What to Know About ChatGPT's Impact on Planetary Resources
One of the primary concerns surrounding ChatGPT is the environmental impact associated with its training and inference processes, which are known for their high energy consumption. Estimates suggest that training ChatGPT generates carbon dioxide emissions comparable to those produced by nearly five average American cars over their entire lifetimes. The energy consumption for training ChatGPT is even more staggering, equating to that of an American household for more than 700 years.
If that surprises you, you’re not alone. This substantial environmental footprint underscores significant concerns about the sustainability of large-scale language models, highlighting the pressing need for more energy-efficient AI practices.
Here’s what journalists should know.
Some People Believe ChatGPT Has Positive Environmental Impacts
One way to look at it is that AI could possibly boost economic growth, which some fear might hinder climate progress, leading to calls for degrowth. While economic activity is linked to higher emissions, deep decarbonization requires significant changes that are more achievable with economic growth. AI has noteworthy potential to drive climate innovation, reversing the slowdown in overall innovation and fostering advancements in various fields. However, AI could also lower the costs of fossil fuel production and other high-emission industries. The key is "directed innovation," where society prioritizes climate-positive advancements through policy and other means. Whether society is ready to focus on climate change will determine the balance of AI's impact.
Data Centers' Impact on the Environment
Data centers, which house the power-hungry servers essential for AI models, consume a remarkable amount of energy and contribute substantially to the carbon footprint. AI developers, including OpenAI, depend on these data centers for training algorithms and analyzing data through cloud computing.
Training a single LLM reportedly results in a carbon footprint of approximately 300,000 kg of CO2 emissions, comparable to 125 round-trip flights between New York and Beijing. Additionally, cloud computing currently accounts for about 0.5% of the world’s total energy consumption, a figure expected to exceed 2% in the coming years. The exact emissions vary based on the energy sources powering these data centers—coal or natural gas-fired plants produce higher emissions compared to solar, wind, or hydroelectric power—making precise figures challenging to determine. As language models continue to expand, it becomes increasingly necessary to explore methods for mitigating their environmental impact to ensure a sustainable future.
Reducing AI’s Impact
One approach to addressing the environmental impact of AI is advocating for greater transparency in machine learning development and operation. Scholars have created foundations to help researchers report their energy and carbon usage, promoting accountability and responsible practices. Online tools have been made available to benchmark energy usage, encouraging eco-friendly trials, regular updates on energy and carbon measurements, and careful assessment of energy-performance trade-offs before deploying models.
Individuals also play a key role by tempering the hype around AI systems like ChatGPT and recognizing the limitations of language models. Contextualizing achievements and acknowledging trade-offs can expedite research that doesn't rely solely on developing larger models. This would encourage responsible AI practices and support developing "greener" AI.
Journalists covering ChatGPT's environmental impact should focus on its substantial energy consumption and carbon emissions during training and inference. They can also focus on the public's role in understanding AI's limitations, promoting responsible AI use, and supporting the development of environmentally sustainable AI technologies. Another key point to go over is the potential conflict between AI-driven economic growth and climate progress, with suggestions for directing AI innovation towards climate-positive solutions.
