GREEN AI: BUILDING ENERGY-EFFICIENT ARTIFICIAL INTELLIGENCE : A Review of Techniques, Trends and Research Gaps

Artificial Intelligence (AI) has become a major asset now in various industries. However, developing and running high-performance AI models usually require massive computing power, which leads to enormous power and energy usage. Green AI aims at significantly reducing these demands and environmental damage by finding ways to make AI models both highly accurate and environmentally friendly. This work gives an overview of Green AI key concepts and the most recent researches done in this direction. It starts by introducing Green AI along with its role in sustainable computing and proceeds to explore the various ways that Green AI is being implemented such as pruning, quantization, model compression, efficient network design, TinyML, edge deployment, and hardware-level optimization. Also, this paper estimates the ecological cost of AI and compares ten articles that have been published in the period 2015 - 2025. The findings highlight several problems yet to be solved for instance the trade-off between model accuracy and energy consumption, no agreement on a standard method of disclosing energy and carbon emissions, the computational burden of massive neural networks, the limited implementation by the industry, and the necessity of research on environmentally friendly Large Language Models (LLMs) and generative AI