The Thermodynamic Barrier to Artificial General Intelligence: A Comprehensive Analysis of Energy Efficiency, Neuromorphic Computing, and the Path to Sustainable AGI
Year: 2025
Type: article
Abstract: The rapid advancement of artificial intelligence has brought unprecedented capabilities in language processing, reasoning, and multimodal understanding. However, the path to Artificial General Intelligence (AGI) faces a fundamental constraint that transcends algorithmic innovation: thermodynamic efficiency. This paper presents a comprehensive analysis of the energy consumption disparity between current AI systems and biological intelligence, examining why the human brain's 20-watt operation represents not merely an engineering benchmark but a physical necessity for scalable, ubiquitous AGI. We analyze the limitations of von Neumann architecture, explore emerging neuromorphic computing paradigms including spiking neural networks and memristive systems, and provide a rigorous timeline for convergence toward biologically-efficient artificial intelligence. Our findings suggest that while algorithmic AGI may emerge by 2027-2029, truly sustainable and democratized AGI requires a fundamental shift in computing substrates, unlikely before 2032-2035.
Author: SÉRGIO DE ANDRADE, PAULO
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7116059910
FWCI:
Cites:
Cited by:
Related to: