A recent study published in the Proceedings of the National Academy of Sciences suggests that our preexisting beliefs deeply influence how we learn new information. By tracking eye movements and decision-making during a simulated news evaluation game, scientists found that people readily learn from rewards that match their existing views but struggle to adapt when rewards challenge their preconceived notions.
These findings provide evidence for the cognitive pathways that allow misinformation to persist. This dynamic explains why simply presenting factual corrections often fails to change minds.
People increasingly rely on social media platforms for news, where algorithms filter content to match users' existing preferences. This digital environment provides a fertile ground for disinformation to spread, raising the question of why individuals continue to believe false content even when fact-checking is available.
"I began seriously considering this line of research in 2021, after witnessing the damage misinformation caused during the COVID-19 pandemic, particularly in relation to vaccination," said study author Stefano Lasaponara, an associate professor at Sapienza University of Rome. "That experience led me to wonder to what extent fake news might affect not only what people believe, but also how they learn from feedback."
Lasaponara and his colleagues sought to understand how a person's preexisting judgments and internal confidence interact with learning from external feedback. They designed the study to test whether our tendency to favor belief-consistent information might be rooted in basic learning mechanisms.
To explore these questions, the scientists recruited 28 healthy young adults, aged 18 to 36, for a three-part experiment. In the first phase, participants viewed 324 news headlines that had circulated on social media. Half contained real news events, and half contained false information. Participants judged whether each headline was true or fake.
They also wagered a virtual amount of money, from zero to 99 cents, on their answer. This bet served as an indicator of their internal confidence. Based on these answers, the scientists grouped headlines into four personalized categories for each participant: true with high confidence, true with low confidence, fake with high confidence, and fake with low confidence.
During this phase, researchers used eye-tracking glasses to measure participants' pupil dilation as they read. Pupil dilation indicates mental effort, focused attention, and physiological arousal.
In the second phase, participants played a computer game where they chose between pairs of the headlines they had rated. The goal was to select the headline that would win a 20-cent reward. Unknown to participants, rewards were not randomly assigned.
In different rounds, an 83 percent chance of winning was tied to specific categories from the initial evaluation. For example, in one round, picking headlines judged as true provided the reward. In another, picking headlines judged as fake gave the reward. Other rounds rewarded choices based on high or low confidence, and one round gave rewards at random for baseline comparison.
The third phase tested whether the learning game changed participants' minds. Scientists showed the original headlines again, along with their initial judgments and confidence wagers. Participants could confirm their original judgment or change their mind. If their final answer matched the actual status of the news, they kept their wagered money.
The outcomes showed that participants learned very differently depending on the hidden rules. When the game rewarded them for choosing headlines they already believed to be true, they learned quickly and earned high scores. Performance dropped when the game rewarded them for picking headlines they believed were fake. Participants also struggled when rewards were tied to confidence levels rather than beliefs about truth.
"One important takeaway is that our prior beliefs can begin shaping our decisions even before we explicitly express a judgment," Lasaponara said. "In our study, these pre-existing convictions were strong enough to influence learning itself."
To understand the underlying mental strategies, scientists used computational modeling. The models revealed that when rewards matched a participant's belief in the truth, they used broad, generalized rules to make choices.
When rewards no longer matched their sense of truth, participants abandoned these strategies. Instead, they reverted to simply reacting to feedback on a trial by trial basis, which was less effective.
The eye-tracking data provided physical evidence that beliefs engage our nervous systems before a conscious choice. In the initial phase, participants' pupils dilated more when looking at headlines they would later judge with high confidence. This suggests strong subjective beliefs trigger an early physical arousal response. During the learning phase, pupils dilated when participants faced a mental conflict, such as choosing between a strongly held belief and a competing reward signal.
"I expected to find pupillary effects related to the moment of decision itself, but I did not expect to observe them at an earlier stage, during the formation of a belief-consistent choice tendency," Lasaponara noted. "That suggests the influence of prior beliefs may begin unfolding before an overt response is made."
When participants received feedback against their established beliefs, their pupils widened, indicating cognitive surprise and increased mental load. In the final feedback phase, participants showed a strong tendency to stick to their original opinions. They rarely changed their minds, especially if they had placed a high confidence wager initially.
High confidence made people resistant to changing their minds regardless of whether the headline was actually true or false. Participants were slightly more willing to update beliefs if they had initially expressed low confidence.
There are limitations to keep in mind. Because participants experienced all reward rules back to back, learned rules from one round might have affected behavior in the next. "An important caveat is that this study does not yet allow us to make strong claims about correcting misinformation, or about when and how people truly change their minds after learning," Lasaponara explained. "Our results show that prior beliefs can bias reinforcement learning, but they do not yet tell us how to reliably undo that bias."
The experiment relied exclusively on political and social news headlines, meaning these learning patterns might look different with neutral topics. Future research could expand by using different types of information.
"Our broader goal is not only to better understand why people believe fake news, but also to identify the conditions under which misinformation becomes less effective," Lasaponara added. "In follow-up studies, we are investigating whether different reinforcement structures can lead to varying degrees of belief updating."
Scientists could also design experiments that explicitly present participants with direct evidence contradicting their beliefs, rather than just changing reward rules. This would help map conditions that might encourage people to update stubborn opinions.
"The title is also a small nod to Metallica, whom I am a big fan of," Lasaponara added. "More importantly, this work would not have been possible without my co-authors, especially Valentina Piga and Silvana Lozito."
The study, "Eye of the beholder: Pupillary response reflects how subjective prior beliefs shape reinforcement learning with fake news," was authored by Silvana Lozito, Valentina Piga, Sara Lo Presti, Angelica Scuderi, Fabrizio Doricchi, Massimo Silvetti, and Stefano Lasaponara.