Researchers at Tufts University School of Engineering have developed a neuro-symbolic AI system that slashes energy consumption by up to 100 times compared to conventional models while nearly tripling task accuracy. The team, led by Professor Matthias Scheutz, combined traditional neural networks with symbolic, rule-based reasoning to create a system that mirrors how humans solve problems by breaking them into logical steps.

The breakthrough targets visual-language-action (VLA) models used in robotics rather than large language models like ChatGPT. In tests using the Tower of Hanoi puzzle, the neuro-symbolic system achieved a 95% success rate compared to just 34% for standard VLA models. Training took only 34 minutes versus more than 36 hours for conventional approaches, consuming just 1% of the energy.

The findings come as AI data centers consume as much electricity as small cities, with global AI energy demand projected to double by 2028. The research will be presented at the International Conference of Robotics and Automation in Vienna in May.