The Entropy of Artificial Cognition: Deciphering the Anthropogenic Footprint and Multilingual Energetic Tax of Large Language Models

Abstract

We are currently facing a growing challenge known as the "hidden cost of intelligence," driven by the immense energy consumption required to develop and practically deploy artificial intelligence (AI) models. This research paper analyzes the environmental impact of AI models and evaluates trends in their energy efficiency by integrating data from existing research and benchmark studies. The analysis shows that inference is the main source of energy consumption, accounting for about 90% of an AI model's total lifetime energy use. It also highlights a major measurement challenge, as energy consumption estimates can differ by up to 2.4 times depending on how the system boundaries are defined. Transparency is still a major issue, as more than 84% of AI models released since 2022 do not provide information about their environmental impact. However, methods like 4-bit quantization show the potential of Green AI by reducing emissions by up to 55% without affecting model accuracy. This study concludes that future AI research should give equal importance to energy efficiency and performance instead of focusing only on improving accuracy. Adopting Green AI practices, along with standardized methods for measuring and reporting energy consumption, will improve transparency, increase accountability, and support the long-term sustainability of AI technologies. Keywords: artificial intelligence; Environmental impact; Sustainability; energy efficiency; Green AI