental costs related to building cooling, computer storage, and server networks.
Efforts towards sustainable AI are gaining momentum. Microsoft, for instance, is investing in research to measure AI’s energy use and carbon impact while striving for efficiency improvements. Companies are also emphasizing the use of clean energy sources to power data centers and aligning with sustainability goals.
AI’s diversity of impact
AI’s environmental impact varies significantly based on its application. Predictive AI models can contribute positively by aiding climate change mitigation, such as detecting deforestation. However, the same technology can exacerbate environmental issues when applied to accelerate resource extraction in industries like oil and gas exploration.
AI can play a pivotal role in combating climate change when used judiciously. It aids in understanding environmental patterns, forecasting natural disasters, and optimizing resource usage in various sectors. While large language models may capture attention, the broader field of AI encompasses a wide range of applications with varying energy requirements.
AI beyond Large Language Models
AI’s impact extends beyond high-profile LLMs, affecting various facets of society. Recommender systems, control algorithms for factories, satellite imagery for agriculture, and financial forecasting algorithms all rely on AI. These applications often consume less energy than LLMs but have a profound societal impact.
The environmental consequences of AI technologies, particularly large language models, demand careful consideration. As the world grapples with the allure and apprehension surrounding AI, understanding its diverse environmental effects is crucial. Striking a balance between harnessing AI’s potential for positive change and mitigating its environmental costs is an ongoing challenge that requires collaboration among researchers, industry stakeholders, and policymakers.
