What Makes Children's Learning Remarkable, and Why Should We Care?
Hande Melis Altunay, Dissertation Plus
What is the most recent thing that you have learned? A word from another language, a new dance, a new opening move in chess or maybe how to take care of a plant. Learning is a process, and as humans, we take part in it almost every day. There is one group of human beings, however, that brings learning to another level at an unbelievably incredible pace: Children.
Any person who has tried to learn a new language or a physical activity like skiing can recall that children pick up on things way faster and way easier than adults do. In fact, I know many, including myself, thinking "I wish my parents would have pushed me harder to learn how to play an instrument when I was a kid" upon realizing how hard learning new things can be. Similarly, every language is difficult to learn as a second language, especially at a mother-tongue level. Conversely, no language is difficult when it is your first language. In fact, "even babies" can learn it.
We know that children are the best learners on earth. One day, they are babies, and by the end of the 4th year, they can talk with you about almost everything. In such a short time, they learn how to crawl, stand, walk, run and even dance. They connect with other people, understand their intentions and their emotions. They acquire language and learn to put their thoughts into words. They dream, create games, play with friends, and even make jokes. How do they learn so much in so little time, with very little instruction? This is one of the fundamental questions of cognitive and developmental psychologists, and they've started to uncover some clues.
Research shows that children's learning process closely resembles that of scientists. Like scientists, children observe, make predictions about the world, conduct experiments through exploration and play, and draw conclusions1. This may sound too sophisticated for a baby, but it simply describes what happens when a baby bumps different toys together to see what kinds of sounds they make.
In one study, researchers showed babies both physically not-surprising events (e.g. a ball rolling and stopping when it hits a wall) and surprising events (e.g. a ball rolling and seemingly passing through a solid wall). They then measured how long babies looked at each scene. The researchers hypothesized that if babies have some early knowledge of the physical world, they should expect that objects cannot pass through a solid wall and would be surprised by the impossible event. This phenomenon, also known as the violation-of-expectation principle, is reflected in longer looking times. Indeed, the study found that babies as young as four months old looked significantly longer at the impossible event2-3.
Another group of researchers took this a step further and asked, "Can babies use these surprising events as learning opportunities?" They showed 11-month-old babies two different scenes—one surprising and one not—each featuring a different object. For example, in one case, a toy car appeared to float in midair after reaching the edge of a platform (a surprising event), while in another, a ball rolled and stopped when it hit a wall (an unsurprising event). Later, researchers gave babies both objects—the toy car and the ball—and observed how they played with them.
They found two key results. First, babies played more with the object that had behaved surprisingly. More intriguingly, they played with it in highly specific ways. If the ball had appeared to pass through a solid wall, babies bumped it against surfaces more often, as if testing its solidity. If the toy car had floated in midair, babies repeatedly dropped it, as if testing whether it would float again.4
This mirrors the routine of a scientist, but in baby steps: Start with some beliefs about how the world works, observe an event, make predictions, test them through experimentation, and update your beliefs accordingly. Scientists don't need to study every single person in the world to draw conclusions about human behavior. Similarly, children don't need endless examples to learn—they infer from just a few. A simple picture book, for instance, can introduce them to the concepts of cats and dogs. Children are the ultimate learning machines.
Speaking of learning machines: Guess who else is trying to learn everything from scratch—and might even be a little envious of babies? Artificial intelligence (AI).
Recent research on machine learning and AI has revealed a simple yet powerful insight: an AI that knows how to explore and learn from available data is far more efficient than one that is pre-programmed with all knowledge5-7. In other words, a baby-minded AI—one that learns flexibly through experience—is more effective than an adult-minded AI that starts with fixed knowledge. However, the way babies and children learn is fundamentally different from how AI operates, posing significant challenges for researchers.
The recent rise of AI is largely due to the enormous amount of data provided by millions of internet users around the world. AI systems rely on immense datasets and clear, well-defined images, whereas babies' experiences are chaotic. Parents rarely provide direct instruction; instead, babies observe, explore, experiment, ask and learn independently, and the data they collect is quite chaotic. Seeing a handful of cats and dogs would be enough for children to differentiate these two animals, whereas AI needs an immense dataset of all sorts of different cat and dog images and definitions and instructions to be trained about this distinction.
Another main ingredient of babies' and children's learning is their insatiable curiosity. Parents often experience this as being asking too many questions and follow-up questions about every possible thing around 3-to-5 years of age. Unlike AI, which passively absorbs data, kids actively choose which question to ask, about what and to whom, which are all crucial for efficient learning.
A final distinguishing factor of children's learning is its deeply social nature. While AI learns through controlled supervision, children selectively imitate those around them, understanding others' goals and intentions. Social interaction also shapes their emotional intelligence, fostering empathy and fairness—qualities that remain major challenges in AI research.
AI may need to embrace these fundamental human traits to achieve true intelligence. Some researchers believe this is possible; others think it’s entirely out of reach. And some, like myself, say—maybe yes, maybe no. But I prefer to ask a different question: Given our limited resources, why invest so much energy in creating artificial learning machines when we already have the original ones—children? If we invested in these powerful learners and their environments—babies, children, caregivers, teachers, and schools—the way we invest in technology, engineering, and design, perhaps we could build the humane future we genuinely need.
There are many other experiments showing how babies and children learn about the world. If you are interested here is a playlist of similar experiments explained by their researchers.
Sources
1) Gopnik, A., Meltzoff, A. N., & Kuhl, P. K. (1999). The scientist in the crib: Minds, brains, and how children learn. William Morrow & Co.
2) Baillargeon, R., Spelke, E. S., & Wasserman, S. (1985). Object permanence in five-month-old infants. Cognition, 20(3), 191-208.
3) Spelke E. S., Breinlinger K., Macomber J., Jacobson K., Origins of knowledge. Psychol. Rev. 99, 605–632 (1992).
4) Stahl, A. E., and Feigenson, L. (2015). Observing the unexpected enhances infants' learning and exploration. Science 348, 91–94.
5) Treger, S., & Ullman, S. (2025). From Infants to AI: Incorporating Infant-like Learning in Models Boosts Efficiency and Generalization in Learning Social Prediction Tasks. arXiv preprint arXiv:2503.03361.
6) Zaadnoordijk, L., Besold, T. R., & Cusack, R. (2022). Lessons from infant learning for unsupervised machine learning. Nature Machine Intelligence, 4(6), 510-520.
7) Zhu, L., Wang, J. Z., Lee, W., & Wyble, B. (2024). Incorporating simulated spatial context information improves the effectiveness of contrastive learning models. Patterns, 5(5).
Artikel kommentieren
Kommentare sind nach einer redaktionellen Prüfung öffentlich sichtbar.