Bernstein member involved: Fred Hamker

On January 28, 2026, a kick-off meeting launched a pilot project on the topic of “Brain-inspired use of efficient shortcuts in artificial intelligence” at Chemnitz University of Technology. Over the next three years, researchers from the Chair of Artificial Intelligence (headed by Prof. Dr. Fred Hamker) at Chemnitz University of Technology and the Chair of Neuropsychology (headed by Prof. Dr. Markus Ullsperger) at Otto von Guericke University Magdeburg will work together to develop a solution for increasing computing and energy efficiency in large modular neural transformer networks. This solution aims to draw inspiration from the brain’s remarkable ability to learn routines.

Background: The development of habitual behavior in both animal and human brains reflects the nervous system’s ability to efficiently delegate cognitive and neural resources. By automating frequently repeated responses, the brain minimizes the need for cognitive effort and decision-making. This allows individuals to focus on other complex tasks. “Just as the human brain automates frequently repeated responses to free up cognitive capacity, AI systems can benefit from habit-like mechanisms to optimize processing efficiency,” says Hamker. By learning and automating frequently used decision patterns, AI can minimize redundant calculations, reduce the computational load, and shorten response times. “This not only improves overall efficiency, but also contributes to energy savings, as neural networks and reinforcement learning models require considerable computing power,” says the Chemnitz-based AI expert.

From the researchers’ point of view, future AI systems could therefore be significantly improved if they enabled both detailed model-based learning and automated habit learning. The team from Chemnitz and Magdeburg wants to translate the shortcut connections that exist in the brain into AI. “We assume that these shortcut-like concepts are advantageous when calculating routine tasks, as they require significantly less computing power and thus reduce energy consumption considerably, while at the same time maintaining the flexibility of AI systems,” says Hamker. The brain-inspired AI model will be compared with existing AI methods in terms of performance and energy consumption, as well as with the cognitive flexibility of humans.

The interdisciplinary project will receive approximately € 365,000 in funding by the German Federal Ministry of Research, Technology, and Space (BMFTR) until December 2028 as part of the “Neurobiologically Inspired Artificial Intelligence” call. The project results are intended to lay the foundation for a novel, computationally and energy-efficient AI system, suitable for learning automation and solving complex tasks in a wide range of application areas.

Translated from German by Elena Reiriz Martínez/BCOS