Environmental Cost of AI Models: Carbon Emissions and Water Consumption
Generative AI models, particularly large language models like GPT-3, have become a major concern due to their significant environmental impact. According to the AI Index Report 2023 by Stanford University, GPT-3 emitted carbon dioxide equivalent to 500 times the emissions of a New York-San Francisco round trip flight in 2022. The report also speaks of similar emissions by other AI models like ChatGPT.
The rising cost of training LLMs and the estimation of carbon emissions of AI systems are some of the factors discussed in the report. The article also highlights the surprising factor of water consumption in AI training and the direct negative environmental impacts, as well as their impact on environmental research. The report emphasizes the need for increased awareness and research to strike a balance between AI advancements and environmental sustainability. This article focuses on the environmental impact of generative AI models, which has become a major concern.
Also Read: Elon Musk’s Urgent Warning, Demands Pause on AI Research
The 2023 Artificial Intelligence Index Report
The Stanford Institute for Human-Centered Artificial Intelligence (HAI) has released its annual AI Index Report, which provides a comprehensive picture of today’s AI world. This year’s report spans 302 pages, a nearly 60% increase from the 2022 report, reflecting the boom in generative AI and the growing effort to collect data on AI and ethics. The AI Index Report compared the carbon emissions of four LLM models. GPT-3 had the highest emissions, even surpassing Gopher, an open-source model trained on large 280B parameters. The multilingual language model BLOOM, with equivalent parameters to GPT-3, produced 25 tonnes of carbon in 2022, which was 20 times lower than GPT-3. Meta’s Open Pre-Trained Language Model (OPT) consumed the least power, with 1/7th the carbon emissions produced by GPT-3.
The Rising Cost of Large Language Models
The capabilities of LLMs like ChatGPT have increased dramatically, but so has the cost of training them. Language models consume the most computing resources among all machine-learning systems, leading to high carbon costs. According to Stanford University’s AI Index Report 2023, GPT-3’s carbon dioxide-equivalent emissions stood at 502 tonnes in 2022, the highest compared to similar-parameter trained models.
Estimating Carbon Emissions of AI Systems
The AI Index team considered factors such as the number of parameters in a model, the energy efficiency of data centers, and the type of power generation used to deliver electricity to estimate the carbon emissions of AI systems. Their analysis concluded that even the most efficient model, BLOOM, emitted more carbon than the average U.S. resident uses in a year.
Water Consumption: A Surprising Factor in AI Training
A new study found that training AI models like ChatGPT and Bard consume significant amounts of water. The amount of water it took to train ChatGPT 3 is equivalent to the water needed to produce 370 BMW and 320 Tesla electric cars. The paper also notes that “ChatGPT needs to drink a 500 ml bottle of water for a simple conversation of roughly 20-50 questions and answers.”
Direct Negative Environmental Impacts: Energy Consumption and Beyond
The environmental impact of generative AI models extends beyond carbon emissions and water consumption. Other direct negative environmental impacts include reducing nature experiences and the risk of worsening the effects of misuse and bias due to simulated authority. The increasing focus on AI technologies can lead to research prioritization effects, widening the digital divide and distracting from environmental research efforts.
Also Read: AI “Could Be” Dangerous – Joe Biden
Our Say
The environmental impact of generative AI models, such as large language models, is a pressing concern as their carbon footprints and water consumption contribute significantly to global emissions and resource depletion. The Stanford AI Index Report 2023 highlights the need for further research and increased awareness about the environmental costs associated with the development and usage of AI technologies. As the world continues to adopt AI in various industries and applications, it is crucial to strike a balance between innovation and sustainability. With AI continuously evolving and becoming increasingly integrated into various aspects of our lives, it is vital to balance the benefits of AI advancements with environmental sustainability.
