# Curiosity as a Driver for Complex Neural Architecture Artificial neural networks reveal how curiosity shapes brain-like structures and cognitive flexibility. - **Source:** [New Scientist](https://www.newscientist.com/) - **Published:** 2026-11-02 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/c20151ae-ff3d-40f0-a2c4-b46dbd602980 ## Curiosity and Neural Complexity Researchers trained artificial neural networks to explore how curiosity influences cognitive development. By modeling learning styles, they discovered that curious models spontaneously organized into complex, modular structures. These systems mirrored human brain architecture, suggesting that the drive to seek new information is a fundamental catalyst for developing sophisticated and flexible cognitive abilities. ## Infant Learning Data The study utilized eye-tracking data from 275 infants aged 8 to 15 months to model human learning behaviors. Infants were presented with sequences of shapes, demonstrating a clear sensitivity to information density. When learning rates dropped, infants disengaged, providing a behavioral baseline for the intrinsic curiosity parameters used in the neural network simulations. ## Simulating Intrinsic Curiosity Artificial networks were programmed with varying levels of curiosity to perform 20 distinct tasks. High-curiosity models switched tasks when learning rates fell below a specific threshold, while low-curiosity models switched randomly. Both groups mastered their training tasks, but only the curious models demonstrated superior performance on novel, unseen tests after the training phase concluded. ## Structural Brain Modeling Researchers embedded these networks into a virtual 3D space, penalizing long-distance connections to simulate biological wiring costs. This constraint, combined with curiosity, produced hierarchical, modular structures similar to human brains. The curious networks exhibited dense connectivity patterns that evolved over time, reflecting the complex, interconnected nature of biological neural systems observed in nature. ## The Pruning Trajectory Curious networks displayed a developmental trajectory where connection density rose and then fell, mimicking human synaptic pruning. In humans, synapse numbers increase during the first year of life before unhelpful connections are lost. This suggests that curiosity acts as a selection mechanism, determining which neural pathways are retained to optimize future cognitive performance. ## Biological Age Comparison By mapping neural density against human profiles, researchers estimated the biological age of the networks. Curious models reached a mature developmental phase equivalent to 20 years, whereas non-curious models remained stuck at a 9-year-old level. This indicates that curiosity is essential for reaching advanced stages of cognitive maturity within the simulated developmental arc. ## Educational Policy Implications The findings suggest that rigid educational environments may hinder cognitive development by limiting exploration. Francesco Poli argues that forcing children to focus on single activities could be counterproductive. Prioritizing curiosity-driven learning, where children are free to choose activities, may better foster the flexible cognition required for complex problem-solving and long-term intellectual growth. ## Future of AI Development Current AI research focuses on scaling system size, but this study proposes building curious models that learn like humans. While large language models demonstrate the power of scaling, incorporating curiosity could lead to more human-compatible systems. Such models might proactively ask questions to clarify user intent, potentially offering more efficient, resource-conscious alternatives. ## Sources covering this story - https://www.newscientist.com/article/2596932-curiosity-may-be-what-drives-our-brains-to-become-so-complex/?utm_campaign=RSS|NSNS&utm_content=home&utm_medium=RSS&utm_source=NSNS — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._