An Illinois-led study suggests opinion polarization caused by data overload
It is commonly observed that the rise of social media seems to have been correlated with a rise in social and political polarization. Many theories have tried to explain the phenomenon, but a recent Illinois-led study concludes that the culprit might simply be information overload, and proposes new algorithms to help people navigate it better.
“The internet and online social media have surrounded us with unprecedented amounts of real-time information,” says Tarek Abdelzaher, the project lead and a faculty member in the Coordinated Science Lab at the University of Illinois at Urbana-Champaign. “Yet, a person’s capacity to absorb information has not increased over the decades. I still read at the same speed my parents did.”
This means humans are chronically overloaded and see an ever-shrinking fraction of the ever-expanding amount of available information. As it becomes harder to navigate the growing menu of information sources and options, the natural instinct is to gravitate towards the familiar – the opinions and sources that support a person’s already-held beliefs. Search engines, like Google, can learn a person’s preferences and rank information for him or her accordingly, further reinforcing existing beliefs. People end up in “bubbles” surrounded by others who think like them, and society ends up more polarized.
People have long felt a kinship with others who think like them, so what’s different now? Social media have made it exceptionally easy to find friends and create communities with those who hold similar worldviews, and also easy to unfriend and block those with contrary opinions. Abdelzaher’s study, conducted in collaboration with other researchers at the University of Illinois, the Rensselaer Polytechnic Institute, the University of California, Los Angeles, and the University of California, San Diego, suggests that the sheer volume of information available today, combined with greater connectivity to the information, makes the effects worse. That was the focus of the team’s recently published paper, “The Paradox of Information Access: Growing Isolation in the Age of Sharing.”
“It’s not the first time that affordances of connectivity and increased access have led to polarization,” said Abdelzaher, a Willett Faculty Scholar and computer science (CS) professor. “When the US interstate freeway system was built, urban socioeconomic polarization increased because connectivity allowed people to self-segregate into more homogenous sprawling neighborhoods.”
The creation of the information superhighway has had similar polarizing effects.
Information volume is a big culprit, the study confirms. When a search engine like Google returns 1,000,000 matches and a person reads the top three, chances are that he or she is not reading a representative sample, even if those three were the best matches, the researchers point out. Top matches do not offer a representative view when they are chosen from a huge pool, and often come from sources the person has visited before. The researchers believe some notion of summarization is needed instead.
“Travel sites, such as TripAdvisor, summarize reviews of individual venues; a traveler does not simply get a ranked list of tens of thousands of reviews,” Abdelzaher explains. “Travelers can find out specifically what others liked or did not like and can separately read the good and the bad reviews. There is no analogy to such summarization in today’s general information search.”
The solution, being developed by Abdelzaher and his collaborators, involves restructuring how search results are presented and sorting out results that include false information. The ideal web search would include a vetted summary of all the results that presents multiple sides of an issue.
To achieve such a solution, Abdelzaher is relying on Illinois CS colleagues Jiawei Han and Heng Ji, along with Alex Schwing of electrical and computer engineering. Han is a data and text mining expert and has made fundamental contributions in the field of hierarchical content abstraction. He can take thousands of results and present them in a connected, tree-like display rather than a ranking system. Ji brings her expertise in statistical linguistics and consistency analysis, allowing her to take massive amounts of information and analyze it for consistency and accuracy, while using algorithms to flag inconsistencies and errors. Schwing, also a CSL member, is a machine learning and vision expert who can extend the summarization to include visual information in addition to text.
With their combined expertise, the researchers are developing algorithms that can better organize search engine results, ensure their accuracy, and provide people with information that they might not otherwise have found, helping break them out of their thought bubbles.
“The work protects democracy itself because democracy is based on informed constituents,” said Abdelzaher. “If we are misinformed or not aware of critical facts, we can make bad decisions. The narrow bubbles we inadvertently end up in may jeopardize the way democracy functions, which is why this work is important.”
“The Paradox of Information Access: Growing Isolation in the Age of Sharing” was published on arXiv.org with support from DARPA contract W911NF-17-C-0099. Abdelzaher was joined in the research by, Heng Ji, Jinyang Li, and Chaoqi Yang at the University of Illinois at Urbana Champaign, John Dellaverson and Lixia Zhang at the University of California, Los Angeles, Chao Xu at the University of California, San Diego, and Boleslaw Szymanski of the Rensselaer Polytechnic Institute. Prof. Jiawei Han and Alex Schwing, at the University of Illinois, collaborated on the study.