right or “anti-woke” content.40 Noting that YouTube is “under-studied relative to other social media platforms,” the Penn team scrutinized the browsing behavior, on and off YouTube, of more than 300,000 Americans from January 2016 through December 2019. Rather than recommendations heightening user engagement with far-right content, the Penn team concluded that “consumption of political content on YouTube appears to reflect individual preferences that extend across the web as a whole.”41 'Different realities' Whatever its precise measure, the reduction in conspiratorial recommendations “does not make the problem of radicalization on YouTube obsolete nor fictional,” Hany Farid, a Berkeley computer scientist, and his fellow researchers argued. “Aggregate data hide very different realities for individuals.”42 Radicalization may be relevant for only a small fraction of visitors to the site, but given the gargantuan size of YouTube’s user base, that fraction represents a sizable absolute number. What’s more, those individuals are the ones society should be most worried about. Researchers at NYU’s Center for Social Media and Politics scrutinized recommendations received by 760 Americans in the fall of 2020. After classifying the recommendations by ideology, they found no evidence that users on average were guided toward extreme content, defined as the rightor left-most 5% of videos. But in a manuscript pending publication, they offered several important qualifications: YouTube recommendations did nudge both Republicans and Democrats toward more conservative and more homogenous content, although this push was “very mild on average.” And the platform’s “conservative bias” increased as recommendations accumulated. “Our aggregate findings mask important and meaningful exceptions” to the tendency of YouTube recommendations to exert only a mild push to the ideological right, the NYU researchers said. “In a small fraction of cases,” they added, “users did ‘fall down a rabbit hole’ in which they were only shown relatively extreme recommendations for the duration of their time on the platform.” ‘To Maximize Engagement, You Maximize Outrage’ YouTube doesn’t intentionally spread racist hatred or anti-vaccine misinformation. To the contrary, it has rules banning such harmful material, and it uses automated and human content moderation to try to enforce its rules. But YouTube’s content moderation function resembles an ambivalent arsonist frantically trying to extinguish conflagrations that he himself helped start. The same is true of Facebook, Instagram, Twitter, and TikTok. Our Center has described this paradoxical situation in past reports, but it’s worth reiterating here. The platforms collect the lion’s share of their considerable revenue from advertisers seeking the attention of consumers. To demonstrate that users remain glued to their screens, platforms prioritize “engagement,” a metric reflecting user watch time and the volume of “likes,” shares, retweets, and comments. But here’s the problem: Content that tends to heighten engagement often does so because it provokes emotions like anger, fear, and resentment.1 So, when software engineers design algorithms to rank and recommend content, the automated systems favor posts that stir these negative emotions.2 Eric Schmidt, the former chief executive officer of Google, YouTube’s parent, and executive chairman of Alphabet, the holding company for both Google and YouTube, candidly explained the phenomenon during an interview on the Armchair Expert podcast in January 2022. “Corporations, at least in social media land, are optimizing, maximizing revenue,” he said. “You maximize revenue, you maximize engagement. To maximize engagement, you maximize outrage.”3 https://www.pnas.org/doi/10.1073/pnas.2024292118 ² https://science.sciencemag.org/content/370/6516/533 ³ https://armchairexpertpod.com/pods/eric-schmidt 1 8 HOW YOUTUBE SPREADS HARMFUL CONTENT – AND WHAT CAN BE DONE ABOUT IT