AI SCIENCE
The smartest thing here might be using less
AI is using a growing share of US electricity.
The International Energy Agency says AI systems and data centres used about 415 terawatt hours of power in 2024, or more than 10% of the country’s electricity production.
With demand expected to double by 2030, concerns are rising over how sustainable current AI systems really are.
Researchers at a School of Engineering have developed a proof-of-concept alternative called neuro-symbolic AI.
The system combines neural networks with symbolic reasoning, allowing it to use both learned patterns and rule-based logic to solve problems more efficiently.
The team is focused on robotics models known as visual-language-action systems, which turn visual input and written instructions into physical actions.
Traditional versions often depend on large amounts of data and repeated trial and error, which can waste energy and still lead to mistakes.
The researchers say symbolic reasoning helps reduce that inefficiency.
By applying rules around things like shape, balance, and order, the system can plan tasks more effectively instead of relying only on statistical guesses.
Here’s what you should know:
AI power use is rising fast, putting more pressure on infrastructure.
The hybrid system outperformed standard models while using far less energy.
Researchers say this approach could support a more sustainable future for AI.
Expensive guesswork
In tests using the Tower of Hanoi puzzle, the neuro-symbolic model achieved a 95% success rate, compared with 34% for standard systems.
It also trained in 34 minutes rather than more than a day and a half, while using just a fraction of the energy.
The findings suggest neuro-symbolic AI could offer a more efficient and reliable path forward as energy demand from AI continues to grow.
Bigger is not always better, babe. Sometimes it is just more expensive.- MG