The researchers conducted a study on the carbon footprint of various AI models, such as chatbots, essay editors, image classifiers and image generators. They measured the energy consumption and the carbon emissions of these models when they performed a given task 1,000 times on a select set of data.
They found that text-based models, such as chatbots and essay editors, were relatively energy-efficient, requiring about the same amount of energy as charging a smartphone to 16%. However, image-based models, such as image classifiers and image generators, were much more energy-intensive, consuming as much energy as 950 smartphone charges (11.49 kWh) for the least efficient image generation model.
The researchers also noted that there was a large variation between the performance of different image generation models, depending on the size and quality of the images they produced. They suggested that more research was needed to understand the trade-offs between the environmental costs and the benefits of AI models.
Managing Environmental Costs Across the AI Industry
The study highlighted the importance of considering the environmental impacts of AI, as the global AI market is projected to reach a value close to $1,600 billion by 2030. The amount of data generated by the digitized economy is also growing exponentially, reaching 163 trillion gigabytes by 2025. These trends imply that the energy demand and the carbon emissions of AI will increase accordingly.
However, AI can also contribute to solving environmental problems, by providing new insights and facilitating effective environmental governance. For example, AI can help monitor and predict climate change, optimize renewable energy sources, reduce waste and pollution, and promote sustainable consumption and production.
Therefore, the researchers called for a balanced approach to AI, that supports its responsible use in the context of climate change. They said that their study was a first step to provide data points that can inform both AI researchers and practitioners, as well as policy-makers who are working towards regulating the environmental impacts of AI models.