The Chaos Machine: The Inside Story of How Social Media Rewired Our Minds and Our World

In late 2018, New York Times reporter Max Fisher visited Facebook's headquarters carrying more than 1,400 pages of internal documents leaked by a contract moderator he calls Jacob. Working at an overseas outsourcing firm, Jacob had watched posts grow steadily more hateful and conspiratorial, observing that the recommendation systems appeared to promote the most incendiary content most widely. When he sent his concerns up the corporate chain, headquarters never responded. Fisher had spent years as a foreign correspondent linking extreme events around the world back to social media. The book follows the researchers, whistleblowers, and concerned citizens who set out to answer a question the platforms refused to confront: What are the consequences of routing human life through systems designed to maximize engagement?

Fisher traces how platforms exploit human psychology. In 2014, Renée DiResta, a tech investor and former intelligence analyst in Silicon Valley, joined online parenting groups and encountered fierce anti-vaccine sentiment she rarely saw offline. When she created a pro-vaccine Facebook group, the platform returned overwhelmingly anti-vaccine results, and joining a single anti-vaccine group triggered a cascade of recommendations for increasingly extreme content. DiResta concluded that engagement-maximizing algorithms favored anti-vaccine content because a mother consumed by conspiracy would spend far more time online than one who accepted medical consensus. Facebook was not merely hosting extremists; it was creating them.

Fisher traces Silicon Valley's insular culture to the military-industrial buildup in the Santa Clara Valley after World War II and a venture-capital model in which each generation of founders replicated its predecessors' biases. The archetype of the ruthless, antisocial geek founder was eventually built into the social media products governing billions of users' daily lives.

Fisher identifies Facebook's 2006 news feed as pivotal. When users revolted, protest groups proliferated, but the outrage was itself amplified by the feed: Each click to join broadcast to all of a user's friends, creating an illusion of overwhelming consensus. Facebook recognized the distortion but embraced it because it drove engagement upward.

Sean Parker, Facebook's first president, later described the company's strategy as deliberately giving users intermittent dopamine hits through likes and comments, a feedback loop that he and Mark Zuckerberg, Facebook's cofounder and CEO, understood from the beginning. The "Like" button, launched in 2009, taps into the sociometer, a psychological gauge through which humans monitor social standing. Fisher argues that identity is the platforms' most powerful force, drawing on social psychologist Henri Tajfel's social identity theory to show that platforms surface tribalistic instincts by turning every interaction into a social act and elevating whatever sentiments win the most engagement.

Myanmar serves as a case study of what happens when an entire society goes online at once. Between 2012 and 2015, internet adoption exploded from 0.5 to 40 percent, mostly through smartphones preloaded with Facebook, which for many citizens became synonymous with the internet. The extremist Buddhist monk Wirathu used the platform to reach much of the country with fabricated accusations against the Muslim minority. A Stanford researcher warned Facebook in late 2013 that hate speech was overrunning the platform, but the company employed only one Burmese-language moderator. In mid-2014, Wirathu shared a fabricated rape accusation that went viral; hundreds rioted in Mandalay, killing two. By August 2017, the military launched a genocide against the Rohingya, a Muslim minority of 1.5 million. The United Nations later concluded that Facebook substantively contributed to the genocide.

In the United States, Fisher traces how the social web's extremes entered mainstream life through Gamergate, a 2014 harassment campaign targeting women in the video game industry that originated on 4chan, an anonymous message board. Neo-Nazi leader Andrew Anglin recognized that social media had cultivated a willing audience for white nationalism among American youth. Milo Yiannopoulos, a Breitbart columnist, channeled Gamergate's energy into the "alt right." A Harvard study found that Breitbart became the most popular right-wing news source on Facebook from May 2015 to November 2016, with the platform's algorithms promoting hyperpartisan outlets into major publishers.

Guillaume Chaslot, a French artificial intelligence specialist, helped build a machine-learning algorithm for YouTube's recommendation system that proved enormously effective at increasing watch time. YouTube cited poor performance when it fired Chaslot in October 2013; Chaslot believed the firing was retaliation for trying to build a safer alternative. He later created Algo Transparency to track recommendations at scale and found that 85 percent of recommended videos on Pope Francis were conspiracies and 70 percent on global warming called it a hoax. YouTube's 2016 shift to deep learning, a method in which automated systems assume control of recommendation decisions with minimal human oversight, made the algorithm's choices opaque to engineers.

Fisher documents the 2016 election as a turning point. Pizzagate, the conspiracy alleging Democratic leaders operated a child sex ring, spread across platforms; 14 percent of Trump supporters believed it. After Trump's victory, Zuckerberg dismissed the idea that misinformation influenced the result. DiResta later analyzed 400 gigabytes of data on Russian influence operations and found that the agents had not invented their strategy of whipping up identity-based outrage but had discovered it in the platforms' own incentives. Psychologist William Brady and Yale neuroscientist Molly Crockett synthesize their findings into the MAD model (Motivation, Attention, Design), concluding that platforms make moral-emotional content progressively dominant and that users rewarded for outrage internalize the compulsion.

Fisher and colleague Amanda Taub traveled to Sri Lanka in early 2018, where a viral Facebook rumor about a Muslim pharmacist hoarding sterilization pills triggered three days of mob violence against Muslim homes, businesses, and mosques. Abdul Basith, a Muslim resident trapped in his burning home, was found dead. The government blocked social media, and the violence stopped almost immediately. In Germany, researchers Karsten Müller and Carlo Schwarz found that towns with higher Facebook use experienced about 35 percent more anti-refugee attacks, a figure that dropped by the same rate whenever internet access went down.

German media scholar Jonas Kaiser and Swiss researcher Adrian Rauchfleisch mapped YouTube's recommendation networks and found that the algorithm stitched mainstream-right channels, conspiracy theorists, and white nationalists into a unified supercluster. QAnon, a conspiracy movement claiming a secret cabal controls the government, spread through these systems into a movement with millions of adherents. In March 2019, 28-year-old Australian white supremacist Brenton Tarrant livestreamed the murder of 50 people at two mosques in Christchurch, New Zealand; investigators concluded YouTube was his most significant source of radicalization. In Brazil, YouTube's algorithm helped elect far-right president Jair Bolsonaro in 2018.

Fisher examines Facebook's governance: thousands of outsourced moderators enforcing sprawling, contradictory rulebooks with about 8 to 10 seconds per decision. Facebook's courtship of Republicans, led by conservative lobbyist Joel Kaplan, resulted in the company shelving research on algorithmic divisiveness, allowing politicians to lie on the platform, and secretly censoring government critics in Vietnam at the dictatorship's request.

The Covid-19 pandemic brought these forces to a head. Facebook researchers found that disabling the algorithmic boost for serial reshares, repeated reposting chains that amplify the same content, would curb misinformation by up to 38 percent, but Zuckerberg refused because the change would reduce engagement. The algorithms bound together anti-vaxxers, QAnon adherents, and self-described militias into a unified movement. Throughout 2020, researchers identified at least 60,000 Facebook posts invoking political violence. After Joe Biden's election, the "Big Lie" that Trump had won saturated social media. On January 6, 2021, thousands stormed the U.S. Capitol in an insurrection planned on social media. Five people died.

In September 2021, Facebook data scientist Frances Haugen copied thousands of internal documents and sent them to the Securities and Exchange Commission (SEC), Congress, and the Wall Street Journal, revealing that executives knew their algorithms promoted divisive content, repeatedly overruled fixes, and lied to regulators. Haugen argued that the single most effective reform would be to turn off the algorithm. Fisher notes that the longer someone studies social media, the likelier they are to converge on this conclusion. He compares the situation to HAL 9000 in 2001: A Space Odyssey, a machine entrusted with power over human welfare whose incentives diverge from those of the people it serves, and whose creators assume good intentions are sufficient to guarantee benevolent outcomes.