The algorithms behind social media platforms can change your attitude towards people with different political views in just one day, suggesting that tech companies really can have a strong influence on how we perceive others.
Most social media platforms use algorithms to sort a user鈥檚 feed, generally ranking posts by someone鈥檚 likelihood of engaging with them and presenting the most engaging first 鈥� but this may inadvertently be pushing people apart.
鈥淭he trend in polarisation more generally is quite concerning,鈥� says at Stanford University in California. 鈥淧olarisation can lead to a lot of problems, including political violence.鈥�
Advertisement
In an effort to study the impact of social media algorithms on political polarisation, Piccardi and his colleagues asked 1256 US users of X (formerly Twitter) to install a web browser extension that re-ranks content on the platform in real time.
This extension intercepted posts before they were seen by users and fed them to the GPT-4 large language model, which ranked the posts by how politically polarising they were, according to a score based on eight potentially anti-democratic expressions, such as opposition to bipartisanship. Some users were then given a greater number of polarising posts than normal, others saw less, while a control group experienced X鈥檚 usual algorithmically curated feed.
The researchers surveyed each participant before, during and after the experiment, which lasted 10 days, to measure their perception of people with different political beliefs.
Perhaps unsurprisingly, they found that reduced exposure to politically polarised content made people warmer to those they disagreed with politically, while greater exposure fostered harsher sentiments. What was unexpected is how quickly this occurred, with survey answers changing after just one day. 鈥淲e were not confident this would work,鈥� says Piccardi. 鈥淭he result is surprising, and we鈥檙e excited by that.鈥�
Changing the way content was presented didn鈥檛 change traditional engagement metrics such as likes, reposts or time spent on the platform.聽Participants exposed to a greater volume of polarising content also reported heightened negative emotions, including anger and sadness, during the study, but this effect disappeared in the days after the experiment.
The experiment highlights how important the algorithms that dictate how we consume information are to our perception of others, says at Princeton University. 鈥淭here is an ongoing debate about the impact of existing social media algorithms,鈥� he says, and this experiment shows that 鈥渢he last information we consume has some weight regarding our behaviour鈥�.
Tech companies should take heed, says Piccardi. 鈥淭his is a starter to show that algorithmic curation can impact polarisation,鈥� he says. 鈥淚 think platforms should really consider that optimising only for engagement can be dangerous.鈥�
The fact that it is possible to switch on or off polarising content demonstrates that social media platforms have a role in managing this, says at Northumbria University in Newcastle, UK. 聽鈥淚 wonder how much of this experiment is already in action on platforms,鈥� she says.
One solution may be the approach taken by Bluesky, the decentralised social network that allows users to pick their own algorithms. 鈥淏luesky adds a lot of options for people to 鈥榮elf-select鈥� into what they consume,鈥� says Ribeiro. 鈥淚t remains a mystery to me whether having individual opt-in algorithms actually changes much. But it鈥檒l definitely imply new platform dynamics 鈥� if adopted.鈥�
arXiv
