A recent UNESCO report warns that artificial intelligence's energy consumption has reached unsustainable levels, with AI demand doubling every 100 days. The report highlights the environmental impact of generative AI tools, which consume significant amounts of energy and water. For instance, ChatGPT receives around a billion requests daily, amounting to 310 GWh annually, equivalent to the annual electricity consumption of three million people in Ethiopia.
The report suggests that AI's annual energy footprint is equivalent to that of a low-income country and is growing exponentially. To mitigate this, the report recommends using smaller models, shortening prompts, and model compression to reduce AI's energy appetite. Task-specific models can reduce energy consumption by up to 90% without compromising performance. Companies like Google, Microsoft, and OpenAI are already offering miniature versions of their large language models.
Concise queries can also cut energy use by more than 50%. Reducing prompt length from 300 to 150 words can make a significant difference. Model compression techniques like quantization can achieve energy savings of up to 44% without affecting performance.
The report emphasizes the importance of reducing AI's energy consumption to ensure sustainable development and mitigate its environmental impact. Smaller, more efficient models can also make AI more accessible in low-resource settings, bridging the technology gap between high-income and low-income regions.
To address this issue, the report urges policymakers to mandate energy transparency, support open-source model development, and integrate environmental sustainability into AI procurement policies. By taking these steps, we can work towards a more sustainable future for AI and reduce its environmental footprint.
