How Bots Took Over Our Lives
All the way back in 2016, when Donald Trump won his first Presidential election, many people had trouble understanding how, exactly, it could have happened. The line of inquisition extended to the political class, where recriminations from talking heads on MSNBC about Hillary Clinton’s ineffective campaigning and tone-deaf messaging abounded. Others, though, gravitated toward a more sinister conspiracy, which involved foreign interference and a network of automated agitators spreading disinformation at scale. The bots, this emerging coalition said, were to blame for Trump’s rise.
During the past decade, so-called bots, a slippery concept from the start, have gone from a conspiratorial curiosity to a familiar force in our daily lives. In the popular imagination, they are suddenly everywhere and seem designed to annoy the hell out of people. Consider the familiar indignity of buying concert tickets only to find that, immediately after a sale opens, every ticket has been purchased by an army of bots, only to be listed on a resale site with a markup of hundreds or even thousands of dollars. This is their function: bots operate on a scale and speed that gives them an unfair advantage over humans, who have to fumble around with mouse clicks and thoughts. They hoard inventory on e-commerce sites to frustrate customers and torture retailers, and scan real-estate websites so they can send automated text messages to homeowners with lowball offers. They trawl the internet to scrape e-mail addresses and then fill inboxes with phishing attempts and spam.
But the impulse to treat bots like digital mosquitoes that need to be swatted away minimizes their role in making the internet work. At their most basic, bots are bits of programming built to run a certain function automatically until something or someone stops them. They are tools, and as much as general attitudes toward them have become negative they are most often used for mundane tasks. Still, our reflexive reaction to bots on social media is irritation and suspicion—to call something a “bot” is to label it as unthinking and nonhuman. That instinct partly originated in the blame game following Clinton’s narrow loss in the 2016 election, with the rise of a general if somewhat vague consensus that a web of coördinated political agitators based in Russia had used social-networking sites like Facebook and Twitter to sow disorder and distrust in the American electorate.
The accusation wasn’t without merit. Following Trump’s victory, the Internet Research Agency, a St. Petersburg-based “troll farm,” was accused of interfering in American elections, a detail revealed in the Mueller report. According to the report, the I.R.A. had been planning this digital-agitprop campaign for several years prior to the election, even travelling to the U.S. to take photos to use in their posts. They deployed bots to pose as American grassroots political organizers, eventually making virtual contact with Trump supporters and campaign officials, and to spread bogus headlines targeted at voters in swing states. According to the report, the I.R.A.’s strategy evolved from a generalized attempt to destabilize American politics to, per the report, “a targeted operation that by early 2016 favored candidate Trump and disparaged candidate Clinton.”
This meant that the disappointed liberals who had grown convinced that Russia tilted the election toward Trump were right—or at least half right. The degree to which the Russian efforts were effective is still an open question, and some reports have cast doubt on social media’s ability to drive political outcomes. A study published in 2023 by N.Y.U.’s Center for Social Media and Politics, in Nature Communications, found that Russian bots on Twitter had limited influence, with one per cent of users being exposed to seventy per cent of the posts created by Russian troll accounts. There was also a study conducted by professors at Dartmouth, Princeton, and the University of Exeter in 2017 which found that the majority of “fake news” promoted on sites like Facebook was being consumed by hard-line conservatives who likely would have voted for Trump anyway. But it didn’t matter. The Mueller report and the spectre of Russian influence provided an easy answer for a stunning political defeat, and the modern mythology of the bots was born.
In the years since 2016, bot networks have become only more sophisticated, and the proliferation of A.I. technology promises to widen their scope. Analysis from the internet-infrastructure giant Cloudflare estimates that nearly sixty per cent of internet traffic is nonhuman. In practice, that means that many exchanges on the internet are just bots talking to other bots in an automated loop. Years ago, as virtual assistants were first taking off, a video recording of what happened when a human asked two Alexa-equipped devices at the same time to start a psychotherapy session. They end up in an infinite recursion of questions without making any progress. Now imagine that process scaled up to the size of the internet, and you have an approximation of the bot-saturated social-media platforms. Researchers at one A.I. company estimated that of a sample of more than a million accounts on X, sixty-four per cent could be bots, a finding that requires no stretch of the imagination, given the sheer inanity of interactions on the site. Under any number of viral posts, you can find accounts responding with a sequence of increasingly brainless rehashings of the original content, like a series of photocopies that slowly deteriorate in quality.
The quantity of bots on Twitter was a key sticking point for Elon Musk during his purchase of the company in 2022, and his subsequent, unsuccessful attempt to back out of the deal. He estimated that twenty per cent of the site was made up of automated accounts, a dynamic that had overvalued Twitter and should therefore allow him to walk away from his hasty acquisition. Twitter, at that point, claimed that the ratio was closer to five per cent. But bots aren’t just sitting around talking to each other; they are also a major factor in what posts become popular, facilitating virality by clustering around certain subjects and liking and reposting specific posts, creating a flywheel of engagement that spreads whatever message someone might be amplifying.
This strategy of manufacturing a froth of interest around a given subject, from Nicki Minaj’s political perspectives to Sydney Sweeney’s jean ads, has recently been especially effective, and has moved culture-war fights from the trenches of bot-infested social networks into more mainstream channels. A New York Times investigation into the social media fracas found that there were relatively few mentions of Sweeney on X until several days after the début of her infamous American Eagle commercial, and less than ten per cent of those early posts were directly critical. It wasn’t until a couple of posts that compared the ad to Nazi eugenics propaganda ended up going viral that the outrage cycle began, and conservative influencers attempted to paint what was essentially a fringe perspective on the ad campaign as a liberal consensus.
“These episodes follow a common pattern: a small incident is framed as ‘proof’ of a broader culture war, then a mix of real users, coördinated accounts, and bots push the same talking points and narratives until it looks like a widespread response,” Emilio Ferrara, a computer-science professor at U.S.C., told me. Extreme voices with high follower counts, such as Chaya Raichik, the woman behind the right-wing X account Libs of TikTok, are usually at the center of the skirmish, and then bot networks swarm to boost the message. It’s like pouring gasoline onto a few embers, which then appears to be a bonfire. Lynnette Ng, a former bot-network researcher at Carnegie Mellon University, explained why the swarm is so effective: “Bots frequently and aggressively hit cognitive-bias triggers in their messages which push the brain to short-circuit people’s ability to reason about the information in front of them.” It doesn’t help that people struggle to decipher bots from humans in the first place during political conversations. The default heat setting for political discourse online is already so high that it tends to pressurize such conversations into a single, highly combustible layer. Bots that are designed to inflame those political disagreements are trained on those conversations and then injected back into the discourse, amplifying the feedback loop of people yelling at bots who are yelling at people, and so on.
This is easy enough to understand; bots are like automated moths to digital flame. But the rationale propelling them is a little harder to decipher. Surely, some of the motive is financial, as engagement on social-media platforms has been monetized. On X, for example, an account is able to participate in X’s revenue-sharing program as long as it is “verified”—which really just means that the owner has paid a small fee for a premium account—and the page meets minimum metrics such as having five hundred verified followers and a certain number of recent “organic” impressions. (“Organic” here means that you didn’t pay for your posts to be promoted like an ad.) There’s no quality incentive here; as long as a post is attracting eyeballs and generating ad impressions, there is a potential for payout. A bot that is programmed to maximize engagement and to interact with accounts that have lots of followers can make money for its operator by appearing next to ads in the X timeline. A.I. has made that feedback loop spin only faster: bots now learn what sort of content drives the most engagement and adjust their outputs to generate likes, reposts, and comments.
Unfortunately, the posts that get the most engagement are invariably the ones that upset us the most. Rage-baiting, as this strategy is often called, is an effective behavior to program into a bot’s temperament, since people are more likely to engage with social-media posts that challenge their beliefs than those that confirm them. Maintaining a system powered by invective discourse also seems to be a practical business model for social-media platforms. A group of researchers found that removing toxic posts from a user’s feed reduced the amount of time they spent on a given platform. The total time that users spend mindlessly scrolling is the lifeblood of social-media outlets, meaning, as Ferrara explained, that they are dealing with a “conflicting system of incentives,” since bots are good for engagement metrics that can be attractive to advertisers but also mar user experience past a certain threshold. Keeping things just toxic enough so that users scroll without getting too upset at the state of the world is the sweet spot for platforms. It’s cynical, but it’s also very good business.
The platform engineers who keep people scrolling are only one cog in the attention-capturing machine. The people who create and deploy bot networks are pulling another set of levers in their attempt to make money or influence public opinion, and conflict is very good at attracting eyeballs. Many bots on social media are trained to maximize engagement, and the number of likes, shares, or comments a post receives is a large component of how the algorithms on X, Instagram, and TikTok decide what to put in front of users next. Those two parties—the platform owners who tweak the algorithm’s currents and the bot operators who ride it—work in tandem to shape people’s experiences online and end up warping reality in the process. There’s a growing weariness about the bot-dominated internet, though. Perhaps it explains why so many people were upset when it was revealed that Geese, a band beloved by the exact kinds of people who consider their music taste to be unassailably authentic, had reportedly worked with a marketing company to create legions of fake fan accounts. This tactic is not a major deviation from decades of creative and manipulative music marketing, but coverage of the controversy was overwhelmingly negative. Critics accused the band’s label of waging a “PsyOp,” one in which bots were capable of building an alternate reality and tricking them into liking something, almost against their will.
Ng compared the mechanics behind the glut of bots inflaming culture-war topics to an automated propaganda war that is currently being won by the individuals and groups more willing to use available technologies to pursue their own ends. “They’re an interesting tool for automated advertising that can be used by people to do good and do harm. Instead of us focussing on bots that are doing harm, we should focus on how we can use the same technology to make them do good.” It’s a pragmatic solution, if a little depressing, to give over the rest of the internet to automated skirmishes against disinformation.
Personally, I’d rather let the bots fight one another and log off forever. And yet the bot revolution is coming—or is perhaps already here—whether we like it or not. It goes far beyond political discourse, as bots now dominate how we experience life online: the customer-support chatbots that pop up when you want to rebook a flight, A.I.-powered summaries at the top of Google searches, automatic censors that flag and remove explicit content. But what happens when bots are all that’s left of digital life? Such a reality doesn’t seem terribly far away. Some in Silicon Valley would hand their entire lives over to “agents,” a dressed-up term for bots which has become one of the industry’s favorite words. Agents, though, are conceived of as independent and capable of reasoning, a spin along the lines of “bots with thoughts.”
In February, a security researcher at Meta named Summer Yue created an A.I. agent using the platform OpenClaw that she tasked with managing her Gmail account and suggesting which messages she should delete. Before she knew it, the bot started deleting all the e-mails in her inbox, like a digital locust. “I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb,” she wrote on X. That a security researcher for Meta was comfortable handing over the keys to her e-mail account to an autonomous bot that quickly went feral was funny, but after reading Yue’s story I found myself staring at my inbox. It was Sunday afternoon, and I had just got back from a weekend away. The number of unread e-mails in my inbox hovered around eighty thousand, and I found myself fantasizing about turning my whole life over to the bots and letting an agent nuke my overflowing inbox. It sounded liberating. Perhaps that’s what the A.I. apocalypse is really shaping up to be—not the elimination of humanity by an omniscient machine mind but an internet overrun by unthinking, error-prone automatons that make the web unusable and destroy it from the inside. After the bots take over, we can all get back to the real world. ♦