What makes successful learners? How Minecraft can help us understand social learning
2025-04-25T09:00:02Z
The ability to learn socially from one another is a defining feature of the human species. Social learning enables humans to gradually accumulate information across generations. And although we are able to build cities full of skyscrapers, send people into sp…
This article has been reviewed according to Science X’s editorial process and policies . Editors have highlighted the following attributes while ensuring the content’s credibility:
A person interacts with a custom-built Minecraft environment used to study adaptive social learning in a virtual foraging experiment. This study, published in Nature Communications, was designed to investigate how humans adapt their decision-making through both social and asocial learning strategies. Credit: Charley Wu
The ability to learn socially from one another is a defining feature of the human species. Social learning enables humans to gradually accumulate information across generations. And although we are able to build cities full of skyscrapers, send people into space, and collectively develop cures for diseases, most studies investigating social learning mechanisms focus on relatively simple, abstract tasks that bear little resemblance to real-world social learning environments.
As a result, little is known about how humans dynamically integrate asocial and social information in realistic, real-world contexts. To investigate this, an international team of scientists from the Cluster of Excellence Science of Intelligence (SCIoI), the Max Planck Institute for Human Development (MPIB), the University of Tübingen, and NYU developed a virtual foraging task programmed in the popular video game Minecraft.
In their study, published in Nature Communications, they found that adaptability (i.e. flexibly using asocial and social learning strategies, rather than fixed strategies) is the most important driver of success.
‘Should I explore on my own or work with the group?’
In the experiment, each participant controls an avatar that destroys Minecraft blocks in order to find resources. Whenever a resource is discovered, a blue splash appears, which is visible to other players, and could potentially provide useful social information about the location of further resources.
At the beginning of each round, the players are informed of whether they will be working alone or in a group of four people who can interact with each other in real-time. Additionally, they are tested across two types of environments.
In “patchy” environments, resources are clustered together, which means that participants can find numerous blocks containing resources close to each other, while in “random” environments resources are spread out.
Thus, social information is particularly valuable in “patchy” environments, since it may reveal other rewards nearby. However, social information has no value in “random” environments, since there is no learnable pattern of resource locations. Each player tries to maximize their own rewards, rather than working towards a collective goal, and thus needs to effectively find rewards using the right balance of individual and social learning strategies.
“Using a game like Minecraft is useful because it simulates real-life challenges. For instance, since you can only see a small part of the game world at a time, you must choose whether to focus on searching on your own or pay attention to what the other players are doing to learn from them,” said Ralf Kurvers, the senior author of the study.
“This means that I am constantly faced with a choice: do I follow my own instinct and go search alone, or do I utilize social information (in this case, the blue ‘splashes’) by following the players who’ve already found something, as they are likely to have found a resource patch?”
New tools for studying the interaction between individual and social learning
Through a newly developed computational method for automating the transcription of visual field data, the scientists measured which objects, events, and other players were observed by each participant, recorded at a rate of 20 times per second. They created a model that brings together where people look, how they move, and the choices they make when foraging.
“In simpler terms, we can now predict which block a participant will choose next by combining individual and social learning strategies, all in one computational framework,” explained Charley Wu from the University of Tübingen.
“This new approach allows us to connect the learning algorithms that power modern AI with flexible social learning mechanisms that adaptively learn from the successful behaviors of others.”
Altogether, the study bridges a decades-long gap between research on individual and social learning. The results show that humans are not just passive imitators or stubborn individual learners. Rather, they dynamically balance these strategies; adaptive mechanisms of individual and social learning amplify one another, and are driven by a common currency of individual performance.
Furthermore, the extent to which each individual was able to adapt their individual and social learning strategies was the best predictor of their performance. This emphasizes that adaptability, rather than fixed strategies, is what drives human intelligence.
This work advances our understanding of the cognitive mechanisms underlying adaptive learning and decision-making in social contexts, opening new pathways for understanding how information spreads in groups, how new innovations emerge, and gives clues on how to design systems that better foster adaptive learning in social settings.
More information: Adaptive mechanisms of social and asocial learning in immersive collective foraging, Nature Communications (2025). On bioRxiv: DOI: 10.1101/2023.06.28.546887 Journal information: Nature Communications , bioRxiv
Auto-posted from news source
About The Author
