A study reveals that our moments of mental wandering could prepare the brain to learn more effectively. Far from being wasted time, this passive exploration would build internal models of the world.
This discovery, published in Nature, emerges from observing mice navigating virtual environments. Researchers at the Janelia Research Campus recorded the activity of tens of thousands of neurons, revealing a previously unknown mechanism: the visual cortex encodes visual patterns without a specific goal, thereby optimizing future learning.
Exploration, fertile ground for learning
The experiment conducted by the researchers compared two groups of mice moving through a virtual corridor with multiple textures. The first group received rewards (water) by associating certain textures with positive stimuli, while the second explored the same environment without any goal or reward.
Using an advanced imaging technique, the activity of nearly 90,000 neurons was recorded simultaneously. The results show that mice in both groups developed similar neural plasticity in their medial visual cortex, a key region for pattern processing.
The surprise came from the subsequent learning speed: mice that simply explored the environment without rewards learned faster to associate textures with rewards in a follow-up task. This suggests that passive exposure prepares the brain to efficiently encode new information, even without immediate motivation.
Two complementary systems
The study distinguishes two mechanisms: unsupervised learning (automatic, via exposure) and supervised learning (goal-driven). The first creates a neural database, while the second associates meanings with it.
Researchers used a two-photon excitation microscope to track neuron activity simultaneously. The data show that medial visual areas activate during passive exposure, while frontal zones only respond to reward-based tasks.
This duality could explain why some learning, such as image recognition or spatial navigation, seems intuitive. The brain would capitalize on pre-established models during exploration, accelerating later training phases.
Article author: Cédric DEPOND