Feedback isn’t just the grit-your-teeth, high-pitched squeal you get when you put a microphone too close to a speaker, though that’s the form many of us are most familiar with. Music is what you hear at a concert when feedback is under control. The squeal is out-of-control feedback.
To an electrical engineer, however, feedback systems are filters and amplifiers.
Knowing how that control happens, and how it fails, will help us to understand the breakdown of social media.
Here’s what’s happening when you hear that squeal. A microphone picks up the sound waves that reach it. That sound is amplified and comes out of the speaker system. Feedback happens when the sound coming out of the speakers is added to the other sounds already present at the microphone.
One set of frequencies (those that arrive at the microphone in phase) is amplified at the expense of others, drowning out everything else. That’s why when you hear feedback, you only hear a single pitch, not an entire soundscape.
And that’s why the pitch changes if you move the microphone. The feedback system is a filter: it amplifies one set of frequencies; and everything else disappears. There’s no music, just a spine-chilling scream.
There’s an important lesson here. By controlling the properties of the feedback loop—in this case, the distance between the speaker and the microphone—you can control the system’s output. It takes very little energy: the feedback loop doesn’t power the amplifier. There’s no obvious point of control. The feedback loop commandeers the sound system and the engineers, with all their knobs and dials, have to fight to bring the system back under control. Until they succeed, you just have to sit back and listen to the noise.
Network propaganda works in much the same way as microphones and music. Propaganda isn’t a new phenomenon. It has existed for ages, probably going back to the dawn of consciousness—certainly to the dawn of political competition.
Only now the mechanics, the mechanisms used to spread disinformation, have changed.
One key to understanding these changes is to see propaganda as a feedback loop—not simply one-way propagation. Feedback is subtle. It isn’t regulated by a central authority. It can be controlled from the outside with a gentle touch. And once started, it can maintain itself. Control the feedback, and you control the news. You’re like the man behind the curtain in The Wizard of Oz: Nobody pays attention—unless a plucky heroine and a cowardly lion (et al) get involved.
We saw those feedback loops in 2016, we’ve been seeing them ever since, and we will certainly see them used in 2020. Seed a group with an opinion, use sockpuppet accounts to amplify responses, and sooner or later, you have a full-fledged feedback loop, living on its own knee-jerk reflexes. Clickbait titles that encourage people to repost without actually reading the content help to get the cycle started. As posts and re-posts arrive faster and faster, the squeal of feedback rises, and it becomes impossible to pay attention to anything else.
For the last few years, we’ve been trapped in a nightmare of feedback loops, where abuse and fake news feed on more abuse and fake news to produce noise. The noise can be targeted precisely at people, at issues, an at organizations. It quickly makes rational discussion impossible; that’s its point.
This process has been weaponized by bad actors, both domestic and international, ranging from teenagers in Macedonia to informally organized networks of “patriots” whose government ties can always be denied (a phenomenon discussed a decade ago in Jeffrey Carr’s Inside Cyber Warfare), to professional groups like Russia’s Internet Research Agency that can mount very sophisticated attacks.
The content of the howl isn’t even important, except to the extent that it destroys the possibility for normal conversation. We’ve seen feedback loops built around every ideology on the right and on the left. To the bad actor feeding the loop, it isn’t the thought that counts; the only important thing is that thought can no longer take place.
Participants in the conversation can either leave, or they can be paralyzed.
While many of our systems are vulnerable to attack, social networks have proven the most vulnerable. There’s no sense in attacking a hard target when a weaker one is available. Attacks on social networks can fly under the radar as “free speech,” even when the speaker is a bot that’s been programmed to manipulate a feedback loop.
The question, then, is how are we to build systems that are resilient against attack, systems in which feedback loops don’t form?
Don’t feed the trolls
There are several solutions for managing feedback in electronics. You can change the properties of the feedback path itself, disrupting the loop before it starts. Sometimes simply moving a microphone changes the path enough to prevent feedback. Or you can introduce “negative feedback” to cancel out the feedback signal. It’s important that neither approach represses feedback frequencies entirely; they aren’t holes in the system’s frequency response. Your goal is to allow all voices to be heard.
Perhaps a more useful way to frame the question is: How do we prevent feedback in our social networks without repressing certain conversations over others?
Many of us have been “free speech absolutists,” and the notion of repressing speech is painful, to say the least. But we’ve also learned that too much speech is a way of repressing speech. When the sound system is howling in a feedback loop, you can’t hear anything, and you can’t say anything. You can’t hear the performers, you can’t even hear your noisy neighbor. In the world of social media especially, preserving the right and ability to pay attention is more valuable than preserving some abstract notion of “speech.” Speech is meaningless when no one can listen.
For the most part, we don’t have a politburo that tells state-sponsored news agencies what news is fit to print. What we do have are systems that use feedback to control what people are exposed to and that shape how they are able to respond.
We already have some tools for attenuating feedback in our social networks. Speech doesn’t entail the right to amplification. Bots and sockpuppets are old by now and may have passed peak usefulness, but getting them off social networks is an important first step.
Limiting likes, retweets, and similar tools would also be useful. Instagram is experimenting with eliminating likes, saying users should focus on the videos and pictures, not on peer approval. On Twitter, I use retweets as much as anyone, but I’d gladly sacrifice that to create room for others to think. Structuring social media to require participating in discussions, rather than simply reflecting and amplifying others’ opinions, would be progress.
“Don’t feed the trolls” is advice that goes back to Usenet and the 1980s, but it’s still applicable here: Responding might feel satisfying, but it accomplishes nothing besides helping that feedback loop to start. And once it starts, it’s very hard to stop.
Defending people who are victimized by the feedback loop is helpful and necessary, but these responses have to be careful. Even mentioning trolls does nothing but encourage them. Research on Usenet has shown that participants can be classified as questioners, helpers, trolls, and flamers based merely on their posting patterns.
De-platforming people whose behavior is unacceptable is an age-old form of feedback control; Ages ago, it was called shunning or exile. Denying abusers a voice: It isn’t pleasant, but it is necessary.
Alternative venues, such as The Great American Debate, are another way of defusing the feedback loop. The Great American Debate is an experiment in providing a platform for reasoned, fact-based, non-partisan political discussion. The goal is to provide maps of important issues, starting with climate change. These maps will present all aspects of all sides the discussion, making it impossible to listen to one voice without hearing the others. If a feedback loop is a filter that eliminates all voices but one from the discussion, this platform is an anti-filter that disarms the loop before it starts.
Projects like Scuttlebutt attempt to reinvent the online experience as a safe space. If our current social media won’t adapt to manage feedback, it’s time to invent new media. That’s not an easy task, but 15 years ago, nobody thought that MySpace and Orkut would be replaced by Facebook, Twitter, and LinkedIn. It can certainly happen again.
Maximizing engagement to drive advertising revenue is at the heart of the feedback problem, as media expert Renée DiResta has pointed out. It is the mechanism that makes our media so easy to subvert. Advertising has always been a key part of the feedback loop; before the invention of social media, it was the only way for media output, modulated by customer response, to become input for the channels that create new media.
Congratulations, you’re engaged!
Then, as now, maximizing revenue is the only metric for which media channels are held accountable, and the result is predictable. The engagement metric has become a tool for radicalizing audiences: It leads to content recommendations that are increasingly extreme, just to keep viewers watching, their adrenaline pumping, and their engagement numbers up.
The problem with engagement is that gaming the system is simple: inject some articles or videos with clickbait titles, create some bots to read or watch the content and make it trend, and sit back as the social networks drive the feedback loop via suggested content.
Chris Wiggins of Columbia University’s Data Science Institute argues that organizations need to take ethical principles into account when formulating their key performance metrics. That isn’t a recommendation that will be welcome to corporations or to their investors; the notion that management is responsible only for maximizing short-term shareholder value is a powerful and convenient way to avoid responsibility. Business models based on subscription, rather than advertising, relieve some (though not all) of the pressure.
Deep fakes are the newest tool for disrupting conversation and starting feedback loops, as we’ve already seen with a doctored video of House speaker Nancy Pelosi. Some progress has been made on detecting deep fakes, but this conflict will certainly escalate, and it’s fair to say that the fakers have the advantage. The significance of the Pelosi video isn’t the technology—it’s at best a shallow fake, easily detectable—but that it started the social media feedback loop.
DiResta suggests that fake news (including fake videos) can be handled using the same methods that have successfully tamed spam email: “Nobody mourns the free speech rights of spammers.”
Non-technical solutions to the problem of fakes certainly involve framing. I’ve seen attempts to defend the Pelosi video as parody or satire. Growing up during the Nixon administration, I saw many TV comedians and satirists say “My Fellow Americans…” I doubt anyone was tricked into thinking they were seeing Nixon on the screen. There was always a frame that made it clear who we were seeing. When Nixon actually made a guest appearance on Dan Rowan’s and Dick Martin’s Laugh-In, we knew it.
Framing isn’t entirely reliable; a quick Google search turns up many cases where un-doctored images were distorted by a misleading caption. But becoming more aware of how an image is framed, and requiring some kind of framing that shows who posted the image and why, is a start. If nothing else, framing requires the attacker to keep the frame in synch with the fake video and the viewer’s expectations. A fake video requires a fake frame that must align properly with the video. It gives the attacker another chance to slip up.
We require “full disclosure” from reporters, essayists, and journalists. We should also require it from our images and videos. Likewise, as consumers we need to distrust media that lacks framing, or where the frame doesn’t match the content.
The struggle is not about free speech; it’s about the right to pay attention, and to think. If nothing else, this is why the feedback metaphor is appropriate. Feedback isn’t helping you to hear the music at a concert; it’s destroying the ability to hear by amplifying one frequency at the expense of all the others.
And make no mistake about this: In our world of social media, it’s not accidental. It is being managed, and the forces managing it are malignant.