Neuro-Symbolic AI slashes energy use and boosts robot task accuracy

A new neuro-symbolic AI system developed by researchers at a leading School of Engineering significantly reduces energy consumption while enhancing the accuracy of robotic tasks. The breakthrough, which combines traditional neural networks with symbolic reasoning, promises to cut energy use by up to 100 times and outperform current AI systems.

Key Highlights

The research, led by Matthias Scheutz, Karol Family Applied Technology Professor, introduces a hybrid approach called neuro-symbolic AI. This method integrates the strengths of both neural networks and symbolic reasoning, mirroring human problem-solving techniques. The team's work will be presented at the International Conference of Robotics and Automation in Vienna in May.

Energy Savings and Performance Boost

According to the International Energy Agency, AI systems and data centers consumed about 415 terawatt hours of power in 2024, accounting for over 10% of the U.S. total electricity production. This demand is projected to double by 2030, raising significant sustainability concerns. The new neuro-symbolic AI system addresses these issues by drastically reducing energy consumption and improving performance on complex tasks.

How It Works: Visual-Language-Action (VLA) Models

The neuro-symbolic AI focuses on visual-language-action (VLA) models, which are used in robotics. These models extend the capabilities of large language models (LLMs) like ChatGPT by incorporating vision and physical movement. VLA models take in visual data from cameras and instructions from language, then translate that information into real-world actions, such as controlling a robot's movements to complete a task.

Symbolic Reasoning Enhances Accuracy

Traditional VLA systems rely heavily on data and trial-and-error learning, often leading to mistakes. For example, a robot tasked with stacking blocks may struggle with shadows or incorrect placements. Symbolic reasoning, on the other hand, uses rules and abstract concepts like shape and balance, allowing the system to plan more effectively and avoid unnecessary trial and error.

"Like an LLM, VLA models act on statistical results from large training sets of similar scenarios, but that can lead to errors," said Scheutz. "A neuro-symbolic VLA can apply rules that limit the amount of trial and error during learning and get to a solution much faster. Not only does it complete the task much faster, but the time spent on training the system is significantly reduced."

Impressive Results in Testing

The researchers tested their system using the Tower of Hanoi puzzle, a classic problem that requires careful planning. The neuro-symbolic VLA achieved a 95% success rate, compared with just 34% for standard systems. When given a more complex version of the puzzle, the hybrid system still succeeded 78% of the time, while traditional models failed every attempt. Training time also dropped sharply, with the new system learning the task in only 34 minutes, compared to more than a day and a half for conventional models.

Massive Energy Savings

Energy consumption was reduced dramatically as well. Training the neuro-symbolic model required only 1% of the energy used by a standard VLA system. During operation, it used just 5% of the energy. These savings have significant implications for the sustainability and efficiency of AI systems, especially in the rapidly growing field of robotics.

References

• AI breakthrough cuts energy use by 100x while boosting accuracy | ScienceDaily