Babies as young as two to three months can learn that kicking one leg makes a mobile above their crib move — and a new computational model now explains how.
The research
Josua Spisak, Sergiu Tcaci Popescu, Stefan Wermter, Matej Hoffmann, and J. Kevin O'Regan built a computer simulation of infant behavior in the classic "mobile paradigm." Their study appeared on arXiv (arXiv:2504.17939, submitted April 24, 2025; revised July 2026) and was accepted in IEEE Transactions on Cognitive and Developmental Systems, volume 18, issue 3, pages 692–707.
In the mobile paradigm, researchers tie a ribbon from an infant's ankle or wrist to an overhead mobile. If the infant kicks or moves that limb, the mobile jiggles — a direct sensorimotor contingency. Across decades of developmental psychology, infants show a consistent pattern: they move the connected limb more than the others, seemingly learning that their action produces a sensory effect.
The new model combines a neural network, action-outcome prediction, an exploration drive, motor noise, a preferred activity level, and biologically inspired motor control. Simulations reproduced the classic finding of preferential movement of the connected limb. Curiously, the model also produced a burst of movement after the mobile was disconnected — mirroring an occasional real-infant result. The model further matched two recent detailed studies where the mobile connection was either gradually introduced or all-or-none.
A series of ablation studies — removing each component one at a time — showed that action-outcome prediction, exploration, motor noise, and biologically inspired motor control were each essential. Without them, the model failed to match infant behavior.
Why it matters for your brain
Sensorimotor contingency detection — noticing that "when I do X, Y happens" — is one of the foundations of cognitive development. It underpins goal-directed movement, imitation, and eventually tool use and language. The fact that a handful of simple mechanisms (prediction, exploration, noise, and realistic motor control) can reproduce infant learning suggests these same ingredients operate in adult brains whenever we learn a new physical skill, from typing to playing an instrument.
Motor noise, often treated as an obstacle, turns out to be useful here: it drives variability, which fuels exploration. Exploration, in turn, generates the data needed to learn the action-outcome link. That interplay — noise → exploration → prediction → learning — is a general principle of adaptive cognition.
What you can do
- When learning a new motor skill, build in small variations rather than repeating the exact same motion. Slight variability helps your brain map cause and effect.
- Pay attention to the sensory consequences of your actions — the sound, sight, or feel — not just the movement itself. Prediction accuracy improves with feedback.
- After learning a contingency (e.g., a key press opens a menu), expect a brief burst of exploratory actions when the contingency stops. That's normal and may help you update the rule.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.