This article was originally published in 2019.
The power of individual choice has long been a hot topic in debates around social media and privacy. If you value privacy above the platform, the argument goes, don’t create an account — easy-peasy, right?
New research, however, shows that this argument doesn’t truly hold water. Researchers from the University of Vermont and the University of Adelaide show that information gathered from users on social media platforms can be used to reliably predict another user’s later tweets. Furthermore, even if this latter user leaves the platform — or never joined in the first place — data mined from social media can be used to predict their future activities.
“There’s no place to hide in a social network,” says Lewis Mitchell, a co-author of the study.
Mitchell and his team gathered over thirty million public posts on Twitter from 13,905 different users/accounts, clumping them in networks of 15 users. Then, using messages sent to and from as few as 8 or 9 of a certain user’s contacts (this user is named the “ego” and the contact its “alters”), they proved that they can predict the content and wording of that user’s later tweets with an
One disturbing finding the team reports on is that “there is so much social information that an entity with access to all social media data will have only slightly more potential predictive accuracy (~64% in our case) than an entity that has access to the activities of an ego’s alters but not to those of that ego (~61%).”
In other words, if you can access the alters’ data, you can predict what the ego is going to say/do roughly as reliably as if you had access to the ego’s data itself. The online posts of the alters provide about 95% of the “potential predictive accuracy” of a person’s future activities, the team writes.
When you sign up for Facebook or another social media platform “you think you’re giving up your information, but you’re giving up your friends’ information too!” says lead author Prof. James Bagrow, a mathematician at the University of Vermont.
Remarkably, this doesn’t really disappear if you delete your account.
“[…] If an individual forgoes using a social media platform or deletes their account, yet their social ties remain, then that platform owner potentially still possesses 95.1±3.36% of the achievable predictive accuracy of the future activities of that individual,” the study reads.
Needless to say, this raises some very concerning questions about the nature of privacy on social media, and about the nature of privacy in today’s world in general.
At least in theory, the paper suggests that a third party can use data that an individual’s friends posts on social media to accurately profile them — raging from where they like to hang out to their political affiliations, favorite products, and even religious commitments. They don’t need to post anything themselves, they don’t even need to have a social media account. If they have friends on social media their privacy is, potentially, compromised.
Information flows on social media platforms have become a huge force in shaping customer choices, election results, or protest movements, the team writes. These platforms contain an immense pool of data pertaining to their users — and, as the study shows, their friends.
The public consciousness is very interested in protecting their online privacy, especially after debacles such as the one involving Cambridge Analytica. Bagrow’s team showed that this concern is valid, and, although there is a mathematical upper limit on how much predictive information a social network can hold, it’s not limited to its users alone.
“You alone don’t control your privacy on social media platforms,” says professor Bagrow. “Your friends have a say too.”
The paper “Information flow reveals prediction limits in online social activity” has been published in the journal Nature Human Behavior.