The Message
Published in

The Message

What’s Wrong with Twitter’s Latest Experiment with Broadcasting Favorites

It Steps over Social Signals While Looking for Technical Solutions

Last month, I was invited to Twitter to give a talk. Twitter has made a huge, positive difference in my life, and changed the landscape of politics in multiple countries I studied, including my home country, Turkey. So to start the conversation on a good footing, I opened my talk with this question:

“Would you like me to discuss why you screwed up the mute/block expansion?”

Yes, I missed charm school, why do you ask?

But I started this way mostly because I care! These platforms have become very important parts of our civic and personal lives, and their design decisions have significant consequences. In truth, the audience was receptive to the discussion and it felt productive, at least for me.

Yesterday, I saw that Twitter has decided to experiment with tweaking people’s timelines, this time by broadcasting favorites—Twitter’s bookmark function—by people you follow into your timeline—as if they were retweeted.

I have already written about problems with over-reliance on algorithmic curation, but let me expand on what’s wrong with this new experiment. I’m not completely against using algorithms in social platforms or experimenting with new features — it’s just that this has to be done with an understanding of the social contexts of these platforms, and without disempowering users. This one, unfortunately, does not do that.

This is the same class of mistakes that tech companies make again and again, not just Twitter, so it’s time to spell it out.

Dear tech companies: Stop stepping over social signals in your search for technical solutions. Technical affordances and social signals have to be evaluated within their totality, not as if they were separate universes. They are inseparably entangled.

These platforms exist at the intersection of both the technical and the social, but for the users, it’s the social signals that drive the system. Too many companies are still looking at things from mostly the technical end. Technical solutions have to be built around and integrated with social signals, not by stepping on them.

Let’s start by recounting what went wrong with mute/block. You may remember that there was a time when Twitter only had block, unless you used a client. A block meant that the person you blocked could not see your tweets when logged in — but if your timeline was public, logging out made your tweets visible to the blocked person. So Twitter decided to replace “block” with “mute,” which meant that the person you muted could still see your tweets, but you wouldn't see theirs.

It sounds perfect, on pixel. In effect, Twitter had decided that the solution was to get the problem person out of your hair, and sought to do this by muting them, without informing them.

I can almost envision the meeting. “Look”, says the advocate. “The offending party can always log out or create another account — so it’s not like block really works. By muting without informing, we solve the problem for the bothered party, and because we don’t really inform the muted person, they’ll keep doing whatever it is they do without being visible and not even knowing they are “shadow-banned,” i.e. made invisible without notice. So they won’t take extra measures to keep up the harassment! Voila! Problem solved.”

It sound logical enough…for C3PO. Here’s why it didn't work for humans — and why the backlash came, leading Twitter, to its credit, to quickly backtrack.

Humans exist in social groups and rely on social signals to make the group work — signals that are symbolic, and appear without teeth, but in essence are among the most powerful dynamics of a human society. “Sticks and stones will break my bones, but words will never hurt me,” lies the saying. It’s not true. We care deeply, deeply about our social status and signals, and will do just about anything for them. And humans will make anything and everything into social signals. Hair, clothes, words, finger nails, tattoos, fork on the left, mimics, body. You name it. Put people in jail and in uniforms, and they will create social signals out of tattoos. In schools with uniforms, slight tweaks to hair will become a signal. Take away everything, and they will signal by the way they walk, or don’t walk, or blink. Social signaling is ingrained into everything we do.

In tech platforms, when our signaling ability is limited to technical affordances, we adopt existing tools and transform them into social signals. Things that start off as utilities, or only “technical” affordances, soon acquire social meaning. In Twitter, this is true for both block and favorite (but not mute because it is not visible — hence it is not a signal to the other party. A signal, by definition, is visible).

So, in essence Twitter’s block had become a social signal — it said, go away. GO AWAY. The point of it was to be visible to the offending party. Of course we know that the person may log out and see our tweets anyway. But, in most interactions, an overwhelming majority of people do respond to social signals. And giving those social signals makes us feel better — that is not without worth. We honk our horn for a reason. Yes, in all human societies, there is a small number of people who do not respond to social signals — some call them sociopaths — and I suspect that small minority are behind most of the problems of abusive and aggressive behavior we see online. In other words, only a very few people will troll; but at a global scale, it adds up to create a real enough problem.

We need social signals to deal with the former (the vast majority who respond to social signals), and technical solutions for the latter (the small minority who don’t). Tech companies keep mixing and conflating the two.

Regular, offline society of course operates the same way and works via social signals. A primatologist once remarked that you couldn’t put 100 chimpanzees in a small, packed metal tin tube and expect them to all disembark in one piece, five cramped hours later. Yet, we easily do this. (Okay, we fight over the recline). We fancy ourselves lions, but we are very compliant herd animals, most of the time. In fact, in contradistinction to the myth of Hobbesian chaos in times of stress, disaster, and protests, most of the time, most people cooperate and act altruistically, especially under stress. (Yes, wars also exist and that’s the other end of the human spectrum: we are a bundle of contradictions). In ordinary times, we do envision harsher punishments and confinement for the minority of people who will not cooperate and respond to social signaling. In truth, the reality of the criminal system operates far from that ideal, and in many countries also acts as coercive control on the marginalized. However, even in very egalitarian and community oriented societies, even at the scale of the pre-modern village, you do see confinement and punishment because there is indeed always a small group who only responds to deterrence beyond the mere social signal.

Twitter claims this change showing favorites to others generates more engagement, and people sometimes click on the “favorited” tweet they have been shown.

I believe that but who cares?

The problem with this course of action is not that it doesn't generate engagement, but that it violates the social expectation of the favoriter, who did not intend to broadcast this act, and steps over the social signal function of favorite, which for many has come to mean: “I’ve seen this [and appreciate it] but am choosing NOT to broadcast this to everyone but only conveying it to you.” It has other uses as well, (in fact, a researcher found 25 functions) but Twitter’s favorite is widely used by many as the opposite of Twitter’s broadcast, the retweet, as an individualized and quiet signal as the “notification” tab will show it to the user whose tweet was favorited, but will not broadcast it to all followers.

If Twitter goes ahead with making public some favorites, at once, it will kill a useful social signal in a system whose main drawback is lack of social signaling affordances and will also violate visibility expectations, which Google Buzz and Facebook famously got in trouble for.

I can almost imagine the meeting for this one, too, envisioned from the point of view of the timeline: “Putting someone else’s favorite adds maybe one percent more content to the timeline so why would people complain? It creates more opportunities for engagement with interesting content.” But Twitter is not just a timeline, it’s a social system, and the issues it needs to deal should not be calculated as “how tiny a change we made to the timeline” (admittedly small) but “what are the social signals we are modifying when we do this?” (which are huge in this case).

As I said, I understand that Twitter has a problem with bringing on board new users, and algorithms can help with some of the challenges it faces. I made some suggestions in my previous post. But Twitter needs more social signals to let users better manage their attention (a timed mute, please, so that I don’t have to choose to forever mute people from whom I don’t want to see content, but for the day or the week not forever), and algorithmic suggestions to help people find engaging and interesting content (not just people like people you follow — that’s the easiest thing to do, and the thing I least need algorithmic help for).

Also, the Twitter blog suggests that Twitter looked at engagement rates to judge the worth of these tweaks. Looking only at engagement is the metric of doom, though I understand why it’s attractive. Tech platforms should always try listening to the users about changes. Experiments to platforms are fine, within ethical confines, of course, but in human systems, one should always always listen to the humans, not just look at imprints of their behavior. A lot of discontent can be hidden by behavior. Recent reports of a deep drop in Facebook use among teens have not surprised a single researcher who works in this field. To be sure, Facebook will remain Facebook — a massive platform that acts like a digital directory. But its privacy and naming missteps have caused deep mistrust among many groups — and especially young people because, more than anyone else, and contrary to the misconception, young people struggle deeply with visibility and privacy because they have little power in this world. “Engagement” can continue, while masking a desperate search for an alternative (at the moment, that’s Instagram, Snapchat and Whatsapp, so that explains Facebook’s billions of dollars shopping list).

So, Twitter, yes, please find ways to make Twitter easier for new users. Create good tutorials. Use algorithmic suggestions in ways that give people more content to choose from. Give us more technical capabilities to manage our attention and to socially signal to each other. Create community based structures (group DM, shared, timed mute and block lists), and empower users — but make it all optional as the core of the system is set, and its social conventions are all built on it. Please, don’t take away the few available social signals in a system that already has too few.



A Pandaemonium Revolver Collection. Season 2 stars @anildash @alanalevinson @ftrain @hipstercrite @itsthebrandi @jamielaurenkeiles @vijithassar @yungrama @zeynep. Season 1 available on DVD shortly.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Zeynep Tufekci

Zeynep Tufekci


Thinking about our tools, ourselves. Assistant prof at UNC iSchool. Princeton CITP fellow, Harvard Berkman faculty associate, Sociology.