Green AI: mitigating the environmental impact of machinelearning

Abstract

Artificial intelligence (AI) and machine learning (ML) are changing the world in exciting ways. Yet there is a growing concern that is often overlooked—the huge amount of energy these technologies use. Training a single large AI model can produce as much carbon pollution as five cars do in their entire lifetimes. This paper explores the concept of Green AI, which is the idea of making AI smarter and more efficient so it uses less energy and causes less harm to the planet. We reviewed many research articles from 2019 to 2025 to identify major challenges, examine the methods used to solve them, and highlight areas that still require further research. Our main findings show that certain techniques—such as making models smaller, removing unnecessary parts, and using cleaner energy sources to power computer systems—can reduce energy use by up to 32% without making the AI less accurate. However, there is still no common system for measuring and reporting how much energy AI uses, which makes it challenging to compare different studies. This paper also suggests new research directions, including brain-inspired computer chips, shared training across many devices, and stronger government policies to push the AI industry toward greener practices.