The Crowd and the Algorithm
How Recommendation Systems Reshaped Public Opinion
- 8 chapters
- 53m
- Sociology
- Free · no sign-up
The book explains how crowd psychology works when people make decisions together, then shows how algorithms amplify certain voices while silencing others. It covers wisdom of the crowd theory, collective behavior patterns, and how human-based computation differs from machine learning approaches. Specific chapters examine how social media use affects political participation and how recommendation systems can radicalize users over time.
Researchers studying these phenomena disagree about whether polarization is inevitable or avoidable. Some argue that algorithmic amplification creates filter bubbles, while others say people actively seek out like-minded content. This balanced analysis helps readers understand the complex relationship between technology and public opinion without taking sides.
Anyone interested in how digital platforms shape political discourse will find this research valuable.
Listen
-
Read this chapter
Overview
In the social sciences, framing refers to how people organize and understand reality through mental shortcuts and shared meanings. It's both a way individuals process information and a method of communication between people. Frames help simplify complex topics by connecting them to what we already know. In social theory, framing works like a mental filter shaped by biology and culture, influencing how we interpret events. These frames come from mass media, political leaders, or social movements and shape public perception. Framing can be positive or negative depending on context, and includes techniques like equivalence frames, where the same facts are presented differently, or emphasis frames, which focus on certain aspects. Journalism uses framing to influence readers without changing facts, often through word choice—like calling a developing human a "fetus" versus a "baby." Politically, framing packages issues in ways that suggest particular solutions, encouraging certain responses over others.
In sociology
Framing theory offers a way to understand how messages shape what people think and do, and analysts have applied it across many fields including communication studies, news reporting, politics, and social movements. In 1995, Johnson-Cartee looked at how framing works in news, while sociologist Bert Klandermans focused on collective action frames. According to Klandermans, the social construction of these frames involves public discourse—meaning the mix of media messages and personal conversations—and persuasive efforts during mobilization campaigns by groups, their opponents, or countermovements. It also includes moments when people become more aware during group actions.
History
The way we choose words has always shaped how we communicate and convince others. The concept of framing, which influences how we make sense of events, comes from Erving Goffman. In his work, he refers to Gregory Bateson as the one who first introduced him to the term “frame,” noting that idea appeared in Bateson’s paper "A Theory of Play and Fantasy." Goffman used the idea to describe mental frameworks—schemata of interpretation—that help people locate, perceive, identify, and label what happens around them. These tools give meaning to experiences and influence how we act. His thinking grew out of his earlier 1959 book, The Presentation of Self in Everyday Life, where he explored how individuals manage the impressions they make. That earlier work built on Kenneth Boulding’s concept of image.
Social movement theory
Framing helps explain how social movements gain traction by shaping how people see issues and aligning those views with the beliefs of participants. When movement messages match what people already think, it creates a powerful resonance that can shift public opinion and open political opportunities. Movements use strategies like bridging, amplification, extension, and transformation to build their narratives, while also adapting to opposition and changing circumstances. For example, in debates over abortion, groups have adopted terms like “Pro-Life” and “Pro-Choice,” which sound positive to most people. But framing can also be used to paint the other side as the villain—like protest signs that say “Abortion Kills.” This kind of language changes how people understand the issue by making the opposition seem morally wrong.
News framing
In communication studies, framing describes how news media coverage shapes public opinion. Richard E. Vatz’s work on rhetorical meaning connects directly to this idea, even though he doesn’t reference framing often. Framing effects are the changes in behavior or attitude that happen because information is presented in a certain way within public discussion. Today, many top communication journals publish research on media frames and how they affect people. These studies usually fall into two categories: some look at framing as something that happens — like how journalists choose which frames to use — while others study framing as something that influences audiences. The first kind is called frame building, and the second is known as frame setting.
Frame setting
When people encounter a new way of talking about an issue, they’re more likely to accept it if they already have ideas about how that issue works—this is called the applicability effect. The better acquainted someone is with a topic, the more power a frame has over them. For example, the more someone knows about the tobacco industry’s deceptive practices, the more persuasive it is to frame smoking-related health problems as the industry’s fault rather than the smoker’s. Researchers have studied many kinds of framing effects, from changes in attitudes and voting behavior to how people feel about issues. Some, like Iyengar, found that news frames can shape who people blame for social problems. Others have looked at how framing influences how deeply people think about an issue or how they evaluate political leaders.
Mass communication research
The way news is presented plays a key role in how people make sense of events. Media outlets shape what seems important by focusing on certain facts, values, or angles. They influence not just what we know, but how we think about it. By choosing which details to highlight, they guide our judgments without forcing a particular view. This process helps define shared meanings and priorities in society. The frames that media use become part of the common understanding of issues. In this way, they help form public opinion by suggesting how to interpret what's happening.
Sociological roots of media framing research
Media framing research comes from both sociology and psychology, with sociological framing looking at how communicators choose words, images, and styles to present information. This kind of research studies how social norms, organizational pressures, interest groups, journalistic habits, and political views shape what makes it into the news. Todd Gitlin was one of the first to study frames this way, looking at how the news media portrayed the student New Left movement during the 1960s. He argued that frames are "persistent patterns of cognition, interpretations, and presentation" that are often hidden and unspoken, shaping not just what journalists do, but also how people understand the world through their reporting.
-
Read this chapter
Overview
The idea that a group can be smarter than any one person is called the "wisdom of the crowd," and it's been around long before the internet. This concept has gained attention through websites like Quora, Reddit, and Wikipedia, where collective knowledge helps answer questions. The theory suggests that when many people give their opinions, the errors in individual guesses balance out, leading to better overall results. Jury trials rely partly on this idea, with a group deciding instead of just one judge or expert. Sometimes, sortition—choosing leaders randomly—is seen as a way to bring wisdom of the crowd into politics. Cognitive scientists have studied how this works with the mind. One of the oldest formal arguments for this is Condorcet's jury theorem from 1785.
Examples
Aristotle first wrote about the “wisdom of the crowd” in his work Politics, saying that while individuals may not be good men, collectively they can be better. In 1906, at a country fair in Plymouth, 800 people tried to guess the weight of a slaughtered ox. Statistician Francis Galton found that the median guess, 1207 pounds, was within 1% of the actual weight, 1198 pounds. This showed how a group’s estimates can center near the truth. More recently, businesses and political researchers have used this idea to gather consumer feedback, design ads, and predict election outcomes.
Surprisingly popular
Scientists at MIT's Sloan Neuroeconomics Lab, working with Princeton University, developed a method called the "surprisingly popular" technique. For a given question, people give two answers: what they think is right and what they think others will choose. The difference between those responses points to the correct answer. In tests, this approach cut error rates by 21.3 percent compared to simple voting, and by 24.2 percent when compared to basic confidence-weighted votes. It also improved results by 22.2 percent over more advanced methods that only use the highest-confidence answers.
Definition of crowd
In the wisdom of the crowd, the word "crowd" means a group brought together by an open invitation to join. Today, tools like Google, Facebook, and Twitter let many people share ideas and opinions at once, forming what some call "intelligent communities." But these digital groups can be affected by unfair representation, overly active users, or fake accounts, which weakens their ability to be wise. Researchers suggest mixing different platforms or using factor analysis to clean out bias and confusion. Crowds also work offline, sometimes with money offered for participation, and in places like the United States, jury duty requires people to take part.
Analogues with individual cognition: the "crowd within"
The idea that crowd guesses form probability distributions leads to comparisons with individual thinking. One suggestion is that people's judgments come from internal probability distributions, so averaging multiple estimates might approach truth better than individual guesses. This works best when estimates are independent, which is why delays help—like Vul and Pashler's 2008 study showing participants guessing three weeks apart were more accurate than those guessing immediately. Hourihan and Benjamin found people with lower memory spans saw bigger gains from repeated guesses, suggesting independence matters. Rauhut and Lorenz showed that asking oneself multiple times doesn't always help and can reduce accuracy. Müller-Trede found repeated self-estimates improved accuracy for certain question types but not general numerical ones. Van Dolder and Van den Assem confirmed within-person averaging improves accuracy, especially with time delays, but gathering estimates from different people remains more effective overall.
Dialectical bootstrapping: improving the estimates of the "crowd within"
Herzog and Hertwig tested whether people could improve estimates using dialectical bootstrapping, combining reasoned discussion with self-reflection. In their 2009 study, participants guessed historical dates like when electricity was discovered, then made a second estimate. Half simply guessed again, while the other half used "consider-the-opposite" strategy, imagining their original estimate was wrong and thinking through what might have been overlooked. Results showed this approach led to better individual estimates than just making two guesses without reflection, though it didn't beat the wisdom of the crowd—averaging each person's first guess with a random stranger's. Hirt and Markman later found that simply considering any plausible alternative, not just the opposite, also improved judgments. However, not all studies supported repeated self-assessments boosting accuracy; Ariely and colleagues found that averaging multiple estimates from the same person didn't significantly improve them.
Challenges and solution approaches
The wisdom-of-the-crowd effect works best when people have diverse opinions and independent judgments, because averaging can smooth out random errors but not systematic biases. Scott E. Page's diversity prediction theorem shows that greater group diversity leads to lower collective error. Yet, people often don't share sincere views, instead voting strategically. Miller and Steyvers found that allowing limited communication between participants improved accuracy, as individuals integrated prior knowledge with their own. Wisdom-of-the-crowd algorithms perform well when answers are precise, like in geography or math, but struggle when there's no clear solution. The WICRO algorithm tries to identify experts by how closely their responses align within a domain. Simple averages can be misleading if they mix knowledgeable and unskilled input. Groups may also fall into premature consensus, reducing accuracy—though diversity in the group helps counter this. Research from the Good Judgment Project shows that prediction polls organized without forcing early agreement produce better results than traditional markets.
-
Read this chapter
Overview
Crowd psychology looks at how people change when they're part of a group, and how that's different from how they act alone. This field studies both individuals in a crowd and the group as a whole. Two key ideas are deindividuation — when people lose their sense of personal responsibility — and the belief that behavior is universal within the group. These effects grow stronger as the crowd gets larger. Important thinkers in this area include Gustave Le Bon, who lived from 1841 to 1931, Gabriel Tarde, from 1843 to 1904, and Sigmund Freud, who was alive from 1856 to 1939. Their theories are still used today, especially in trying to predict how crowds behave during normal or emergency events — like preventing stampedes or crushes.
Origins
In the late 19th century, debates over crime and human behavior were heated between Italian and French schools of thought. Enrico Ferri’s biological theory suggested that criminality was inherited, but critics like M. Anguilli and Alexandre Lacassagne argued that social environment played a key role. At the 2nd International Congress of Criminal Anthropology in Paris, August 1889, Professor Lombroso's ideas about the “born criminal” were challenged by figures such as Léonce Pierre Manouvrier, who dismissed them as outdated phrenology. Later, Scipio Sighele and Gabriel Tarde discussed how to determine responsibility within crowds, with Sighele writing The Criminal Crowd and Tarde La criminalité comparée. Earlier still, Charles Mackay’s Extraordinary Popular Delusions and the Madness of Crowds (1841) and Hippolyte Taine’s The Origins of Contemporary France (1875) had shaped how people thought about crowd behavior.
Types of crowds
Crowd psychology lacks consensus on categorization, though scholars like Momboisse (1967) and Berlonghi (1995) have attempted it, with Momboisse distinguishing casual, conventional, expressive, and aggressive crowds, and Berlonghi grouping them as spectator, demonstrator, or escaping. Sociologist Herbert Blumer described four emotional levels: casual, conventional, expressive, and active. A casual crowd is simply people in the same place at the same time, without shared goals. Conventional crowds gather for specific events like lectures or concerts, acting in predictable ways. Expressive crowds come together to display emotion—such as at a political rally or Mardi Gras. Active crowds may become violent, as seen in mobs. Crowds can also be grouped as active or passive, with active ones including aggressive mobs like those during football riots or the 1992 Los Angeles riots, escapist mobs fleeing danger such as at the 2021 Astroworld Festival, acquisitive mobs fighting over scarce resources, and expressive mobs gathering for a cause like civil rights sit-ins or religious revivals.
Le Bon
Gustave Le Bon described crowds as moving through three stages: submergence, where individuals lose their sense of self and responsibility; contagion, where ideas and emotions spread like a disease among the crowd; and suggestion, where shared unconscious beliefs take hold. He believed this collective mind could only lead to destruction, because people in crowds feel less accountable for their actions. Critics have challenged his view, noting that studies show crowds don’t act as one unit, but rather consist of smaller groups with different intentions. For example, after a panic at a 1979 The Who concert, Norris Johnson found that people were mostly trying to help each other. Le Bon’s theory also overlooks how social and cultural context shapes crowd behavior, and some argue that participants vary widely in their willingness to follow norms.
Freudian theory
Freud believed that when people join a crowd, they unlock their unconscious mind, with the super-ego or moral center being replaced by a charismatic leader. McDougall agreed, saying that in crowds, simple emotions dominate and complex ones are rare, reducing shared feelings to a primitive level. He described this as resembling the "primal horde" of pre-civilized society. Freud argued one must rebel against the leader to reclaim individual morality. Theodor Adorno countered that mass spontaneity was artificial, shaped by modern life's structures. He said the bourgeois ego dissolves into the Id and a "de-psychologized" subject emerges. Adorno also claimed the bond between masses and leaders through the spectacle is fake. As he put it: “They do not really identify themselves with him but act this identification, perform their own enthusiasm, and thus participate in their leader's performance.”
Deindividuation theory
In crowd situations, people can lose a sense of who they are through deindividuation, a concept first spelled out by American social psychologist Leon Festinger and colleagues in 1952. The idea builds on Gustave Le Bon's work and suggests that anonymity, group unity, and arousal weaken personal controls like guilt or self-evaluation. This can make individuals more sensitive to their surroundings and less likely to think things through rationally, sometimes leading to antisocial behavior. Philip Zimbardo expanded on this, noting that when people can't clearly see themselves as objects of attention, they're freed from normal social restraints. He explored how sensory overload and lack of constraints blur mental input and output. Though his Stanford Prison Experiment was later criticized as unscientific, it illustrated the power of deindividuation. Studies have shown mixed results on aggression, with behavior depending more on the situation's norms—like if someone is deindividuated as a KKK member or a nurse. Zimbardo also applied the idea beyond groups, linking it to acts like suicide and murder.
Convergence theory
Crowd behavior, according to convergence theory, doesn’t come from the crowd itself, but from people who are already alike coming together. Floyd Allport said that “An individual in a crowd behaves just as he would behave alone, only more so.” This theory suggests that crowds form from people with similar views, and their actions get stronger when they’re in the group. It claims that what happens in a crowd is not irrational—it’s the logical expression of shared beliefs. But critics point out that this view ignores how social forces shape identity and action. Some research from the 1970s riots even found that participants were less likely than others to have prior criminal records, challenging the idea that crowds lead people to act against their usual selves.
Emergent norm theory
Crowds don’t start out with shared ideas or behaviors—according to Ralph H. Turner and Lewis Killian’s emergent norm theory, norms develop from within the group as it forms. At first, people mill about without clear direction, but then key individuals step forward with distinctive actions or personalities that draw attention. When others follow without objection, those leaders gain influence and shape what becomes the crowd's behavior. This process is influenced by conformity, as shown in Sherif’s and Asch’s studies, and by the idea that if everyone is doing something, it must be right—a concept Allport called the universality phenomenon. These norms can be either positive or negative, depending on the leader’s influence. Critics argue that crowds often lack self-awareness and that existing sociocultural norms play a bigger role than this theory allows.
-
Read this chapter
Overview
Algorithmic amplification spreads content widely on Facebook, YouTube, TikTok, and X through recommendation systems that score posts based on user engagement likelihood, then show them to more people regardless of follow status. A video gaining views reaches new audiences, with early attention feeding back into the system. While helping content discovery and creator visibility, these systems are linked to misinformation, extremist material, and mental health concerns, especially among young users. Researchers found mixed results on how much ranking affects user attitudes, with some studies suggesting user behavior matters more than recommendations. Governments in the EU, UK, US, and China are working on different rules, with the EU's Digital Services Act requiring platforms to assess risks and offer non-profiled options, while the UK's Online Safety Act requires algorithmic risk assessments and child safety measures. In the US, lawmakers have debated liability and free speech issues, and China has mandated users can opt out of personalized recommendations.
Terminology
Algorithmic amplification lacks clear definition, according to Jonathan Stray and colleagues' 2024 review. They explain platforms spread content via information cascades driven by user sharing and algorithmic recommendation, calling this "amplification," yet the term is hard to operationalize as many definitions reduce to simply showing content once, requiring disputed baselines. The concept appears in legislative discussions, such as a 2025 House of Commons report. In the EU, the Digital Services Act flags recommendation systems as possible systemic risk. Meanwhile, the U.S. Filter Bubble Transparency Act proposes platforms using "opaque algorithm" must offer version with "input-transparent algorithm," avoiding user-specific data unless explicitly provided by user for that purpose.
Development of recommendation systems
Recommendation systems began in the early 1990s, when they were first tested for filtering email and news, with systems like Tapestry and GroupLens leading the way. By the 2000s, platforms like Facebook and YouTube began sorting and highlighting content, with Facebook’s News Feed introduced in 2006 and YouTube changing its algorithm in 2012 to favor watch time. TikTok, launched in 2018, relies almost entirely on algorithmic curation through its For You page. These systems also expanded beyond social media—Spotify uses behavioral data to suggest music, while Amazon adopted item-based filtering in 1998. Law professor Amy Adler noted that adult content platforms, especially those controlled by Aylo, use similar algorithmic methods to shape what users see, shaping not just visibility but individual preferences through feedback loops and categorization.
Mechanisms
A platform needs a system to pick what content to show you, ranking pieces based on how useful they might be. It uses two main methods: looking at similar users' behavior to guess your tastes, and machine learning to predict engagement based on past actions. These systems score content by click, share, like, watch, or time-spent likelihood. What you actually do online doesn't always match what you say you want. Early engagement makes content more visible, creating feedback loops that can make some things get shown much more than others, even if they start equally good. Over time, this reduces variety in what people see and makes everyone's preferences more alike.
Beneficial and public-interest uses
Recommendation systems help people find content in vast digital libraries, improving discovery. In public health, platforms spread medical information quickly but also risk sharing false claims. During emergencies, social media becomes key for updates and relief efforts, though it sometimes spreads misleading reports. On music platforms, these systems act as cultural intermediaries, shaping what listeners hear. A 2023 UK government report found most music creators worried about bias favoring certain artists, while only half of consumers expressed concern for their own listening habits. The report noted 70% of music streamed is still chosen by users themselves. Musicologist Georgina Born and computer scientist Fernando Diaz argue these systems should promote diversity and shared experiences, not just individual engagement, drawing on traditions from public broadcasters like the BBC.
Effects on information ecosystems
Algorithmic amplification changes how information spreads online, shaping the whole information environment. Studies look at how this affects the spread of misinformation and harmful content, as well as who gets seen and paid more—creators and news outlets alike. Other research explores how these systems create filter bubbles, deepen political divisions, and impact young users’ mental health. Some work also examines how state actors use recommendation systems to influence public opinion.
Misinformation and harmful content
In 2018, Soroush Vosoughi and colleagues discovered that false news spread faster and reached more people than true stories on Twitter, with bots moving both kinds of content at similar speeds, pointing to human behavior as the key difference. Stray and others later questioned whether speed was driven by recommendation algorithms or user choices, distinguishing between radicalisation and polarisation while noting online recommendations involved chat rooms, personal connections, and life situations. Studies generally show recommenders push content toward engagement but offer little proof of radicalising effects due to weak statistical power. One example cited was the military's treatment of the Rohingya in Myanmar, where Amnesty International claimed Facebook features helped fuel anti-Rohingya hate before 2017. In early 2019, Joe Whittaker and colleagues tested YouTube and Reddit using automated accounts. On YouTube, a far-right account was twice as likely to see extreme content, while neutral ones saw much less. Reddit showed no significant difference. An analysis of Gab's timelines was exploratory due to technical problems, but its Popular feed seemed to favor fringe over moderate material.
Creator visibility and economic effects
A small number of major platforms now control where audience attention goes algorithmically, acting as gatekeepers whose system flaws strongly influence what people see and share. They grow value by boosting audience size, time spent, and interaction—data that helps advertisers target users. Sociologist Zeynep Tufekci notes how online civic space shifted around 2005 from personal blogs to centralized platforms where owners decide what users encounter through ranking systems. This gatekeeping affects not only what readers see but also what publishers create. Juliane Lischka and Marcel Garz studied Facebook and Twitter posts from 37 German news outlets between January 2013 and December 2017, finding that clickbait averaged 5.9% of Facebook posts and 2.8% of tweets, with a few outlets using it much more. Supply and interaction followed an inverted U shape, peaking before dropping off. Testing whether Twitter's algorithmic ranking increased clickbait, they found no link—clickbait was already growing before the platform introduced ranking. From 2023 onward, generative AI tools lowered content creation costs, increasing material for platforms to rank. Marketing scholars Tianxin Zou, Zijun Shi and Yue Wu modeled this using game theory, predicting that modest quality improvements in low-quality content would crowd out high-quality work, reducing both consumer welfare and creator profits—especially on platforms that screen for quality.
-
Read this chapter
Overview
Recommender algorithms on platforms like YouTube and Facebook are designed to keep users engaged by showing them content they’re likely to watch, like, or share. These systems track how long people spend with certain videos, what they click on, and how they react, then use that data to serve up more of the same. Over time, this can lead people toward increasingly extreme material. The effect is worsened by echo chambers—where users are repeatedly exposed to content that confirms their existing beliefs. This process may contribute to radicalization, though it's debated whether algorithms are truly to blame. Social media companies often don’t remove these channels because they drive engagement, even when the content is divisive or dangerous. Studies have shown mixed results about how much algorithms actually promote extremism.
Social media echo chambers and filter bubbles
Social media platforms shape what you see by learning what you like, keeping you engaged in a constant scroll known as a filter bubble. This setup leads users into echo chambers—closed groups where beliefs are amplified and reinforced by like-minded people. These spaces spread information without challenge, feeding confirmation bias. According to group polarization theory, being surrounded by similar views can push individuals and groups toward more extreme positions. The National Library of Medicine notes that online users tend to seek out ideas matching their worldview, dismiss opposing views, and form tight-knit groups around shared stories. When polarization runs high, false information spreads quickly.
Facebook
Facebook's algorithm prioritizes content that generates clicks, comments, or shares, ranking posts based on friend interactions, virality, and divisive material while personalizing feeds according to user interests, potentially trapping people in echo chambers with troubling content. Users often remain unaware of the interests the algorithm uses—74% of Facebook users had never seen their "Your ad Preferences" page until researchers pointed it out. The platform employs artificial intelligence to control what users see, with internal documents showing engagement is prioritized above all else. In August 2019, Facebook admitted its platform mechanics aren't neutral, stating that to make money, they must optimize for user interaction. The memo noted that "the more incendiary the material, the more it keeps users engaged, the more it is boosted by the algorithm." A 2018 study found false rumors spread faster and wider than true information, especially in politics.
YouTube
YouTube has been around since 2005 and now has more than 2.5 billion monthly users. The platform's recommendation system focuses on what you've watched, liked, or favorited to suggest new content. About 70% of what people see comes from that algorithm. A 2022 study by the Mozilla Foundation found users have little control over what appears in their suggestions, including videos about hate speech and livestreams. The site has been linked to spreading radicalized content, with groups like Al-Qaeda using it for recruitment. In a study published in the American Behavioral Scientist Journal, researchers looked into how the algorithm decides what to recommend and found that videos with radical keywords in their titles were more likely to be shown. In February 2023, a case called Gonzalez v. Google asked whether Google could be held liable for its algorithms promoting ISIS content. Section 230 generally protects platforms like YouTube from being sued over user-generated content. Studies have also found little evidence that the algorithm pushes far-right content to people who haven't already engaged with it.
TikTok
TikTok's algorithm pushes users toward engaging content, showing personalized feeds based on past behavior. This has led to more radical material appearing on users' "For You Pages." The platform has faced criticism for hosting misinformation and hate speech that drive high engagement. In 2022, TikTok said it removed hundreds of thousands of videos for breaking its rules. Studies show people can be drawn into extremism through app content. In early 2023, Austrian authorities stopped a plot against an LGBTQ+ pride parade involving teenagers inspired by jihadist videos on TikTok. One 14-year-old was influenced by Islamist content and took part in planning an attack. In 2024, more teens in Vienna were arrested for plotting a terrorist act at a Taylor Swift concert, with TikTok among platforms spreading extremist views to them.
Self-radicalization
Someone who carries out a terrorist act alone, without support or direction from a government or organized group, is called a "lone-wolf" terrorist. The U.S. Department of Justice uses this term. In recent times, this kind of terrorism has grown, tied to how online recommendation systems push people toward extreme ideas. Social media platforms create spaces where radical views are not only accepted but quickly adopted by others. These beliefs are strengthened through forums, group chats, and internet communities that reinforce each other. Such environments help individuals move toward more extreme positions, encouraging self-radicalization.
The Social Dilemma
The Social Dilemma is a 2020 docudrama that shows how social media algorithms can addict users and shape their thoughts, feelings, and actions. It highlights the spread of conspiracy theories and false information by focusing on psychological manipulation. The film uses terms like “echo chambers” and “fake news” to illustrate how people’s views become distorted. In the story, Ben gets caught in a cycle where the algorithm decides his page has a 62.3% chance of keeping him engaged, so it recommends more content. As he watches more videos, he becomes more immersed in propaganda and conspiracy theories, growing more extreme with each recommendation.
United States: Weakening Section 230 protections
In the Communications Decency Act, Section 230 says that online platforms can't be held like publishers for what users post. It shields them from lawsuits over third-party content, including illegal stuff. Critics say this protection lets companies push radical material for profit without fear of legal consequences. They argue it reduces incentives to remove harmful content. Proponents counter that before Section 230, courts had already ruled in Stratton Oakmont, Inc. v. Prodigy Services Co. that choosing what to moderate could make providers liable as publishers. In October 2021, House Democrats Anna Eshoo, Frank Pallone Jr., Mike Doyle, and Jan Schakowsky introduced the "Justice Against Malicious Algorithms Act" as H.R. 5596. The bill died in committee but would have stripped Section 230 for services using recommendation algorithms that knowingly or recklessly cause serious harm.
-
Read this chapter
Overview
Collective behavior describes social events that don’t follow normal rules or structures—what emerges suddenly and unexpectedly. It shows up in many forms, from the way birds fly in formation to the way people act during a riot or a passing trend. These actions often break typical societal expectations, and they’re shaped by group influence, pushing individuals to do things they wouldn’t normally consider. Whether destructive or harmless, this kind of behavior comes from the power of the crowd, not from established laws or traditions.
Defining the field
The study of collective behavior began with Franklin Henry Giddings and was taken up by scholars such as Robert Park, Ernest Burgess, Herbert Blumer, Ralph H. Turner, Lewis Killian, and Neil Smelser. Among them, Turner and Killian pioneered the use of photographs and motion pictures to provide visual proof of how crowds act, moving beyond the shaky reliance on eyewitness testimony that had come before. Their approach drew heavily on Herbert Blumer’s idea that social “forces” are not real forces at all. According to Blumer, individuals actively interpret the actions of others and then respond based on that interpretation.
Examples
Collective behavior spans moments both large and strange, from the stock market crashes of 1929 to the "phantom gasser" events in the mid 1940s, and from the hula-hoop craze of 1958 to the Los Angeles riot of 1992. Not all sociologists accept that these events belong to one field of study, but Blumer and Neil Smelser did, along with others who have shaped the discipline. This shared perspective shows how some of the most respected thinkers in sociology have found value in grouping such diverse occurrences together.
The crowd
Crowds are one of the few things scholars agree on when discussing collective behavior. Clark McPhail saw crowds and collective behavior as synonymous, and his empirical studies helped identify different types. Gustave LeBon studied the French Revolution and argued that crowds cause people to lose rational thought temporarily, a view later echoed by Sigmund Freud in Group Psychology and the Analysis of the Ego. LeBon's ideas were supported by events like the tulip mania in Holland in 1637, which Charles MacKay documented in Extraordinary Popular Delusions and the Madness of Crowds. At the University of Chicago, Robert Park and Herbert Blumer agreed crowds are emotional but can express any kind of emotion. Some scholars expanded the idea to include dispersed groups, calling them diffuse crowds—such as revivals or panics. Neil Smelser and John Lofland proposed three basic emotions—fear, joy, and anger—and matched each with a crowd type: panic, craze, and hostile outburst, each possible in compact or diffuse settings.
The public
When Boom talks about the difference between a crowd and a public, he’s not using the word “public” like we normally do. A crowd shows shared emotion, but a public is more specific—it’s a group that comes together to talk about one issue. Park and Blumer explain that each issue gives rise to its own public. A public only exists while people are discussing it, and it ends when they reach a decision. So the number of publics depends on how many issues people take up at once.
The mass
Blumer identified three types of collective behavior, including the crowd and the public, but he also introduced a third form: the mass. This type is different from the others because it’s not defined by how people interact with each other. Instead, it's shaped by the efforts of those using mass media to reach an audience. The first mass medium was printing.
The social movement
When we think about collective behavior, the social movement is different—more structured and enduring than other forms. Blumer outlines two types: active movements such as the French Revolution, which aim to transform society, and expressive ones like Alcoholics Anonymous, which focus on changing their members. Unlike more fluid kinds of collective behavior, social movements are less likely to shift quickly and can eventually become fixed institutions. The field gained momentum after the appearance of key works that brought the topic back into the spotlight for American sociologists. This renewed interest followed social unrest in the U.S. during the late 1960s and early '70s, which pushed scholars to test earlier theories against real-world events.
-
Read this chapter
Overview
Human-based computation, also known as human-assisted computation, ubiquitous human computing, or distributed thinking, is a method where computers solve problems by handing certain tasks over to people, often through microwork. This technique uses the different strengths and costs of humans and machines to create a partnership. It's especially important for tough tasks like image recognition, where it helps train artificial intelligence systems, and has been called human-aided artificial intelligence. In regular computing, a person gives a problem to a computer. But in human-based computation, the computer asks people to help solve a problem, then gathers and combines their answers. This creates large networks of humans and computers working together, where code runs partly in human minds and partly on silicon processors.
Early work
Human-based computation began with early work in interactive evolutionary computation, where humans guided algorithms using their own judgment. In Richard Dawkins’ The Blind Watchmaker (1986), the Biomorphs software used human preference to evolve shapes, making people the fitness function. Victor Johnston and Karl Sims later expanded this by gathering input from many users, enabling complex art and face designs. This approach reversed typical computer-human roles, with computers coordinating human efforts instead of directing them. Another precursor was Moni Naor’s automated Turing test from 1996, which used problems without efficient algorithmic solutions to distinguish humans from machines. Human-based genetic algorithms then allowed people to contribute creatively, not just evaluate, participating in all stages of evolution, even where no computational operators existed.
Classes of human-based computation
Kosorukoff introduced a way to classify human-based computation in 2000, organizing it into three main classes based on how humans and computers share tasks. In this model, one class uses humans for innovation and computers for selection; another switches those roles; and a third has both humans and computers contributing to both. The system uses two-letter codes—HC, CH, HH—to identify each type, with a lowercase h indicating limited human involvement in selection. These classifications come from an evolutionary computation model and focus on the roles each plays rather than who’s doing the work. This framework helps explain how recommendation systems shape what we see and think online.
Human-based computation as a form of social organization
Human-based computation works as flexible social organization more effective than traditional companies, adapting to human spontaneity while embracing creativity and mistakes. This approach lets participants feel free without sacrificing function, making people happier. Techniques scale better than older methods, allowing tasks to be spread across thousands of people easily. Some call this mass outsourcing "crowdsourcing," but others argue true human-based computation requires mixing humans and computers in the same system. Michelucci says just distributing tasks among many agents isn't enough—there must also be a blend of human and computer workers to qualify as human computation. Platforms like Mechanical Turk use APIs, task prices, and software protocols to connect employers with workers, but this often leads to automated management instead of personal attention, leaving workers frustrated.
Criticism
Human-based computation has drawn criticism for being exploitative and deceptive, with concerns that it weakens our ability to act together as a group. In social philosophy, it's been argued that this approach treats people as a kind of online labor. The philosopher Rainer Mühlhoff identifies five ways human effort gets captured within hybrid systems connecting people and computers. These include gamification, trapping and tracking like CAPTCHAs, social exploitation such as tagging faces on Facebook, information mining, and click-work found on platforms like Amazon Mechanical Turk. Mühlhoff notes that much of this human input feeds into AI systems, which he calls "human-aided artificial intelligence."
-
Read this chapter
Overview
Social media use in politics involves how online platforms shape political life, from local governance to global conflict. The Internet changed how news spreads, and social media has altered not just what we hear, but how political dynamics play out. Platforms like Facebook and Twitter have become central to civic engagement, even if they’re often biased. These tools affect elections and campaigns, with influencers and trolls spreading views and misinformation. For example, it was reported that Russia flooded American social media during the 2016 presidential election with fake news. Studies found that in the months before the vote, articles favoring Trump were shared 30 million times, compared to only 8 million for Clinton.
Participatory role
Social media let anyone with an internet connection become a content creator, giving people new ways to take part in political conversations. Some call this "new media populism," where citizens who were once left out can now join the discussion. Howard Rheingold said these tools challenge traditional power by letting people control communication, and scholar Derrick de Kerckhove described how online networks shift power from producers to users. Still, studies show most people are passive consumers, while only a few create content. In 2020, TikTok users tricked a Trump rally in Tulsa by buying tickets and not showing up. The U.S. government later banned TikTok on federal devices in 2025, raising questions about free speech and foreign platforms influencing domestic politics.
Polarization (Affective Polarization)
Recent analysis shows that digital discourse is shifting from people simply disagreeing over policies to actively disliking and distrusting those on the other side. Social media platforms play a role by favoring content that mocks or attacks opposing groups. This kind of media gets much more engagement than posts promoting one's own views. By 2025, studies suggest that shared reality among voters is less affected by lack of information and more shaped by how emotions are stirred up in the presentation of that information. This emotional feedback loop builds hostility online, which pushes out moderate voices and rewards extremism. Effective polarization has been tied to lower social trust and a growing belief that the other side threatens the nation. As media moves away from policy debate toward radicalized content driven by algorithms, it's changing the emotional tone of political life, making compromise harder.
As a news source
Social media has become a major source of political news in the U.S., especially during elections. A 2019 Pew Research study found that one in five adults get their political news primarily from social platforms, with 18% using them specifically for election information—48% of whom are aged 18 to 29. In 2022, a small study by McKeever et al. showed that 269 out of 510 participants said most of their gun violence news came from social media. Facebook, Twitter, and Reddit are the top platforms for news, with 66% of Facebook users, 59% of Twitter users, and 70% of Reddit users accessing news on those sites. According to the Reuters Institute Digital News Report in 2013, online news users who blog about news issues range from 1% to 5%, while social media use for commenting varies from 8% in Germany to 38% in Brazil. Still, most people mostly share stories or talk about them offline. The speed at which information spreads can hurt political reputations—like when Congressman Anthony Weiner's inappropriate tweets contributed to his resignation.
Role of political influencers
By 2026, digital news had shifted so that influencers were increasingly marketing themselves as political figures, a trend especially embraced by Generation Z. While platforms like Facebook and X still served as foundations for many, independent creators were building massive followings and gaining widespread support globally. These influencers wielded strong power over personal decisions, often shaping choices based on their branding rather than traditional institutional authority. Unlike conventional news outlets, they frequently merged their political views with their personal identities, leading to a wide range of information quality and messaging.
Attention economy
Social media functions within what Columbia Law School’s Tim Wu calls the “attention economy,” where content drawing more attention gets seen and shared widely. This system influences political views because people are exposed to more ideas online, and news use can shift opinions. Yet it also undermines trust—while reading newspapers boosts social trust, watching TV news weakens it. Younger generations are growing more politically involved through social media, but these platforms are filled with bias. In May 2016, a former Facebook Trending News curator named Benjamin Fearnow said his role was to “massage the algorithm,” though he denied any intentional bias from people or machines at the company. He was fired after leaking internal debates about Black Lives Matter and Donald Trump.
As a public utility
Social media platforms like X, Facebook, and Instagram let politicians and influencers connect with millions instantly, raising questions about whether these tools should be treated as public utilities. While they can be seen as an impure public good—non-rival but excludable because platforms control what users see—these systems are dominated by a few large companies such as Google and Facebook. These firms shape the digital environment based on profit goals rather than public dialogue. TikTok's algorithm, for instance, promotes engaging content that can go viral quickly, helping younger users share political messages on issues like healthcare, climate change, and racial justice during the 2020 U.S. election. The platform also brought global concerns such as protests in Hong Kong and the Israeli-Palestinian conflict to wider attention. As Zeynep Tufekci argues, online services are natural monopolies that risk privatizing our shared public spaces.
Government regulation
Regulation of social media is becoming more common as people worry about platforms like Facebook and Twitter acting like monopolies, and about privacy, censorship, and how information is stored. Some say the government should enforce “algorithmic neutrality,” letting search engines rank content without human input. But others argue that regulating these platforms like public utilities could hurt innovation and free speech, especially since they’re not like traditional utilities. Countries are also fighting misinformation and AI-generated content, with 13% of those holding federal elections having been targeted by hackers or political actors trying to influence voters. While social media helps governments connect with citizens and share important information, too much state control over political content could silence democratic debate. The challenge is balancing tech progress with ethical oversight to protect fair elections and open dialogue.
Read
Free to download, keep and share. For general information only — not professional medical, legal or financial advice. Please consult a qualified professional.