Brain-inspired AI cuts energy use by 90% while boosting performance

University of Surrey researchers have developed a brain-inspired approach to artificial intelligence that dramatically improves performance while cutting energy consumption. The breakthrough, called Topographical Sparse Mapping, mimics how the human brain organizes neural connections, offering a more sustainable path forward as AI models continue growing in size and energy demands.

How it works: The new method connects each artificial neuron only to nearby or related neurons, mirroring the brain’s efficient information organization rather than creating vast networks of unnecessary connections.

• Researchers published their findings in Neurocomputing, demonstrating that this approach significantly improves performance in generative AI and modern models like ChatGPT.
• An enhanced version introduces a biologically inspired “pruning” process during training, similar to how the brain refines its neural connections as it learns.
• The technique eliminates redundant connections while maintaining accuracy, creating more efficient AI systems.

Why this matters: Current AI training methods consume enormous amounts of energy, making sustainability a critical concern as the technology scales.

• “Training many of today’s popular large AI models can consume over a million kilowatt-hours of electricity,” said Dr. Roman Bauer, senior lecturer at the University of Surrey.
• “That simply isn’t sustainable at the rate AI continues to grow.”

The big picture: This brain-mimicking approach could revolutionize how AI systems are built, moving away from brute-force computational methods toward more elegant, biologically inspired solutions.

• The research team is exploring applications in neuromorphic computers, a computing approach directly inspired by the human brain’s structure and function.
• Dr. Bauer emphasized the broader implications: “Our work shows that intelligent systems can be built far more efficiently, cutting energy demands without sacrificing performance.”

What’s next: Researchers are investigating how Topographical Sparse Mapping could be applied to other AI applications and more realistic neuromorphic computing systems.