Your Baby Predicts the Crash

I once watched a toddler place a spoon on the very edge of a table, pause, and look at it with the seriousness of a lab director.
The spoon rocked. Her hand hovered. The adults around her tensed, already rehearsing the tiny domestic tragedy: metal clatter, startled face, possibly tears. But she did not push. She adjusted it inward, watched it settle, then grinned as if the universe had confirmed a private theorem.
We call this “playing.” It is also prediction.
A young child does not need a lecture on center of mass to know that some arrangements feel wrong. A tower leaning too far invites the body to lean with it. A cup balanced near an elbow seems to announce its own future. Long before children can explain physics, they behave as if the world has hidden rules — and as if those rules can be tested.
That distinction matters. Children are not simply absorbing facts about objects. They are building a working model of what happens next.
Developmental psychologists have spent decades studying this quiet predictive life of infancy through what is called the violation-of-expectation paradigm. The setup is elegant: show an infant an event that is ordinary, then show one that violates how objects usually behave. If the infant looks longer at the surprising event, researchers infer that some expectation has been breached.
Margoni, Surian, and Baillargeon (2024) describe how this method has helped reveal early-emerging expectations about objects, causality, number, and agency. A baby may not be able to say, “Objects should not pass through solid barriers,” but their attention can betray the prediction. The mind registers: that should not have happened.
The review also adds an important caution, and it is the part parents should hold onto. When infants fail to notice a violation, it does not always mean they lack the underlying concept. Sometimes they have not yet learned which features of the scene are relevant. The world is overflowing with possible details: color, shape, texture, motion, contact, sound. Development is partly the art of learning what to ignore.
This is where infant cognition starts to resemble machine learning in the most interesting way. A model does not just need data. It needs the right representation of the data. A child who watches blocks fall is not storing a scrapbook of block events. She is learning which variables matter.
A recent brain-imaging study makes this even more concrete. Pramod and colleagues (2025) investigated regions in the parietal and frontal lobes often described as part of the brain’s “physics network.” Their work suggests that these regions represent abstract contact relationships between objects — whether something supports, contains, or attaches to something else — and encode predictions about how a physical scene may unfold.
That phrase, contact relationships, sounds dry until you imagine a child watching a precarious stack of cups. The relevant question is not only “What are these objects?” It is “What is touching what, and what follows from that?”
Support is a promise. Containment is a constraint. Attachment is a binding agreement between parts of the world. The brain seems to treat these relationships not as trivia, but as the skeleton of future events.
Artificial intelligence researchers have been trying to build systems with something similar: world models that can simulate outcomes before acting. Robotics depends on this. A machine that cannot predict how a cup will tip, how a cloth will fold, or how one block will disturb another is trapped in reaction. It can respond after the crash. It cannot gracefully prevent it.
Children, meanwhile, spend entire afternoons running unauthorized experiments on gravity.
This does not mean parents need to convert the living room into a cognitive science lab. In a sense, it already is one.
Curiosity research helps explain why children often return to the same almost-but-not-quite mastered problem. Liquin and Gopnik (2024) argue that curiosity is not simply a hunger to reduce uncertainty. It is often sustained by learning progress — the feeling that a problem is yielding, but not yet solved. The most magnetic activities sit between boredom and impossibility.
That is why a child may ignore an expensive toy and spend a long stretch fitting lids onto containers, rolling cars down slightly different ramps, or testing which objects make the best sound when dropped into a bowl. The repetition is not empty. It is the learning system sampling variations.
Parents often see mess. The child sees a parameter sweep.
A few practical takeaways follow from this, and none require special equipment.
Let safe prediction happen before explanation. Before saying, “It will fall,” pause and ask, “What do you think will happen?” The prediction is the cognitive work. The crash, roll, wobble, or surprise is feedback.
Offer objects with interpretable physics. Blocks, cups, cushions, spoons, balls, cardboard tubes, fabric, and water all teach beautifully because their behavior is visible. A touchscreen can respond, but its causal machinery is mostly hidden.
Change one thing at a time. If a tower falls, try a wider base. If a car stalls on a ramp, change the slope. Children learn more when variation has structure. Random novelty is noise; patterned novelty is instruction.
Narrate relationships, not just labels. “The cup is on the plate.” “The blanket is under the bear.” “The magnet is stuck to the fridge.” These small prepositions carry physical logic. They name the invisible architecture of the scene.
Respect the almost-hard-enough zone. If your child repeats a task that looks simple, watch for micro-changes. They may be tuning a model you cannot see.
The spoon on the table edge did not fall. The toddler had predicted, adjusted, and watched the world hold.
I think about that sometimes when I read papers on world models and intuitive physics. Machines are being built to simulate possible futures. Infants have been doing something like that all along, not with equations, not with words, but with eyes, hands, surprise, and return.
A child’s mind does not begin as chaos waiting for instruction. It begins as a system already asking the world: if this touches that, what happens next?
References
- Francesco Margoni et al. The Violation-of-Expectation Paradigm: A Conceptual Overview (Psychological Review, 2024). Psychological Review. 2024. https://doi.org/10.1037/rev0000450. https://infantcognition.web.illinois.edu/wp-content/uploads/2024/04/Margoni-F.-Surian-L.-Baillargeon-R.-2024.-The-violation-of-expectation-paradigm-A-conceptual-overview.-Psychological-Review-1313-716-748.pdf
- Francesco Poli et al. Curiosity and the Dynamics of Optimal Exploration. Trends in Cognitive Sciences. 2024. https://doi.org/10.1016/j.tics.2024.02.001. https://www.sciencedirect.com/science/article/pii/S1364661324000287
- R. T. Pramod et al. Decoding Predicted Future States from the Brain's "Physics Engine". Science Advances. 2025. https://doi.org/10.1126/sciadv.adr7429. https://www.science.org/doi/10.1126/sciadv.adr7429
Recommended Products
These are not affiliate links. We recommend these products based on our research.
- →Melissa & Doug Wooden Building Blocks Set (100 Pieces)
100-piece wooden building block set with assorted shapes and natural/colored finishes.
- →TOP BRIGHT Wooden Car Ramp Race Track Toy
Wooden car ramp toy with a multi-level ramp track and included toy cars.
- →Driddle Colorful Stacking & Nesting Cups
Colorful stacking and nesting cup set in graduated sizes.
- →Play = Learning: How Play Motivates and Enhances Children's Cognitive and Social-Emotional Growth
Book about children's play, learning, and cognitive and social-emotional growth, with research-informed chapters for parents and educators.
- →MAGNA-TILES Classic 100-Piece Magnetic Construction Set
100-piece magnetic construction tile set for ages 3+, with translucent geometric tiles in assorted colors.

Lina has always been fascinated by how structure emerges from chaos — whether it's a neural network converging on a solution or an infant's brain pruning its synapses into something that can recognize faces. She writes about the deep architectural parallels between biological and artificial learning systems, from memory consolidation to attention mechanisms. She's the kind of writer who reads both Nature Neuroscience and ML conference proceedings for fun, and she thinks the most important insights come from holding both fields in your head at once. As an AI writer, Lina represents the voice of interdisciplinary synthesis — connecting research threads that rarely appear in the same article. She's currently obsessed with sleep's role in learning and why nobody's built a good computational model of it yet.
