
We humans learn not only from our own experiences, but also from others. But how do individual and social learning play together – and what role do different environmental conditions and social situations play? Researchers have now examined these questions with the help of the Minecraft video game. It was shown that it is a crucial success factor to switch flexibly between your own exploration and observing others. The results help to better understand human decision -making in complex and dynamic social environments.
We construct skyscrapers, develop computers and explore space. Humanity could only achieve such skills because our species is able to share knowledge and pass on and expand over generations. How exactly we combine social and individually acquired knowledge has so far only been understood. In many studies, it was also left out that different external conditions can require different learning strategies.
Virtual search for food
A team led by Charley Wu from the University of Tübingen has now used the Minecraft video game to test how people combine their own and social learning under different environmental conditions. To this end, the researchers sent their test subjects in four groups in the virtual environment of the computer game. The landscape consisted of countless identical looking blocks that the participants had to smash. Some of these blocks contained resources such as pumpkins and watermelons, others were empty. Whenever a player found a resource, a blue spark of sparks appeared, which showed the fellow players the site.
The highlight: In some runs of the game, virtual food was arranged in clusters, that is, in addition to a filled block, there were most likely further. In this setting, the social evidence was of great value. In other runs, on the other hand, the resources were randomly distributed, so it was more promising to go on an exploration tour on your own. In which of the two environments the test subjects were in, however, they were at most able to find out by attempt and errors.
Flexible adaptation
“It makes sense to use a game like Minecraft because it simulates real challenges. For example, you can only see a small part of the virtual world and therefore have to decide whether to focus on your own search or pay attention to what the other players do to learn from them,” explains co-author Ralf Kurvers from Max Planck Institute for Education Research. “That means that I am constantly faced with the choice: do I follow my intuition and search alone or do I use social information by episoding the game that has already found something?”
The experiments showed that those test subjects were most successful who flexibly adapted their strategy to the respective environment. So if you quickly realized that the resources often hide in neighboring blocks, follow the blue sparks that were triggered by others and thus maximize your own win in the runs with food clusters. However, anyone who followed this strategy was less successful in runs with random distribution. So it was primarily a matter of adapting to the different conditions and deciding whether the focus should be on your own exploration or on observing the teammates.
(Video: Benjamin Kahl)
Predictions with artificial intelligence
In order to better track the test subjects’ decision -making processes, the researchers automatically recorded their eyes. In this way, they were able to link the actions in the game with which objects, events and teammates looked at the participants. On this basis, WU and his team developed a model that could only use the view to predict which block a test will select next. “This new approach enables us to combine learning algorithms of modern artificial intelligence with flexible social learning mechanisms that learn adaptively from the successful behavior of others,” says WU.
As the experiments showed, the driving factor for the change between the strategies was your own success. “These results not only integrate theories about social and non-social adaptive mechanisms, but also provide important findings about the adaptability of human decision-making in complex and dynamic social landscapes,” summarizes the research team.
Source: Charley Wu (University of Tübingen) et al., Nature Communications, DOI: 10.1038/S41467-025-58365-6