Artificial intelligence is in trouble, and it’s time for humanity to step up and help it out, the Green Screen Coalition warns through Save the AI, an online spoof that launched during the AI Action Summit in Paris earlier this month.
“Artificial intelligence is facing a crisis: humans are consuming far too many precious resources that AI needs to thrive,” the website warns. “Every sip of water you take and every light you turn on could be sustaining the AI systems that uphold your digital conveniences.”
To safeguard those conveniences, and AI itself, the solution is obvious, the site adds: “Reduce your own consumption: drink less water, take shorter showers, and sit in the dark to ensure AI has enough resources to keep going,” it urges. And “raise awareness: join our campaign to stop reckless human consumption and put AI first.”
The site lays out the resources artificial intelligence needs for routine operations: 500 millilitres (17 ounces) of water and 140 watt-hours of electricity to write a 100-word email, 5.4 million litres of water to train a model like ChatGPT-3, and 50 gigawatt-hours of power to train ChatGPT-4, enough to run 6,000 average U.S. homes for a year.
With these and other examples, the spoof gets serious, asserting that energy efficiency improvements alone won’t address AI’s environmental impact. “Water and carbon footprints are complementary rather than interchangeable when addressing environmental impact, as optimizing for one may worsen the efficiency of the other,” the site explains. So “environmental impact needs to be addressed holistically: Addressing the water footprint along with the carbon footprint, emissions, and other impacts is critical.”
It adds that, “while the water or electricity consumption per request may decrease in the future, the total water and energy consumption is likely to rise further as a result of the growing demand for AI services and the increasing scale of AI applications.”
Save the AI calls for full transparency and comprehensive reporting on the technology’s resource consumption and environmental consequences. “AI companies must be required to publicly disclose the electricity, water, and emissions footprint of AI models, including off-site (Scope 2) and supply chain (Scope 3) usage.”
