The world of AI (Artificial Intelligence) is a powerhouse of innovation, but when it comes at a cost – an especially hefty one for the planet. Using AI to generate images, write emails, or chat with bots is an energy guzzler. In fact, crafting an image using sophisticated AI models consumes as much energy as juicing up your smartphone completely. However, there's a stark difference: generating text with AI is far less energy-intensive, only sipping 16 per cent of your phone's battery for a thousand prompts.
Image used for representational purposes only.
Eye-Opening Findings
Researchers from Hugging Face and Carnegie Mellon University took a deep dive into this eco-dilemma, unveiling the significant energy and carbon footprint of AI. Their study, though not as yet peer-reviewed, marked the first attempt to calculate carbon emissions caused by employing AI for various tasks.
The findings were eye-opening. Among 10 popular AI functions like text generation and image classification, generating images emerged as the top energy burner. Using hefty AI models for image crafting churns out emissions equivalent to driving about 4.1 miles in a gas-guzzling car for 1,000 images. Conversely, text generation leaves a far smaller carbon footprint, akin to driving a mere 0.0006 miles in a similar vehicle.
Image used for representational purposes only.
While these colossal AI models get trained once, their everyday use emits more carbon than their training. This worrying trend looms large as tech giants infuse these models into products ranging from emails to word processors, clocking millions, if not billions, of daily usages.
A Way Out
There's a glimmer of hope. Opting for more tailored and less energy-intensive AI models could be a game-changer. Using smaller models crafted for specific tasks, like classifying movie reviews, can slash energy consumption by 30 times, compared to hefty, multitasking AI models.
Image used for representational purposes only.
These revelations underscore the urgency for a more conscientious use of AI, steering away from colossal models for every minor task. Yet, the responsibility doesn't solely fall on users, but also on the companies creating these energy-intensive marvels. These studies call for consumer awareness, nudging companies towards sustainable AI practices. As the tech landscape evolves, understanding the environmental price tag of AI becomes essential for a greener future.
Innovative Solutions
In a bid to find eco-solutions to the AI footprint problem, the MIT Lincoln Laboratory Supercomputing Center (LLSC) has been innovating to minimise energy use in data centres, implementing techniques from power-capping hardware to cutting-edge AI training tools. Their eco-friendly practices promote green-computing research, championing a culture of energy efficiency. With a focus on transparent sustainability, LLSC is leading the charge in greening data centres, guiding advancements in energy-saving AI techniques, and setting a new standard for industry-wide energy awareness and accountability.