GENEVA A groundbreaking report by UNESCO, unveiled at the AI for Good summit in Geneva, offers a simple yet powerful solution to drastically reduce AI energy consumption by up to 90% use shorter prompts and smaller, specialised AI models.
The global study addresses mounting concerns over the surging energy demands of generative AI tools like ChatGPT, Google Gemini, and Microsoft Copilot, all of which rely on vast computing infrastructure. According to the report, optimising the way we interact with AI specifically by reducing prompt length and choosing lightweight models can make a significant environmental difference.
UNESCO’s findings revealed that shortening prompts from 300 to 150 words and switching from a large general-purpose model to a smaller, task-specific model led to energy savings of nearly 90%, without any loss in output quality.
With AI usage skyrocketing globally, the environmental footprint is becoming harder to ignore. OpenAI CEO Sam Altman noted that each ChatGPT prompt consumes an average of 0.34 watt-hours far more than a standard web search. With over one billion prompts daily, ChatGPT alone uses an estimated 310 gigawatt-hours annually, equal to the yearly energy consumption of three million Ethiopians.
UNESCO cautioned that the energy demand from generative AI is doubling every 100 days, stressing its strain on global energy systems, water resources, and rare minerals.
However, the industry is adapting. Major tech firms have released miniaturised AI models to improve efficiency. Google’s Gemma, Microsoft’s Phi-3, OpenAI’s GPT-4o mini, and Mistral AI’s Ministral are purpose-built for performance with significantly lower energy costs.
UNESCO’s report calls on governments, tech companies, and users to prioritize eco-friendly AI practices, especially as AI becomes embedded in everything from online searches and digital assistants to automated content creation tools.
“With mindful use and intentional design, AI can power sustainable development but its rising environmental footprint must be responsibly managed,” the report concluded.
Comments (0)
No comments yet. Be the first to comment!
Leave a Comment