AI model replicates infant learning in mobile paradigm experiment
Neural network with action-outcome prediction mimics how babies discover limb control.
A team of researchers built a computational model that replicates infant behavior in the classic "mobile paradigm," where a mobile connected to an infant's limb prompts preferential movement of that limb. The model uses a neural network with action-outcome prediction, exploration, motor noise, preferred activity level, and biologically inspired motor control. Simulations matched classic findings, including an occasional burst of movement after disconnection, and replicated data from two recent studies with gradual or all-or-none connections. Ablation studies showed that action-outcome prediction, exploration, motor noise, and biologically inspired motor control were essential for replicating infant behavior.
- Model uses neural network with action-outcome prediction and motor noise to replicate infant limb preferences.
- Simulations matched classic mobile paradigm results and explained the post-disconnection movement burst.
- Ablation studies showed exploration and prediction mechanisms were essential for accurate behavior.
Why It Matters
Bridges developmental psychology and AI, offering testable computational principles for how intelligent systems learn from interaction.