Greening Resource-Hungry AI to Protect the Climate
AI tools like ChatGPT are extremely energy- and resource-hungry, threatening the climate. Researchers are finding ways to "green" AI by making data centers more energy efficient and redesigning AI models to require less computing power.
Summary
- AI demand is surging with the use of energy-intensive tools like ChatGPT and Stable Diffusion image generator. A single ChatGPT question takes 10x more energy than a Google search.
- AI's energy use relies heavily on fossil fuels, releasing greenhouse gases that worsen climate change. If all Google searches used AI, emissions would equal a country like Ireland.
- Larger AI models aren't always better but use exponentially more computing resources. Software improvements like "power capping" reduce energy waste in data centers.
- Green electricity can power data centers but competes with other uses and needs water for cooling. Water scarcity also limits locations.
- Researchers are finding ways to build accurate models that need less computing, inspired by efficiencies in nature.
- Reimagining AI design focuses on sustainability from the start instead of maximum capability at any cost. Metrics and reporting lead to responsibility.