In a recent study, researchers have shed light on the carbon footprint of AI tools, and the results are raising eyebrows faster than an AI-powered eyebrow-raising tool. It turns out that the energy consumption associated with using AI has been largely overlooked, like that forgotten piece of broccoli on your dinner plate. According to Jesse Dodge, a research scientist at the Allen Institute for AI, newer and larger generative models are carbon-intensive culprits, making their older AI counterparts look like energy-saving superheroes in comparison.
Remember when Google estimated that an average online search used as much electricity as driving 0.0003 miles in a car? Well, forget about that, because those numbers are likely way off now. Google has integrated generative AI models into its search, and Vijay Gadepally from the MIT Lincoln lab believes the energy usage has skyrocketed since then. It seems like we’ve all been unknowingly contributing to the AI’s carbon footprint while searching for cat videos or the latest conspiracy theories.
But it’s not just the energy consumption during training that we need to worry about; it’s the day-to-day emissions from using AI that are even more concerning. The researchers found that the emissions associated with using AI far exceed those from training large models. In fact, it would take over 590 million uses of Hugging Face’s multilingual AI model, BLOOM, to reach the carbon cost of training its biggest model. Talk about a carbon-heavy burden.
So, what does all of this mean? It means that AI is not as green as we thought, and the responsibility falls on the companies creating these models. As Dodge aptly puts it, they need to be held accountable for their energy usage and emissions. Perhaps it’s time for consumers to start asking questions and demanding greener AI options. After all, why settle for an AI assistant that’s environmentally unfriendly when you can have one that’s both smart and carbon-conscious?
HEALTH
WORLD AFFAIRS
TECHNOLOGY
CREATIVE
