AI Breakthrough Cuts Energy Use by 100x While Improving Accuracy

Artificial intelligence already consumes over 10% of US electricity, and the demand is accelerating. Now, researchers at Tufts University have unveiled a radically more efficient approach that could slash AI energy use by up to 100 times while actually improving accuracy. The system combines neural networks with human-like symbolic reasoning, helping robots think logically instead of relying on brute-force trial and error.

The Problem of AI's Energy Appetite

Large AI systems, from ChatGPT to autonomous robots, rely on massive neural networks that process billions of parameters. These models are powerful but extraordinarily energy-hungry. AI-supporting server facilities can consume as much energy as a small to mid-sized city.

The Tufts team, led by professor Matthias Scheutz, focused on Visual-Language-Action (VLA) models used in robotics. These systems take visual data from cameras and language instructions, then translate them into real-world actions, controlling a robot's wheels, arms, or fingers.

Neuro-Symbolic AI: The Best of Both Worlds

The team's approach, called neuro-symbolic AI, breaks down tasks the way humans do: into logical steps and categories. Instead of learning everything from scratch through massive data consumption, the system applies structured reasoning alongside neural networks.

What Is Neuro-Symbolic AI?

Traditional AI (neural networks) learns by processing enormous datasets and recognizing patterns, like a child who learns by watching millions of examples. Symbolic AI reasons using rules and logic, like following a recipe. Neuro-symbolic AI combines both: the pattern recognition of neural networks with the logical reasoning of symbolic systems. The result is a system that needs less data, less energy, and makes fewer mistakes.

Practical Impact

The research will be presented at the International Conference on Robotics and Automation in Vienna in May 2026. If the approach scales beyond robotics to other AI applications, it could significantly reduce the enormous energy footprint of artificial intelligence.