Why is "What Pretending To Be Crazy Looks Like" flooding YouTube recommendations?

Since the end of May, there has been one question on everybody’s mind: What does pretending to be crazy look like? It’s not a question that anyone asked, really—it’s a question that YouTube recommended to them. On May 26, 2021, YouTuber JCS – Criminal Psychology published the video “What Pretending To Be Crazy Looks Like.” The hour-long explainer of the tics, mannerisms, and tells that Parkland school shooter Nikolas Cruz exhibited during the interrogation that followed his arrest is at the top of many users’ recommended YouTube videos, regardless if they’re into this form of pseudo-criminal psychology or not. So why is YouTube recommending “What Pretending To Be Crazy Looks Like” to so many people?

The video’s prevalence is something of a mystery online. As Ryan Broderick wrote in the Garbage Day newsletter, it got “stuck in the YouTube algorithm.” Like many anomalies that break or highlight the mechanics of social networks, the video is now a meme, with frames of Cruz’s interrogation becoming an meme-worthy image macro remixed with other cultural references and jokes. The recognition of the image shows its power. Even if people haven’t seen the whole video, they probably recognize the thumbnail.

YouTube users who ignore the impulse to find out “what pretending to be crazy looks like” aren’t missing much. JCS – Criminal Psychology compares Cruz’s defense with convicted murderer Jerrod Murray, who, at age 18, shot and killed classmate Generro Sanchez in 2012 and later successfully pleaded insanity. The narrator picks up on how Murray handles himself during his police interrogation, noting how coldly and emotionlessly Murray confesses to the crime. JCS compares Murray’s disposition with the Parkland shooter, who displays his mental state through self-harm and claims of hearing voices. JCS picks apart Cruz’s defense and how the interrogating police officer pokes holes in it.

This bit of pseudoscience makes sense in a culture obsessed with true crime, and JCS knows how to exploit that. Other videos on the channel feature more explainers with clickbaity titles, like “Jennifer’s Solution,” “There’s Something About Casey…,” and “The Legend Of ‘Jeff’,” which thankfully points out who Jeff is in the thumbnail. JCS’s video titles have improved since the channel began producing these videos in 2019. Their first batch of video titles play variations of “The Case of [blank]” or “The [blank] Case of [blank].” The promise of answers—or the case of “Jennifer’s Solution,” the promise of a question—proves that the channel’s curator knows what they’re doing. JCS’ revamped video titles helped things along, but it doesn’t account for how this one video ended up in feeds; “pretending to be crazy” received twice as many views as its immediate predecessor, “Wrath Of Jodi.”

“There’s different things that could be happening,” says Christo Wilson, an associate professor in the Khoury College of Computer Sciences at Northeastern University. Wilson’s work focuses on “algorithm auditing,” in which he and his team attempt to reverse engineer algorithms to “understand what data they’re collecting, how that data is being used, and, ultimately, what these systems use to shape people’s experiences.”

Typically, Professor Wilson, who says that the school just started looking at YouTube in earnest, uses a browser extension that a survey company distributes to a representative demographic of Americans. Participants take a survey and install the browser extension on their computer. The extension then tracks their web usage. Much of what you assume goes into these systems finds their way into algorithms. “A lot of the recommendations are based just on your subscriptions, so if you’re subscribed to these channels you’re going to see more videos recommended from them,” Wilson says. But that doesn’t necessarily apply to “What Pretending To Be Crazy Looks Like” because, as Wilson puts it, “It’s being shown to everybody.”

Wilson says that in a case like this, he looks at “contextually relevant recommendations,” meaning that YouTube is examining at your recent YouTube history and comparing it with the rest of its users. It’s not just recommending true crime videos because true crime is widespread on the site; it’s doing so because of “social recommendations.” “YouTube is looking at all the people on the platform and comparing them. Am I similar to somebody else? Even if I am not watching true crime right now, if my interests, broadly speaking, overlap with yours to a significant degree, and you’re watching true crime, well, maybe they should recommend true crime to me as well.”

The video tends to get stuck, as Broderick put it, because the more people get recommended the video, the more people watch it, and the more people watch it, the more it gets recommended. But, as Wilson points out, one thing we have to be careful of is “survivorship bias,” or, in other words, just because we’re noticing this video, that doesn’t mean this phenomenon isn’t happening all the time. “There could actually be lots of videos like this that are being recommended to lots of the YouTube population, but we haven’t noticed because nobody made a meme about it.” Wilson says that this video just resonated with people on social media, and we collectively realized it, but the video might not be an anomaly. It could be happening all the time.

Regarding “What Pretending To Be Crazy Looks Like,” Wilson says that it’s a confluence of factors. True crime is an evergreen topic that appeals to many people, and the quality of the JCS videos keeps people engaged. “I could easily see the recommendation system thinking most people will be interested in this,” Wilson says. “Couple that with this particular channel having a very high production value. It’s a long-form video, and despite the content being total pseudoscience, it’s very well put together.” If the channel’s 3 million subscribers watch it and watch the whole thing, it sends a signal to YouTube that this is a good video. “It is in this category that has broad appeal, I could very easily see it then being put in front of millions of people’s eyeballs.”

“What Pretending To Be Crazy Looks Like…” isn’t so unique online, but it is hitting a lot of the requirements to become a grassroots viral video. But as Christo says, we may only notice it because the title is so grabbing and the content so salacious. What social benefits we glean from knowing what pretending to be crazy looks like is anyone’s guess. As we’ve seen over the five or six years, these systems aren’t very good at considering the ethical considerations of foisting videos on people. “I think it’s pretty concerning. Teaching pop psychology to people that they can dismiss someone who appears to be in mental health crisis because they’re ‘faking it,’ that’s not a particularly great message.”

“Boiling it all down to these trivial things, all the research tells us that you can’t actually tell when people are lying, and yet this is very much promoting this message of ‘You can go with your gut, right?’ You know when someone’s lying to you, right?” Understanding “what pretending to be crazy looks like” isn’t so clear-cut, even if the source is highly recommended.