Scientists at the University of Cambridge, inspired by the structure and function of the human brain, imposed physical constraints on an artificial intelligence system. Led by Jascha Achterberg and Danyal Akarca from the Medical Research Council Cognition and Brain Sciences Unit (MRC CBSU), the team aimed to replicate the developmental constraints faced by biological brains in their AI model.
COMPUTATIONAL NODES AND VIRTUAL CONSTRAINTS
Departing from real neurons, the researchers utilized computational nodes mirroring the functions of biological neurons. The physical constraint imposed involved giving each node a specific location in a virtual space, creating a spatial hurdle for communication—the farther nodes were, the more challenging communication became. The team then tasked the system with a maze navigation challenge, a simplified version reminiscent of tasks given to animals like rats and monkeys.
AI SYSTEM DEMONSTRATES CHARACTERISTICS OBSERVED IN HUMAN BRAINS
Published in Nature Machine Intelligence, the study reported that the AI system, responding to physical constraints, developed hubs—nodes with high connectivity facilitating information flow. This adaptive behaviour mirrors strategies employed by real human brains. The study emphasizes the importance of considering both structure and function in computational models, shedding light on optimization principles governing the brain’s organization, functionality, and energy efficiency.
ADAPTIVE BEHAVIORS BEYOND SPECIALIZED FUNCTIONS
Remarkably, the individual nodes within the AI system exhibited a “flexible coding scheme.” Unlike traditional systems where nodes specialize in specific maze properties, these nodes showcased adaptability, encoding various maze aspects at different moments. This behaviour, reminiscent of complex animal brains, introduces a new dimension to the adaptability and problem-solving capabilities of AI systems inspired by natural constraints.
This groundbreaking study opens avenues for developing AI systems that not only mimic the complexity of the human brain but also adapt and find effective solutions within specified constraints, offering a promising glimpse into the future of artificial intelligence.