,
14 tweets,
5 min read
Read on Twitter
Social networks can distort decision making and produce undemocratic outcomes, our latest study nature.com/articles/s4158…
We describe “information gerrymandering” — how a social network's structure can bias a collective decision towards one party, even when both parties have equal sizes and each person has the same influence.
Joint work with great colleagues, led by @al_cibiades.
Joint work with great colleagues, led by @al_cibiades.
In the paper we study the process of group decision-making on an influence network. We describe how one group, or party, can gain an advantage from information gerrymandering arising from the topology of who influences whom. 
First we studied a mathematical model of individual behavior in a collective decision.
Then we verified the predicted effects of information gerrymandering in social network experiments on thousands of human subjects playing a simple voting game.
Then we verified the predicted effects of information gerrymandering in social network experiments on thousands of human subjects playing a simple voting game.
Information gerrymandering systematically skewed the vote outcomes by 20% in our experiments, even when both sides had equal membership and each person the same influence.
Zealot bots, strategically placed on the network, also induced information gerrymandering and biased the vote outcomes in our experiments.
Finally we analyzed real-world influence networks, including online political discussions prior to US elections, and bill co-sponsorship in US and European legislatures. We found extensive information gerrymandering, similar to what induced large vote skews in our experiments.
So how can network structure produce undemocratic outcomes?
The idea is analogous to electoral gerrymandering, where congressional districts are drawn to give one party a disproportionate number of seats.
The idea is analogous to electoral gerrymandering, where congressional districts are drawn to give one party a disproportionate number of seats.
Information gerrymandering arises from asymmetry in the structure of influence edges between individuals.
A party is at a disadvantage when it concentrates most its influence in tight bubbles, while leaving some members in bubbles dominated by the opposition. Video by Stewart
A party is at a disadvantage when it concentrates most its influence in tight bubbles, while leaving some members in bubbles dominated by the opposition. Video by Stewart
We can measure the extent of information gerrymandering in a network with a quantity we call the "influence gap" which, we show, correlates with vote outcomes across diverse networks.
Information gerrymandering poses a threat to democratic decision making, even in the absence of misinformation or "fake news".
Also, the capacity to gerrymander information will tend to exacerbate deadlock as multiple parties seek to use it to their advantage.
Also, the capacity to gerrymander information will tend to exacerbate deadlock as multiple parties seek to use it to their advantage.
Future work will hopefully address how to combat information gerrymandering in real-world social networks; and how to reverse the trend towards polarization as parties deploy information gerrymandering.
For other perspectives see the commentary from Bergstrom & Bak-Coleman (nature.com/articles/d4158…), an editorial (nature.com/articles/d4158…), an exposition by Alex Stewart and me (theconversation.com/heres-what-hap…), and the research paper and supplement (nature.com/articles/s4158…).
And for a detailed walkthrough of how the study came to be see this from the lead author,
