Researchers Develop Neuro-Symbolic AI to Reduce Energy Use and Enhance Efficiency
Researchers at the School of Engineering have created a proof-of-concept for neuro-symbolic AI systems that can use 100 times less energy than current AI models. This new approach combines neural networks with symbolic reasoning, improving task accuracy and reducing training time significantly. Testing demonstrated a 95% success rate on a Tower of Hanoi puzzle, compared to 34% for standard models. The findings will be presented at the International Conference of Robotics and Automation in Vienna in May.

Researchers at the School of Engineering have developed a neuro-symbolic AI system that could potentially reduce energy consumption by 100 times compared to current models. The International Energy Agency estimates that U.S.
AI and data centers consumed approximately 415 terawatt-hours of power in 2024, a figure projected to double by 2030. The new system, created in the lab of Matthias Scheutz, combines conventional neural networks with symbolic reasoning.
In tests, the neuro-symbolic model achieved a 95% success rate on a Tower of Hanoi puzzle, significantly outperforming standard visual-language-action (VLA) models, which only managed 34%. The neuro-symbolic model required just 34 minutes of training and used 1% of the energy needed for VLA training, continuing to save energy during task execution.




Comments