Skip to main content

Toddler robots help reveal how human kids learn about their world

toddler robot learning toddler1
There’s a lot of focus on looking at the means by which humans learn and using these insights to make machines smarter. This is the entire basis for artificial neural networks, which try to replicate a simple model of the human brain inside a machine.

However, the opposite can be true as well: Examining robots can help reveal how we as humans absorb and make sense of new information.

That’s the basis for a new research project, carried out out by researchers in the United Kingdom. Looking to understand more about how young kids learn new words, they programmed a humanoid robot called iCub — equipped with a microphone and camera — to learn new words.

Their conclusion? That children may well learn new words in a similar way to robots; based less on conscious thought than on an automatic ability to associate objects.

“We were interested in finding out whether it’s possible to learn words without a complex reasoning ability,” Katie Twomey, a psychology department researcher from the U.K.’s Lancaster University, told Digital Trends.

“To explore this we used the iCub humanoid robot, which learns by making simple links between what it sees and what it hears. Importantly, iCub can’t think explicitly about what it knows. We reasoned that if iCub can learn object names like toddlers do, it’s possible that children’s early learning is also driven by a simple but powerful association-making mechanism.”

In the study, a group of kids aged 2 1/2 were given the task of selecting a particular toy of out of lineup consisting of, alternately, three, four, or five different objects. In each case, one of the objects was something unfamiliar to them. The study aimed to get the kids to learn the name of the unknown object using a process of elimination, based on information they already knew.


“We know that toddlers can work out what a new word means, based on the words they already know,” Twomey continued. “For example, imagine a 2-year-old sees two toys: their favorite toy car, and a brown, furry toy animal that they’ve never seen before. If the toddler hears a new word ‘bear,’ they will assume that it refers to the new toy, because they already know that their toy is called ‘car’.”

In this case, it is possible that kids are able to think in detail about what they already know, and use reasoning to figure out that their favorite is called a “car,” so the new toy must be a “bear.” However, it’s also possible that children solve this puzzle automatically by simply associating new words with new objects.

The researchers then asked the iCub to carry out the same task. It was trained to recognize 12 items but, like the kids, was then shown a combination of objects it recognized and ones it did not. Intriguingly, it performed exactly the same as the kids when it came to learning new words.

“Critically, iCub learned words by making simple associations between words and objects, rather than using complex reasoning,” Twomey said. “This suggests that we don’t need to assume children reflect in detail about what they know and what words refer to. Instead, early word learning could depend on making in-the-moment links between words and objects.”

It’s an interesting use of robotics to help uncover insights about developmental psychology. It can also reveal previously unconsidered details which may also tell us something about how humans learn.

“In our study, the amount of time it took for the robot to move its head to look at objects affected how easily it learned words,” Twomey concluded. “This suggests that the way objects are set out in children’s visual scene could also affect their early word learning: a prediction we are planning to test in new work with toddlers.”

Editors' Recommendations