Book excerpt: Manipulating likes, comments, shares, and follows

(This is an excerpt from drafts of my book, "Algorithms and Misinformation: Why Wisdom of the Crowds Failed the Internet and How to Fix It")

“The systems are phenomenally easy to game,” explained Stanford Internet Observatory’s Renee DiResta.

The fundamental idea behind the algorithms used by social media is that “popular content, as defined by the crowd” should rise to the top. But “the crowd doesn’t have to be real people.”

In fact, adversaries can get these algorithms to feature whatever content they want. The process is easy and cheap, just pretend to be many people: “Bots and sockpuppets can be used to manipulate conversations, or to create the illusion of a mass groundswell of grassroots activity, with minimal effort.”

Whatever they want — whether it is propaganda, scams, or just flooding-the-zone with disparate and conflicting misinformation — can appear to be popular, which trending, ranker, and recommender algorithms will then dutifully amplify.

“The content need not be true or accurate,” DiResta notes. All this requires is a well-motivated small group of individuals pretending to be many people. “Disinformation-campaign material is spread via mass coordinated action, supplemented by bot networks and sockpuppets (fake people).”

Bad actors can amplify propaganda on a massive scale, reaching millions, cheaply and easily, from anywhere in the world. “Anyone who can gather enough momentum from sharing, likes, retweets, and other message-amplification features can spread a message across the platforms’ large standing audiences for free,” DiResta continued in an article for Yale Review titled "Computational Propaganda": “Leveraging automated accounts or fake personas to spread a message and start it trending creates the illusion that large numbers of people feel a certain way about a topic. This is sometimes called ‘manufactured consensus’.”

Another name for it is astroturf. Astroturf is feigning popularity by using a fake crowd of shills. It's not authentic. Astroturf creates the illusion of popularity.

There are even businesses set up to provide the necessary shilling, hordes of fake people on social media available on demand to like, share, and promote whatever you may want. As described by Sarah Frier in the book No Filter: “If you searched [get Instagram followers] on Google, dozens of small faceless firms offered to make fame and riches more accessible, for a fee. For a few hundred dollars, you could buy thousands of followers, and even dictate exactly what these accounts were supposed to say in your comments.”

Sarah Frier described the process in more detail. “The spammers ... got shrewder, working to make their robots look more human, and in some cases paying networks of actual humans to like and comment for clients.” They found “dozens of firms” offering these services of “following and commenting” to make content falsely appear to be popular and thereby get free amplification by the platforms wisdom of the crowd algorithms. “It was quite easy to make more seemingly real people.”

In addition to creating fake people by the thousands, it is easy to find real people who are willing to be paid to shill, some of which would even “hand over the password credentials” for their account, allowing the propagandists to use their account to shill whenever they wished. For example, there were sites where bad actors could “purchase followers and increase engagement, like Kicksta, Instazood, and AiGrow. Many are still running today.” And in discussion groups, it was easy to recruit people who, for some compensation, “would quickly like and comment on the content.”

Bad actors manipulate likes, comments, shares, and follows because it works. When wisdom of the crowd algorithms look for what is popular, they pick up all these manipulated likes and shares, thinking they are real people acting independently. When the algorithms feature manipulated content, bad actors get what is effectively free advertising, the coveted top spots on the page, seen by millions of real people. This visibility, this amplification, can be used for many purposes, including foreign state-sponsored propaganda or scams trying to swindle.

Professor Fil Menczer studies misinformation and disinformation on social media. In our interview, he pointed out that it is not just wisdom of the crowd algorithms that fixate on popularity, but a “cognitive/social” vulnerability that “we tend to pay attention to items that appear popular … because we use the attention of other people as a signal of importance.”

Menczer explained: “It’s an instinct that has evolved for good reason: if we see everyone running we should run as well, even if we do not know why.” Generally, it does often work to look at what other people are doing. “We believe the crowd is wise, because we intrinsically assume the individuals in the crowd act independently, so that the probability of everyone being wrong is very low.”

But this is subject to manipulation, especially online on social media “because one entity can create the appearance of many people paying attention to some item by having inauthentic/coordinated accounts share that item.” That is, if a few people can pretend to be many people, they can create the appearance of a popular trend, and fool our instinct to follow the crowd.

To make matters worse, there often can be a vicious cycle where some people are manipulated by bad actors, and then their attention, their likes and shares, is “further amplified by algorithms.” Often, it is enough to merely start some shilled content trending, because “news feed ranking algorithms use popularity/engagement signals to determine what is interesting/engaging and then promote this content by ranking it higher on people’s feeds.”

Adversaries manipulating the algorithms can be clever and patient, sometimes building up their controlled accounts over a long period of time. One low cost method of making a fake account look real and useful is to steal viral content and share it as your own.

In an article titled “Those Cute Cats Online? They Help Spread Misinformation,” New York Times reporters described one method of how new accounts manage to quickly gain large numbers of followers. The technique involves reposting popular content, such as memes that previously went viral, or cute pictures of animals: “Sometimes, following a feed of cute animals on Facebook unknowingly signs [people] up” for misinformation. “Engagement bait helped misinformation actors generate clicks on their pages, which can make them more prominent in users’ feeds in the future.”

Controlling many seemingly real accounts, especially accounts that have real people following them to see memes and cute pictures of animals, allows bad actors to “act in a coordinated fashion to increase influence.” The goal, according to researchers at Indiana University, is to create a network of controlled shills, many of which might be unwitting human participants, that are “highly coordinated, persistent, homogeneous, and fully focused on amplifying” scams and propaganda.

This is not costless for social media companies. Not only are people directly misled, and even sometimes pulled into conspiracy theories and scams, but amplifying manipulated content including propaganda rather than genuinely popular content will “negatively affect the online experience of ordinary social media users” and “lower the overall quality of information” on the website. Degradation of the quality of the experience can be hard for companies to see, only eventually showing up in poor retention and user growth when customers get fed up and leave in disgust.

Allowing fake accounts, manipulation of likes and shares, and shilling of scams and propaganda may hurt the business in the long-term, but, in the short-term, it can mean advertising revenue. As Karen Hao reported in MIT Technology Review, “Facebook isn’t just amplifying misinformation. The company is also funding it.” While some adversaries manipulate wisdom of the crowd algorithms in order to push propaganda, some bad actors are in it for the money.

Social media companies allowing this type of manipulation does generate revenue, but it also reduces the quality of the experience, filling the site with unoriginal content, republished memes, and scams. Hao detailed how it works: “Financially motivated spammers are agnostic about the content they publish. They go wherever the clicks and money are, letting Facebook’s news feed algorithm dictate which topics they’ll cover next ... On an average day, a financially motivated clickbait site might be populated with ... predominantly plagiarized ... celebrity news, cute animals, or highly emotional stories—all reliable drivers of traffic. Then, when political turmoil strikes, they drift toward hyperpartisan news, misinformation, and outrage bait because it gets more engagement ... For clickbait farms, getting into the monetization programs is the first step, but how much they cash in depends on how far Facebook’s content-recommendation systems boost their articles.”

The problem is that this works. Adversaries have a strong incentive to manipulate social media’s algorithms if it is easy and profitable.

But “they would not thrive, nor would they plagiarize such damaging content, if their shady tactics didn’t do so well on the platform,” Hao wrote. “One possible way Facebook could do this: by using what’s known as a graph-based authority measure to rank content. This would amplify higher-quality pages like news and media and diminish lower-quality pages like clickbait, reversing the current trend.” The idea is simple, that authoritative, trustworthy sources should be amplified more than untrustworthy or spammy sources.

Broadly this type of manipulation is spam, much like spam that technology companies have dealt with for years in email and on the Web. If social media spam was not cost-effective, it would not exist. Like with web spam and email spam, the key with social media spam is to make it less effective and less efficient. As Hao suggested, manipulating wisdom of the crowd algorithms could be made to be less profitable by viewing likes and shares from less trustworthy accounts with considerable skepticism. If the algorithms did not amplify this content as much, it would be much less lucrative to spammers.

Inside of Facebook, data scientists proposed something similar. Billy Perrigo at Time magazine reported that Facebook “employees had discovered that pages that spread unoriginal content, like stolen memes that they’d seen go viral elsewhere, contributed to just 19% of page-related views on the platform but 64% of misinformation views.” Facebook data scientists “proposed downranking these pages in News Feed ... The plan to downrank these pages had few visible downsides ... [and] could prevent all kinds of high-profile missteps.”

What the algorithms show is important. The algorithms can amplify a wide range of interesting and useful content that enhances discovery and keeps people on the platform.

Or the algorithms can amplify manipulated content, including hate speech, spam, scams, and misinformation. That might make people click now in outrage, or perhaps fool them for a while, but will cause people to leave in disguist eventually.