In a manner akin to the necessity for the human mind to expand and operate within physical and biological confines, researchers from Cambridge have illustrated that imposing natural limitations on an artificial intelligence system enables it to develop features akin to the brains of complex organisms to tackle tasks efficiently.
Neurological systems, such as the brain, face the challenge of balancing various needs while structuring themselves and establishing connections. The network in physical space necessitates energy and resources for growth, maintenance, and optimization for information processing. This trade-off influences the formation of neurons across different species, potentially explaining the convergence of certain brain organizational solutions.
Jascha Achterberg, a Gates Scholar at the University of Cambridge’s Medical Research Council Cognition and Brain Sciences Unit (MRC CBSU), emphasized the brain’s exceptional problem-solving ability with minimal energy consumption. Their recent study delves into how the configuration of the hippocampus can be understood by considering the body’s problem-solving capacity and its inclination to minimize resource usage.
The research team, led by Achterberg and Dr. Danyal Akarca from MRC CBSU, proposed a novel approach using artificial intelligence to simulate a simplified version of the brain by incorporating physical constraints. Through this method, the program gradually developed fundamental traits and strategies reminiscent of human brain functionality over time.
Unlike real neurons, the system utilized computing networks where each node processes input and generates an output, akin to the function of neurons. By imposing a “physical” constraint on the system, wherein distant networks had difficulty communicating, the researchers mirrored the spatial arrangement of cells in the brain.
The system was tasked with a maze navigation challenge, akin to those presented to animals like rats and macaques in cognitive studies, requiring the integration of information to determine the shortest path. As the system learned from feedback and adjusted connection strengths between networks, akin to the dynamic changes in brain tissue during learning, it gradually improved its task performance.
The physical barrier enforced by the researchers led the synthetic system to develop intricate features similar to human brains, such as the creation of highly connected hubs to facilitate data transfer. Nodes within the system exhibited a flexible coding scheme, encoding multiple components of the maze task simultaneously, akin to the hippocampus in complex species.
The team anticipates that their AI model will shed light on how these constraints influence individual brain variations and potentially offer insights into mental health conditions. By simulating artificial neurons, researchers can explore the complexities observed in real brain cell tasks, paving the way for innovative experiments and comparisons.
The implications of this research extend to the development of future AI systems, offering valuable insights for creating more efficient models, particularly in scenarios with physical constraints. By leveraging principles from neurology, AI researchers aim to design neural systems that excel in flexibility and efficiency, akin to natural brain structures optimized for specific tasks.