Below, Damon Centola shares five key insights from his new book, Goodthink: The New Science of Collective Intelligence.
Damon is a professor at the University of Pennsylvania, where he directs the Network Dynamics Group and studies how ideas and behaviors spread through social networks, and how those networks shape our collective intelligence.
What’s the Big Idea?
Social influence is not inherently wise or foolish, unifying or polarizing, creative or conformist. What determines the outcome in any of these challenges is having the right structure to shape the flow of influence into intelligence. Once we understand how these structures work, we can design networks that help people learn from one another, make better decisions, and solve problems none of us could solve alone.
Listen to the audio version of this Book Bite—read by Damon himself—in the Next Big Idea App, or buy the book.
1. Overturning the myth of groupthink.
The idea of “groupthink” was popularized by psychologist Irving Janis. After studying decision-making in the White House during the Kennedy and Johnson administrations, Janis argued that loyalty and camaraderie among highly educated advisers suppressed critical thinking. For instance, before the Bay of Pigs invasion, President Kennedy’s advisers confidently made several incorrect assumptions about Fidel Castro’s military strength. The operation became a historic fiasco.
For Janis, the lesson seemed clear: when you put people together and let them influence one another, even the smartest individuals will start to make the same mistakes in judgment.
But while the Bay of Pigs fiasco shows that there is indeed a problem, it does not prove that social influence was the culprit. My research over the last two decades points to a more precise explanation: Smart people do indeed become less intelligent when a single voice dominates a discussion network. But contrary to the theory of groupthink, people actually become much smarter as a result of peer influence in their networks when that influence is distributed evenly throughout the group—that is, in networks where everyone has equal power.
To test this idea, my team and I created 32 online communities discussing vaccine safety. We placed a disinformation agent inside each one, repeatedly claiming that the number of life-threatening allergic reactions was nearly 10,000 people, when the true number was just 35.
In centralized networks (or “firehose” networks, in which a single person sprays their ideas across the group), the disinformation agent occupied the influential position at the center. As expected, group error increased dramatically, by nearly 50 percent, and somewhat unexpectedly, the conversations became angrier and more emotional.
“Smart people do indeed become less intelligent when a single voice dominates a discussion network.”
However, in egalitarian networks, where influence was distributed equally, something remarkable happened. Even though the same disinformation agent made the same inflammatory claims, communities in these networks became more accurate, with error decreasing by nearly 25 percent. Conversations were more civil and more people felt comfortable offering opinions.
In both kinds of networks, people influenced one another. Yet one structure amplified falsehood, while the other allowed the community to correct it. The problems that arose in the Kennedy and Johnson administrations were not an inevitable consequence of people thinking through things together. Rather, those problems resulted from people thinking together in a hierarchical structure that inevitably centralized social influence within the room and, in doing so, undermined the intelligence of those advisors.
2. Overturning the myth of echo chambers.
The second myth of collective intelligence is that talking only to people who agree with us inevitably makes us more polarized. In other words, if Democrats talk only with Democrats and Republicans talk only with Republicans, we assume that both groups will become more extreme and move farther apart. These like-minded groups are often called echo chambers.
That assumption sounds intuitive. But when my team tested it directly in an experiment with nearly 1,500 Democrats and Republicans, we found exactly the opposite: when influence is distributed equally within echo chambers, conversations among like-minded people can actually reduce polarization and move people closer to understanding the same facts.
In our study, we asked Democrats and Republicans politically charged factual questions about immigration, military spending, unemployment, and the Iraq War. We then placed them into politically homogeneous groups: Democratic echo chambers and Republican echo chambers. Conventional wisdom predicted that these echo chambers would magnify partisan bias.
But across a dozen repeated experiments, egalitarian echo chambers increased collective intelligence in both groups. Even more remarkably, polarization between Democrats and Republicans decreased by nearly 50 percent. Both groups moved closer to the facts, and, in doing so, closer to one another.
But why? Because in echo chambers, people were willing to learn from trusted peers. Most importantly, influence was distributed equally, so no single extreme voice could dominate the conversation. When we tested to see what would happen if echo chambers were centralized, we found that any bias at the center pulled both groups in opposite directions toward the extremes. Error increased in both groups, and polarization grew.
Echo chambers become dangerous when a small number of voices control the flow of information, when power is centralized around a firehose. But when influence is egalitarian, having an exchange of ideas take place within trusted communities of similar peers can make people feel safer, and help them to expose—and correct—their biases.
3. Overturning the myth that experts work best alone.
Take doctors, for instance. Medical errors are often treated as individual failures: perhaps a doctor lacked knowledge, time, training, or motivation. The common solution is to give doctors more time, money, and training. But giving doctors more resources and rewards does not reduce the rate of medical errors. In fact, most professional errors—and this is true for fields ranging from medicine to stock investing—are actually due to shared patterns of beliefs that are reinforced throughout professional networks. Money doesn’t change that, and can even reinforce the problem.
My team and I ran a study in which we brought together more than 3,000 clinicians from across the United States and asked them to evaluate clinical cases involving heart disease, diabetes, back pain, and geriatric care. In one set of experimental groups, doctors reflected independently. In the other groups, doctors were embedded in anonymous, egalitarian information-exchange networks. In the networks, doctors could see several of their peers’ diagnostic assessments, but not their treatment decisions.
After about five minutes of exchanging opinions in the network groups, doctors’ diagnostic accuracy doubled in quality compared with the independent groups. This translated directly into better treatment decisions. In the networks, nearly one-third of doctors who initially chose the wrong treatment subsequently changed to recommend the correct treatment for their patient.
“Giving doctors more resources and rewards does not reduce the rate of medical errors.”
We then looked even deeper into how networks can fix medicine. This time we studied whether the same network structure could help address race and gender bias in medical care. We recruited nearly 1,000 clinicians to evaluate two medically identical patients: a white man and a Black woman with the same condition and same guideline-recommended treatment. Strikingly, at the start of the study, clinicians were three times as likely to send the Black woman home rather than admit her to the hospital. But after doctors exchanged their clinical estimates within egalitarian networks, everything changed. Clinicians’ estimates became more accurate, which moved their recommendations out of “the gray” area where unconscious bias could shape patient treatment. Within just a few minutes, disparity in patient treatment had disappeared.
None of the doctors in our studies received additional medical facts, or more money, or more time. And none of them were lectured about bias and culture. The network alone changed how doctors interpreted the information they had and led them to apply what they already knew more expertly and equitably.
4. Overturning the myth that popularity is mindless herding.
The fourth myth rests on a common belief about popularity: that bestselling books, hit songs, and celebrated ideas succeed largely because people blindly follow whatever is already popular. But popularity can sometimes be a powerful form of collective intelligence.
Take Harry Potter. Did you know J.K. Rowling’s first book, Harry Potter and the Philosopher’s Stone, was rejected by 12 publishers before it was accepted? But once it was published, it became an instant classic. Or consider Dr. Seuss. His first book was rejected by 27 publishers. He was walking home to burn the manuscript when he happened to encounter an old friend who had just joined a publishing house. That chance encounter led to a contract, and eventually to The Cat in the Hat, How the Grinch Stole Christmas, and many other classics.
Are cultural successes merely accidents of history, or can crowds recognize quality that experts miss? Research suggests that cultural popularity can amplify the highest-quality products and help crowds discover ideas experts initially overlooked. The key to making this work is that the market needs to have a structure that allows the initially overlooked alternatives to stay alive long enough to be discovered.
One of my friends in graduate school, Matt Salganik, conducted a fascinating (and now famous) study of how this happens. In 2004, he built a music-downloading website called MusicLab and recruited over 10,000 teenagers to listen to and download songs by unknown bands. Matt divided participants among eight initially identical worlds, each with its own Billboard chart of popularity by download, and watched how popularity evolved. Would each world “randomly” choose its big hit, or would each independent crowd herd toward the same big winner?
Remarkably, in all eight worlds, the same song found its way to the top in every case. Even more striking, when Matt and his team ran another experiment in which they manipulated the Billboard ranking by placing the best song—the one that was consistently popular in all eight worlds—all the way at the bottom of the rankings, the crowd of teenagers nevertheless discovered it within just a few hours, and it quickly started climbing back to number one!
Social feedback doesn’t just help people discover good options; it can actively correct rankings when the official tally is artificially manipulated. That’s an amazingly robust finding for the intelligence of popularity. But crowd herding can go astray when initially unpopular options are eliminated too quickly. For instance, if Harry Potter or Dr. Seuss had disappeared after those early rejections, readers would never have had the opportunity to discover them.
Popularity is not necessarily proof of quality. But when cultural markets are fair and allow all options to be discovered, then the popularity that comes from reinforcing social feedback can offer a very powerful form of collective intelligence, which can (and many times does) rescue great ideas from obscurity.
5. Overturning the myth that faster communication drives innovation.
The final myth is that faster communication increases the rate of innovation. When organizations want more creativity, they often try to connect everyone to each other, create open office spaces, and give teams access to as much fast-spreading information as possible.
Yet this strategy is often self-defeating. Creativity flourishes in locally connected networks, where information spreads more slowly, and people repeatedly interact with a small number of diverse peers instead of seeing everyone’s work all at once.
Consider MIT’s legendary Building 20. It was a temporary plywood structure built during World War II that was used as overflow housing for a wide array of researchers arriving to campus from disciplines ranging from acoustics to applied engineering to astronomy. Building 20 was an island of misfit toys. It was unattractive, confusing, and so dilapidated that researchers could cut holes in walls and rewire the electrical panels in their offices. The long, pedestrian hallways forced scientists from different fields to wander around, often encountering unlikely colleagues from diverse disciplines, striking up unusual conversations, and developing unorthodox collaborations. The famous scientist Noam Chomsky, whose Building 20 office was especially unpleasant, wandered the halls and struck up conversations with researchers in biology, psychology, and computer science, fields that later helped shape his revolutionary work in linguistics.
“Creativity flourishes in locally connected networks.”
Building 20 combined two things that rarely coexist: intellectual diversity and limited local interactions. Its residents encountered people who thought differently, but at the same time they were not forced to engage with the vast array of ideas being pursued across the MIT campus. Nor did they typically interact with the expansive network of colleagues back in their home department. Building 20 created highly interactive networks within local hallways, yet limited information flow to the ideas in those hallways, which protected heterodox ideas emerging from Building 20 from being crowded out by approaches and solutions that were more commonly accepted, or looked most promising to established researchers.
To see whether the same principle would hold in a controlled setting, my team and I built a competition in which we placed data scientists onto teams to solve highly complex data science problems. There were eight competitions in total. On half the teams, everyone could see everyone else’s work, and a leaderboard highlighted the best current solutions. On the other half, participants could see only a few nearby peers (on their hallway), so it took a lot longer for any new idea to propagate around the entire team network. The teams only had 15 minutes to explore a strategy space with over 15,000 solutions.
The fully connected teams quickly converged on good solutions. The locally connected teams continued exploring, and soon discovered truly exceptional solutions. In fact, in half the competitions, the locally connected teams found the best possible solution, while none of the fully connected teams ever did.
When it comes to creativity in problem solving, increasing the flow of information can make everyone chase the same promising dead end. Innovation requires diversity, but it also requires slowing the spread of information to give an oddball idea time to catch up to the status quo and mature into a groundbreaking innovation.
Enjoy our full library of Book Bites—read by the authors!—in the Next Big Idea App:
