Sep 4, 2014
Why Twitter Should Not Algorithmically Curate the Timeline
It’s the Human Judgment of the Flock, Not the Lone Bird, That Powers It
Twitter’s CFO made some headlines recently by suggesting that Twitter was going to tweak the reverse chronology of the feed and introduce algorithmic curation:
Twitter’s timeline is organized in reverse chronological order, a delivery system that has not changed since the product was created eight years ago and one that some early adopters consider sacred to the core Twitter experience. But this “isn’t the most relevant experience for a user,” Noto said. Timely tweets can get buried at the bottom of the feed if the user doesn’t have the app open, for example. “Putting that content in front of the person at that moment in time is a way to organize that content better.”
Many on my Twitter feed were strongly opposed to the possibility.
Many, like me, though had reasonably sized follower lists, and would almost certainly benefit as such algorithms tend to inevitably optimize for engagement (or eyeballs, aka, what advertisers care for).
Such systems reward the already-rewarded, in other words.
And while we were left wondering if the Twitter CFO was taken out of context or misconstrued given Twitter CEO’s objection here, the unease remained among many.
So, why the distaste for a change that would benefit many of them? It’s simple: Twitter’s uncurated feed certainly has some downsides, and I can see some algorithmic improvements that would make it easier for early users to adopt the service, but they’d potentially be chopping off the very—sometimes magical—ability of mature Twitter to surface from the network. And the key to this power isn't the reverse chronology but rather the fact that the network allows humans to exercise free judgment on the worth of content, without strong algorithmic biases. That cumulative, networked freedom is what extends the range of what Twitter can value and surface, and provides some of the best experiences of Twitter.
And hence my first reaction to the possibility:
As an example, consider the case of how Twitter figured out the death of Osama Bin Ladin.
On May, 2011, the President of the United States announced an emergency presidential address later that night. What a surprise! Twitter, obviously, went into overdrive with speculation. There were many possibilities. The Libya air attacks were ongoing, and Gaddafi was not yet captured. Could it be Osama Bin Ladin? Something else? Could be anything.
At exactly 10:24pm, Keith Urbahn, ex-chief of staff for Donald Rumsfeld, a Twitter user without a large following or presence, tweeted the following:
Within minutes — minutes — the network had figured out this was the answer. The first step was Brian Stelter who recognized who Urbahn was, and then it was the rest of us who had realized that Stelter, a reliable journalist, would not be retweeting this without cause, and so it went. (If I recall correctly, Urbahn at the time had followers in the three digits so he was not a very visible node—but no matter. The Network is the power, not the node).
Look at the speed with which that judgment spread in the network:
I honestly doubt that there is an algorithm in the world that can reliably surface such unexpected content, so well. An algorithm can perhaps surface guaranteed content, but it cannot surface unexpected, diverse and sometimes weird content exactly because of how algorithms work: they know what they already know. Yet, there is a vast amount of judgement and knowledge that is in the heads of Twitter users that the algorithm will inevitably flatten as it works from the data it has: past user behavior and metrics.
I have witnessed Twitter network’s ability to surface unexpected content again and again, from matters small to large. It’s true, Twitter can be rife with rumors, some false, especially at times of protests, disaster or other crises. But the speed with which the correct information is also surfaced is even more impressive, and only possible because the network quickly surfaces it, with each node filtering news through judgement and experience. (Of course—this depends on your network and perhaps Twitter can help users construct a better, more diverse, less homophilic network by suggesting different kinds of users, rather than the ones already like the ones a person follows. Redundancy restricts networks whereas diversity enriches them).
Twitter brims with human judgment, and the problem with algorithmic filtering is not losing the chronology, which I admit can be clumsy at times, but it’s losing the human judgment that makes the network rewarding and sometimes unpredictable. I also recently wrote about how #Ferguson surfaced on Twitter while it remained buried, at least for me, in curated Facebook—as many others noted, Facebook was awash with the Ice Bucket Challenge instead, which invites likes and provides videos and tagging of others; just the things an algorithm would value. This isn’t a judgement of the value of the ALS challenge but a clear example of how algorithms work—and don’t work.
Algorithms are meant to be gamed—my Facebook friends have now taken to posting faux “congratulations” to messages they want to push to the top of everyone’s feeds, because Facebook’s algorithm pushes such posts with the phrase “congratulations” in the comments to top of your feed. Recently, a clever friend of mine asked to be faux congratulated on her sale of used camera equipment. Sure enough! Her network reported that it stayed on top of everyone’s feed for days. (And that’s why you have so many baby/marriage/engagement announcements in your Facebook feed—and commercial marketers are also already looking to exploit this).
For another thing, algorithmic curation will make writing to be retweeted, which already plagues Twitter much worse. I’m not putting down the retweetable quote; just the behavior that optimizes for that above everything else — and I know you've seen that kind of user. Some are quite smart. Many are very good writers. Actually, many are unfortunately very good writers. They are also usually insufferable. I can see them taking over an algorithmic Twitter.
Bleargh, I say.
But the bigger loss will be the networked intelligence that prizes emergence over engagement and interaction above the retweetable— which gets very boring very quickly. I know Twitter thinks it may increase engagement, but it will decrease engagement among some of its most creative segments.
What else will a curated feed optimize for? It will almost certainly look more like television since there is a reason television looks like television: that’s what advertisers like. There will be more celebrities. There will be more pithy quotes. There will be even more outrage, and even more lovable, fluffy things (both are engaging, and remember, algorithms will optimize for engagement). There will be more sports and television events. There will be less random, weird and otherwise obscure content being surfaced by the collective, networked judgement of the users I choose to follow.
Does Twitter have a signal-to-noise problem? Sure, sometimes. But remember, one person’s noise is another’s signal. Is the learning curve too steep? Yes, it is. Is there a harassment issue, especially for the users with amplified footprints? Absolutely.
But there are many, many things Twitter could do to address all of that without breaking its networked, human-prioritizing logic. Much much better tutorials seems like such an obvious step (I have hardly seen good ones). Better suggestions for users to follow, perhaps a dozen at a time, and better ways of trying following groups of people. Right now, it’s all individual and arduous, and that should remain the core option, but the entry ramp could be much faster. Better filtering, too, especially of mentions would be very welcome. I’m craving a timed mute, for example—let me mute out someone who I don’t happen to want to listen that day or that week, without having to mute them permanently. Group chat for DM? Woohoo. DM is among Twitter’s most powerful features because it only allows contact from people one chooses to follow which is a better filter than email, but not as strict a one as Facebook which operates differently. Also, brevity makes DM more powerful. And lists! Twitter can do so much more to make lists more useful to its users to let users decide how to deal with signal/noise and interest ratios.
There is so much Twitter can do try to improve the user experience, for both the experienced and the beginner. But I hope that it does not algorithmically curate the feed, not because I love the chronology per se, but because I value people’s judgement. Yes, Twitter can make it easier to access that judgment in more varied ways but stepping between people I choose to follow and me is not the answer.
Never forget: the algorithm giveth but it also taketh away. Don’t let it take away the network because it’s the flock, not the bird, that provides the value.
