Published on December 1, 2023, 6:12 am

AI – The Hidden Carbon Cost

Using artificial intelligence (AI) has become an integral part of our daily lives. Whether it’s generating images, writing emails, or interacting with chatbots, AI is everywhere. But there is a hidden cost to all this convenience – the impact on the planet.

A recent study conducted by researchers at Hugging Face and Carnegie Mellon University sheds light on the environmental implications of using AI. The study reveals that generating an image using a powerful AI model consumes as much energy as fully charging your smartphone. However, the energy required for text generation is significantly less, with 1,000 text prompts consuming only 16% of a full smartphone charge.

The research shows that while training massive AI models is energy-intensive, it’s just one piece of the puzzle. The majority of carbon emissions come from the actual use of these models. This is the first time researchers have quantified the carbon emissions associated with different AI tasks.

The team examined ten popular AI tasks on the Hugging Face platform, including question answering, text generation, image classification, captioning, and image generation. They ran experiments using 88 different models and measured energy consumption with a tool called Code Carbon. The results were eye-opening.

Generating images proved to be the most energy-intensive AI task, releasing significant amounts of carbon dioxide into the atmosphere. For example, generating 1,000 images with a powerful AI model equates to driving approximately 4.1 miles in an average gasoline-powered car. In contrast, text generation was far less carbon-intensive.

These findings are crucial in understanding AI’s true carbon footprint and identifying concerning trends in its upward trajectory. With big tech companies integrating powerful generative models into products like email and word processing software, the demand for generative AI has skyrocketed. Millions or even billions of generative AI tasks are performed every day.

The study highlights that larger generative models consume substantially more energy compared to smaller models designed for specific tasks. Using a generative model for classifying movie reviews, for instance, consumes around 30 times more energy compared to using a fine-tuned model specifically created for the task. Generative models attempt to accomplish multiple tasks simultaneously, such as generating, classifying, and summarizing text, contributing to their higher energy consumption.

The research aims to encourage users to be mindful of their AI usage and opt for specialized models whenever possible. There’s no need for large, all-encompassing models if you’re performing a specific task like searching through emails. Selecting smaller, task-specific models can help reduce carbon emissions significantly.

One missing piece in understanding AI’s carbon footprint has been the energy consumption associated with using AI tools daily. Comparing the emissions from newer generative models with older AI models is also crucial in recognizing the increasing carbon intensity of AI systems over time.

As technology evolves, Google estimated that an average online search now uses more electricity due to the integration of generative AI models into its search engine. Understanding these environmental implications will prompt consumers to inquire about companies’ energy usage and emissions associated with their AI products.

It’s clear that there is both an immediate and long-term ecological impact of using AI. While training large AI models requires substantial energy, it pales in comparison to the ongoing emissions from day-to-day usage. The responsibility lies with companies creating these models to be accountable for their energy usage and take steps towards reducing their carbon footprint.

Studies such as this provide valuable insights into quantifying AI’s environmental impact and raising awareness about its hidden costs. By making informed decisions about our AI usage and demanding more sustainable practices from tech companies, we can mitigate the detrimental effects on our planet while still benefiting from the power of AI.


Comments are closed.