How Twitter Got Angry

Twitter is suffering from a systemic harassment problem. This isn’t news—it’s been written about over and over again, and has become a trope in the cultural mainstream. “We suck at dealing with abuse and trolls on the platform, and we’ve sucked at it for years,” the company’s CEO wrote in an internal memo last year. Recently, Paul Ford summed it up: “Google has search. Facebook has the social network. Twitter has the angry social network.”

To understand how we got here, it helps to look back to the early days. In 2008, after a user went public about her harassment and the service’s inability to stem it, Twitter executive Jason Goldman wrote: “What we believe is that Twitter is a recipient-driven utility; you choose what content appears in your timeline. We offer tools like block so that users can distance themselves from others with whom they have disputes or disagreements.” Which fits in with cofounder Ev Williams’s explanation of replies around the same time:

My followers will only see my update if they follow both of us (if they have their setting on the default).
We’re trying to avoid the situation of you hearing someone answer a question when you didn’t hear the question (for instance). Also, you don’t have to hear answers to the question from people you don’t want to hear from.

Dismissing the real harassment going on was bad, but the Twitter execs were right about how Twitter didn’t systematically amplify it. The underlying model for following meant that we all saw a slightly different Twitter, as I wrote in 2009:

The idea of Following means that the natural view we see on Twitter is different for each of us, and is of those we have chosen to hear from. In effect we each have our own view of the web, our own public that we see and we address. The subtlety is that the publics are semi-overlapping—not everyone we can see will hear us, as they don’t necessarily follow us, and they may not dip into the stream in time to catch the evanescent ripples in the flow that our remark started. However, as our view is of those we choose to follow, our emotional response is set by that, and we behave more civilly in return.

That we had to go to a little extra effort to see the replies then—they were in a separate tab, weren’t under posts, and were not sent as notifications—reinforced this cultural feedback loop. If you’ve read @AllTheTwits, the account that reposts tweets from 2007, you can tell how different the atmosphere was. The service had built a way to dampen anger and encourage discourse. Suw Charman-Anderson makes an architectural analogy to describe what she calls a valuable “privacy gradient”:

The idea of a privacy gradient comes from architecture and refers to the way that public, common spaces are located by the entrance to a building and as you progress through the building the spaces become more private until you reach the most private “inner sanctum.” . . . As one moves along a privacy gradient, one is also moving along a parallel trust gradient. As you invite me deeper into your house, so you are displaying increasing trust in me. . . . Six years ago, I thought that Twitter had a basic, but basically sufficient, privacy gradient. And, indeed, it might have been sufficient for the network in 2010, but it is now completely insufficient.

That might have seemed unintuitive given notions of the freewheeling, public web, but it made Twitter a productive medium for conversation.

If you accept Habermas’s assumption of a single common public sphere this makes no sense—surely everyone should see everything that anyone says as part of the discussion? No. This has never worked at scale, and elaborate systems ensured that only a few can speak, and only one person can speak at a time, because a speech-like, real-time discourse was the foundational assumption of debate.

This public worldview became the default assumption of communication online too. We see it now when privileged speakers decry the use of anonymity in the same tones as 19th-century politicians defended hustings in rotten boroughs instead of secret ballots. The tactics of shouting down debate in town halls show up as the baiting and trollery that made YouTube comments a byword for idiocy; when all must hear the words of one, the conversation often decays. First with blogs and then on Twitter another way emerged—many parallel overlapping conversations that spread ideas memetically rather than through Robert’s Rules.

This sieving and amplification—mutual media making sense of the world—is what gives Twitter its power to shape events and make news. Retweeting and responding, @ replies, and hashtags help to bridge ideas, jokes, links, and other memes between these different publics. Some take off, most vanish, but there is a connection there.

The destabilizing changes came about when engagement was taken as a goal in itself—many product changes have focused on this rather than on striking a balance, as the company’s earlier statements implied. The two key ones that changed the cultural context seem like natural efficiencies: notifying when someone @ replies you, and showing all replies under a post.

But this reverses the sense of publics, collapsing them all. We are back in the world where any passing remark by someone you haven’t chosen to follow is shown to everyone—the tragedy of the comments. The notification takes us back to the email inbox problem that Twitter had seemed to transcend: Suddenly, you are interrupted by strangers responding, and their priorities shape yours.

Now, part of this is valuable. If you are a company or brand, having a direct channel to your most frustrated customers like this can be very important. If you are trying to gather political momentum to speak truth to power, quickly amassing replies and comments can be valuable for amplifying an important message. These are the cases Twitter touts to businesses and projects to the media as their source of relevance, built on the free speech ideals that they say they want to protect.

However, this leads to a perceptual disconnect, in a variant of the “my friends have more friends than me” paradox that prevails on many social media platforms. On Twitter, the least typical tweets are the ones that get the most eyeballs, a dynamic that shapes users’ mental model of the platform. This is amplified by the on-boarding process that encourages users to follow famous people, and even to some extent by the ways that brands are successfully using the platform for better customer service engagement. Twitter thus is seen as the place where you shout at people or companies, and are jealous of those whose louder shouts you see every day.

There is some cultural inertia from the previous “multiple publics” model of Twitter, in the same way that there are those of us who remember when an email from a stranger was likely to be interesting and not spam. But Twitter’s attempts to preserve this earlier, friendlier culture, by creating a separate public for verified users, are proving weak.

Typically, in the op-ed tradition, here’s where I’d suggest the magic quick fix that solves this, ideally in 140 characters.

Unfortunately, anonymity isn’t the cause of abuse, as @Jason suggests, so much as perceived social norms. As one study of Twitter demonstrated, some trolls appear to become more hostile when they use their real names.

“Our results also do not support claims that prohibiting online anonymity will make the online world a better world,” Stahel explained by email. “The main point is that prohibiting anonymity online will not settle this ‘problem’ of firestorms.” Indeed, for some trolls, online aggression is rewarded in their social networks, and is often a deliberate public signal. People are actually trying to enforce social norms against a perceived violation by a public figure or group.

What is needed is an analysis and modeling view that bridges literary theory, computer engineering, and social science. I saw how the “everyone sees everything” worldview of web comments was first transcended by Twitter, driving its initial vibrant culture and rapid growth. But then the explicitly engagement-amplifying changes brought in many of these old assumptions implicitly, bringing back those problems.

The new “quality filter” given to those who are verified Twitter users—and, as of this week, rolled out to the rest of the Twitter community—only hides the abuse for those who use it. The abuse is still visible to everyone else by default, where it has the corrosive effect on public mood that is all too visible.

What if, under a popular tweet, instead of seeing all the replies by default, we saw only those from people both ourselves and the post author followed? Or a more nuanced display, with replies varying in visual weight depending on the degree of connection between us and the responder, or the connection between the original post author and the responder?

For these anomalously popular posts, what to show becomes a much more subtle optimization problem. Now that the “abuse filter” has been given to everyone, the second order effects mean that it may likely soon be reverse-engineered and nullified, in the same way that every Facebook algorithm tweak is pounced on and exploited, as the experimental conditions in which it was tested no longer apply. Like all attempts to fight abuse, what really matters is not crude technical signals, but working out how to reward those that amplify a culture of conversation rather than one of anger.

By each reading whom we choose to and passing on some of what we see to others, we are each others’ media and mediators. We are the synapses in the global brain of the web of thought and conversation. Although we each only touch a local part of it, ideas can travel a long way. Twitter is a key part of this consciousness, and it needs to become less prone to seizures.

Kevin Marks tweets at @kevinmarks