Nearly all climate scientists agree that humans are causing global warming. Yet, a recent survey by the Yale Program on Climate Change Communication found that only 59% of American registered voters accept this conclusion. Lack of awareness could be part of the problem—many people may not realize that there is an overwhelming scientific consensus on global warming. But perhaps an even greater problem is a process that psychologists call belief polarization. When our deeply held beliefs are challenged, we often double down, accepting evidence that confirms the beliefs and dismissing evidence that refutes them. For highly politicized issues such as climate change, the upshot of this biased assimilation of evidence is the emergence of groups with opposing views that are equally convinced they are right.

New research on collective intelligence demonstrates an innovative approach to combating belief polarization. The core idea of collective intelligence is that collaboration can improve both individual and group performance, depending on characteristics of group members and on conditions that facilitate (or inhibit) the exchange of information. In a recent study, researchersfrom the University of Pennsylvania’s Annenberg School for Communication set up social networks and found that they could eliminate belief polarization between liberals and conservatives on the issue of climate change. The key to the change was keeping people’s political affiliation hidden, a finding consistent with the idea that collective intelligence is greatest under conditions that promote the free exchange of alternative viewpoints. The study provides a proof-of-concept for how social media can be used constructively to improve people’s understanding of climate change and other urgent issues, while eliminating polarization.

The 2,400 participants were selected for the study based on their political ideology; on a pre-screening questionnaire, half identified as liberal and the other half conservative. Through Amazon’s Mechanical Turk, these participants were presented with a NASA graph showing the average amount of monthly Arctic sea ice between 1979 and 2013 and asked to make a forecast for 2025. The graph shows a “saw-tooth” pattern, with ups and downs over short spans. However, the overall trend is downward; therefore, the correct answer is a value lower than the final point on the graph. To be exact, according to NASA, the monthly ice average will be 4 million square kilometers in 2025—a reduction of 43% since 1979. There were three rounds of the task—an initial estimate and two opportunities for revision.

Each participant was randomly assigned to a control condition, or to one of three bipartisan “network” conditions that included an equal number of liberals and conservatives. In the control condition, participants worked on the task in isolation. In the network conditions, after providing their initial estimate, participants were exposed to information about their four network “neighbors,” two of whom were conservative and two of whom were liberal. In the first of these conditions, along with their own estimate, participants saw on their screens the average of their neighbors’ estimates. In the second condition, which primed participants to think about politics, these two values were displayed above logos of the Democratic Party and Republican Party—an elephant and a donkey. In the final network condition, which primed participants to think specifically about their neighbors’ politics, the values were presented along with the neighbors’ screennames and political affiliation (conservative or liberal).  

In the control condition, liberals gave more accurate estimates in the first round than did conservatives (74% vs. 61%), and neither group improved much across rounds. By contrast, in the network condition with no political cues, accuracy was higher for liberals than for conservatives in the first round, but both groups improved to greater than 85% by the third round, eliminating the difference between the groups. In the two other network conditions, which included political cues, both groups showed improvement across rounds. However, belief polarization persisted, with liberals outperforming conservatives in all rounds. Taken together, the results indicate that collective intelligence can emerge in bipartisan social networks, especially in the absence of political identifiers.

In this study, which focused on climate change, liberals were initially more accurate in their interpretation of evidence than conservatives. However, for other issues, just the opposite may be true. For example, some of the most strident opposition to vaccination comes from liberals. For these issues, there is every reason to think that bipartisan networks can be as effective in eliminating belief polarization as they appear to be for climate change.

As a 2018 report by the United Nations’ Intergovernmental Panel on Climate Change implored, rapid, far-reaching changes in human behavior are needed to limit global warming. In the United States and elsewhere, the polarization of people’s beliefs over the cause of global warming is a major impediment to implementing and sustaining such changes. This new study shows how social media can be used to harness the power of collective intelligence to combat this pernicious problem. The more general message of this study is one that we would do well to remember at this particular moment in history: When we set aside politics, there can be real value to having conversations with people with whom we disagree.