Professor Hyowon Gweon on ‘Curious, Cooperative and Communicative: How We Learn From Others and Help Others Learn’
By Karla Kane, journalist and former Bing parent
A child watches an adult play with blocks. When she’s given some similar toys to try, she soon realizes hers are not behaving as expected. “It’s not sticking,” she says, noting correctly that her blocks lack the adhesive property of the ones she saw the adult use. After communicating this, and trying some problem-solving, she quickly pivots to using the toys in a different way alongside her brother.
“All of this is happening in parallel, in less than a minute,” said Stanford University associate professor of psychology Hyowon Gweon, presenting video footage of the study described above at the 2022 Bing Nursery School Distinguished Lecture, held on May 12.
What makes human learning so “distinctive, powerful and smart?” That’s the question driving Gweon’s work. And a crucial component of human learning, her research shows, is connection with other humans.
“If part of the answer is that we learn from other people, what are the basic cognitive capacities that we share as humans to support these processes?” she asked at the lecture.
Gweon, who heads Stanford’s Social Learning Lab, said humans can interpret data generated in a social context from a young age, in both formal and casual settings. They’re also able and motivated to communicate their knowledge about the world—and the self—to others.
In her presentation, titled “Curious, cooperative and communicative: How we learn from others and help others learn,” Gweon offered examples (including some from studies run at Bing) of the sophisticated ways in which children draw inferences from observing others and decide how to share information with those around them.
Children are not “lone scientists” learning about the world through isolated first-hand experience, she said. Nor are they merely “little imitators,” blindly copying what they see others do. Instead, they’re constantly absorbing, adjusting to and analyzing data generated both by themselves and their fellow humans.
While anyone who has spent time at Bing likely already recognizes children as active learners and teachers, Gweon said that advances in computational cognitive science have allowed researchers to develop probabilistic models of learning, among other laboratory methods, that empirically demonstrate this. “I feel like parents and teachers know this already,” Gweon said, “But it’s one thing to note this intuitively and another thing to study these behaviors as a discipline of scientific inquiry.”
Children’s awareness of someone else’s intentions can influence their behavior. In one study, when researchers presented themselves as “experts” on a toy and demonstrated a noise-making feature, those children left to explore on their own were more likely to focus solely on replicating that feature. When researchers did not present themselves as experts, however, children were more inclined to explore broadly and discover other features of the toy. Their understanding of the researcher’s intentions and knowledge—pedagogy vs. accidental discovery—influenced their actions.
Sensitivity to the intentions of others and the ability to draw inferences from observed data emerge early, said Gweon. She described a study in which children saw a researcher casually pull three blue toys, which squeaked, from a box of assorted yellow and blue toys. The children expected that the researcher pulled them out randomly and that therefore all the toys, of either color, might squeak. When the researcher pulled three blue toys out of a box of mostly yellow, though, children expected that the researcher must have pulled the blue toys out intentionally, and that the yellow toys would not squeak. “They're able to draw different inferences depending on the context,” she said.
Gweon and her colleagues are also interested in children’s abilities as teachers, and their capacity to generalize data that would be helpful for others. “As soon as our children are starting to learn, they are also thinking about how they can help others learn,” she said.
She described a study in which children were shown a toy bearing numerous buttons, only three of which, when pressed, would result in music playing. When children were asked to teach someone else how to use the toy, researchers wondered whether they would advise them to painstakingly try all the buttons or to just show the three functioning ones. As it turns out, the research suggests that children as young as 5 can determine how much information is necessary and adjust their lesson accordingly. If a learner is already familiar with similar toys, the child may save time and energy by doing selective teaching. If someone is completely unfamiliar with the toy, exhaustive teaching would be more appropriate. A helpful teacher, Gweon said, will prioritize sharing information that will save students trouble and cost and maximize their reward. “Humans,” she said, including young children, “have the ability to reason about others’ utilities.”
To explore this, another study presented children with two toys. One, a simple box with one button that activated music, researchers classified as low-cost, low-reward. The second, a more complicated box featuring multiple buttons that activated a very pleasing light-up-and-spin feature, researchers classified as high-cost, high-reward. Once children understood how to use both toys, they were asked to decide which toy they wished to teach to a friend with no prior experience. The friend would explore the other toy without instruction.
Researchers developed a computational model that predicts children’s responses based on the learner’s expected utilities—in other words, the expected rewards attained by the learner minus the costs incurred—and compared the model’s predictions against children’s responses. Results from the study were consistent with what the model predicted: Children more often opted to teach the difficult toy with the more exciting payoff.
“Children are readily able to think about the cost and the rewards of others’ learning,” Gweon said. This has implications for how to best pass on cultural knowledge, she said, because to be most effective, teachers and learners must strike a balance between what learners can discover on their own and what would be best transmitted by a teacher. “Utility-based reasoning can be a basis for accumulation and curation of useful knowledge over time,” she said.
Emotional responses, including facial expressions and vocal cues, are also key to learning.
Gweon played video footage from a study run at Bing in which an experimenter showed children a box with an obvious button at the top that lit up the toy when pressed. The researcher then turned the toy away from the child and gasped in delighted surprise at something (unseen by the child) that the toy did, exclaiming, “That’s so cool!” When left to explore the toy, the child was curious as to what function of the toy elicited the excited reaction from the researcher and was motivated to discover it himself (“What’s so cool?”).
This is an example of what Gweon called the “common ground condition,” because both child and researcher had a presumed shared knowledge base about the object. Under the “no common ground” condition, an adult with what appeared to the children in the study to be no prior experience with the toy came and expressed surprise after the children had already tried it. In that case, the children assumed that this adult had simply discovered the same obvious function that they themselves were already familiar with. When left to play, they therefore had no reason to expect any “hidden function” and simply activated the button they already knew about.
Sensitivity to emotional cues as sources of information is apparent in younger children, too. A study of 13- to 18-month-olds involved an experimenter pulling a ball out of a box filled with mostly red balls and just a few white. The assumption was that if pulled at random, a red ball would be more likely to be extracted. Knowing from prior studies that infants tend to look at unexpected outcomes longer, the experimenter first peeked at the ball before showing it to the child, in some cases showing surprise and sometimes a neutral face. Researchers wondered whether they could cause a reversal of pattern of looking time depending on the emotional reaction of the experimenter. And indeed, they could. The babies used the emotional response to predict an outcome. “It shows that not just older children but even young infants are able to use other people’s emotional expressions … as useful information to make a prediction about what they couldn’t see before,” Gweon said.
Children can also be motivated by the desire to manage what others think about them. In a study on this, children were shown a red toy and a green toy by a researcher. They were informed that another researcher, Anne, would watch them play. The first researcher activated a music feature on the toy but, by design, the children were unable to do so on their first two tries, with Anne watching. After Anne left, they were finally able to succeed. Gweon referred to this as the “absent condition,” with the critical aspect being that the children believed Anne left the room thinking that they were incapable of activating the red toy. The children were then given the chance to try the green toy and had the same experience: two failures followed by success, the only difference being Anne was not present to observe. With another group of children, the “present condition” group, everything was the same except the timing of Anne’s departure, with Anne leaving the room knowing the children could activate the red toy. The children were then asked which toy they would like to show Anne upon her return. For those in the absent condition group, there was a higher tendency to pick the red toy so that they could show off to Anne and prove that they were able to succeed with it after all.
“What children are trying to do here is try to figure out what information can really change what Anne thinks of them,” Gweon said, adding that feedback from others is an important factor in how humans learn about themselves. She and her colleagues are interested in how humans analyze that feedback. Input from one source might have a different meaning or value than from another. That’s what another study at Bing explored.
Children were asked to create two tracings, which were then concealed in two separate envelopes. The children were shown a video in which a “contingent” or “selective praise” teacher looks at a series of children’s tracings—some objectively “better” than others—and deems a few “really good,” earning stickers. The rest she deems “OK” and does not award stickers. A second teacher, on the other hand, overpraises the artwork, awarding a sticker to everything. The researchers then told the children they were showing their work to the same two teachers and that they could bring back their “best” tracing back to show their friends.
Children showed a preference for choosing the one deemed “good” by the selective-praise teacher (They also identified the overpraise teacher as the one trying to be the “nicest.”). In cases when the selective-praise teacher rewarded the objectively “bad” tracings, though, the children made different choices, proving they weren’t merely tracking the frequency of praise. “They’re able to distinguish informative versus uninformative feedback,” Gweon said. “They’re learning from others about what kinds of feedback they provide, and they use it in order to learn about the self.”
The motivation to help others also emerges early. Ongoing studies are looking at how children use their intelligence to help others learn, including a parent-child study at the Palo Alto Junior Museum and Zoo in which toddlers are able to help a parent who is struggling with a toy.
All these examples from Gweon and her colleagues’ research aim to show that the nature of human learning, even very early in life, is rooted in social context. “The power of human intelligence,” she said, “comes from our ability to figure out how to use our intelligence for ourselves and for others.”
Professor Hyowon Gweon
Hyowon Gweon, PhD, is an associate professor in the Department of Psychology at Stanford University and currently serves as the Director of Graduate Studies for the Department of Psychology and the Symbolic Systems Program. She received her PhD in cognitive science from MIT, where she continued as a postdoc before joining Stanford in 2014. Awards and honors include: CDS Steve Reznick Early Career Award, APS Janet Spence Award for Transformative Early Career Contributions, Jacobs Early Career Fellowship, James S. McDonnell Scholar Award for Human Cognition, APA Dissertation Award, and Marr Prize (best student paper, Cognitive Science Society). Gweon has been named as a Richard E. Guggenheim Faculty Scholar.
About the Author
Karla Kane is an award-winning local journalist; former preschool teacher; and the singer, songwriter and ukulele player for the band The Corner Laughers. She has a bachelor’s and a master’s degree in anthropology and lives in Redwood City with her husband and bandmate Khoi, her cats and her daughter, Octavia (a proud Bing alumna).