A new study reveals that personalization algorithms can influence how people learn new information. The research found participants who started from zero tended to browse less when the clues or information they saw were curated by the algorithm. As a result, they absorb a narrowed and often incorrect picture, but still feel very confident about the conclusions.
Research conducted on content recommendation platforms, such as YouTube, shows that when algorithms determine the information that appears in a learning process, participants focus more on a small part of the material. The lack of exploration causes them to answer questions incorrectly in follow-up tests, even though they still express high self-confidence.
Giwon Bahg, a researcher who conducted this study in his dissertation at Ohio State University, said these findings raised concerns. “But our research shows that even when you don’t know anything about a topic, these algorithms can immediately start building biases and can lead to a distorted view of reality,” said Bahg, who is now a postdoctoral researcher at Pennsylvania State University.
The results of the study were published in the Journal of Experimental Psychology: General. Brandon Turner, one of the authors and a psychology professor at Ohio State, explains that users tend to draw broad conclusions from the limited information that algorithms provide.
“People lose information when they follow an algorithm, but they think that what they know can be generalized to other features and other parts of the environment that they have never experienced,” he said.
The researchers provide a simple illustration. Someone who has never watched a film from a country starts trying to watch it. When a streaming service makes a recommendation, the initial choice will prompt the algorithm to show similar films. This situation can limit users’ understanding of the diversity of genres and narrow their view of the culture in the film.
To test this phenomenon experimentally, the research team recruited 346 participants in a fiction learning task. They studied the characteristics of “alien” crystals with six different features. Under certain conditions, the personalization algorithm directs participants to keep seeing the same features over and over again, so they miss other information.
As a result, algorithm-directed participants saw fewer features and developed narrow observation patterns. When tested with new examples, they misclassified more often, but were more confident in their answers. “They were even more confident when they were completely wrong in their choice than when they were right,” Bahg said.
Turner added that these findings are relevant in the context of everyday life. “If you had young people who were really trying to learn about the world… what would happen?” he said. Vladimir Sloutsky from Ohio State University co-authored the study. (Science Daily/Z-2)
