Twitter has a problem that you may have noticed. Its simple stream–what was once essentially a big reverse chronological conversation–doesn’t scale well. Sure, it was fine back when you followed a few dozen people. But with hundreds, or thousands, of people babbling in the stream, the conversation becomes cacophony. And if you’re anything like me, you’re forced to skip past the last few hours of conversation you missed and just jump into a new one.
The company has tried to find solutions. Hashtags let you track specific topics. Trends point out the hashtags that have gotten big. And amongst this global conversation, @ symbols grab your attention through the crowd with laser-like precision. But it’s still not enough to find the relevance in the noise, which is why two things have happened.
Last year, Twitter admitted that a more Facebook-like feed, prioritized by algorithm instead of time, might be a good idea.
This week, Twitter acquired Whetlab, a startup launched by Harvard, Toronto, and Sherbrooke researchers, that promises to make implementing machine learning into a product at the scale of weeks instead of months.
The question becomes, can AI fix Twitter without turning Twitter into a Facebook or Google Now clone?
It’s safe to say that if Twitter turned into Facebook tomorrow, we’d all revolt (or in reality, we’d just tweet that we would). This is because Twitter has differentiated itself by letting the user be in complete control. Like a Vegas buffet, customers can choose to overload their plate, slide the excess aside when no one’s looking, then start the process anew. It’s not efficient, but it’s all there if you can eat it.
But it is possible for AI to improve the Twitter experience without destroying the premise. Even before the Whetlab acquisition, you’ve probably seen that Twitter started to float popular tweets that you missed “While you were away” to the top of your feed. Technically, these tweets bend Twitter’s steadfast rules of chronology, but they don’t really break them. Twitter’s UI implements clear language to signify that “While you were away” content is a special case, while addressing any Twitter user’s biggest fear, that in skipping to the top of their feed, they’re going to miss out on an important nugget of information.
So eventually, maybe the platform isn’t even suggesting you read a tweet with a lot of favorites and RTs, but that you read a tweet about a new cross stitching technique, because Twitter recognizes that you historically clicked on cross stitching links, and had you been around two hours ago, there was a 50% chance you’d have checked it out.
Drawing The Line Is Impossible
However, there’s a catch. If we begin to follow this AI paradigm to its ultimate conclusion, and if we actually believe that AI could learn our preferences better than ourselves, then why wouldn’t we ask–as users–for the AI to only show us only the tweets that we’d care about? Take my social circle. Analyze their conversations. And cut them down to the stuff you know I’d be eavesdropping on and clicking.
And Twitter, while you’re at it, feel free to take into account my location. If you know I just left the theater after watching a certain movie, sure, maybe I’m more interested in seeing what my friends are saying about that movie. Or if you recognize that I tweeted about my dinner, show me more dinners that my friends are eating.
That’s right, Twitter, prune my feed. Leave the chronology, give me the option to see an unadulterated feed (I’ll never use it), but let the AI dig in. Just show me what is relevant at any given moment of my life. This is gonna be great!
But…wait. That sounds a lot like Google Now, and Apple Proactive, and Facebook’s algorithmic news feed. It’s not just a matter of copycatting. This is design darwinism at play, what happens when a lot of smart companies find the same solution when working toward the same goal.
The Perfect Feed Is Now A Competitive Business
The problem with the most perfectly designed feed, that lists only relevant information when you need it, is that every company that matters is trying to build it. The differences become minor selling points that are mostly invisible to the user–Google sees into my Gmail, Twitter knows about my hashtag obsession, and Facebook can analyze the last slideshow of photos I uploaded. The exact balance of utility and social may change from service to service, but it’s all melding into one, perfectly pruned, list of stuff we need to see. It’s the presentation of pure insight.
But maybe that’s why Twitter’s hedging its bet with another initiative the company is working on: Project Lightning. Launching in the coming month, a staffed Twitter news room will organize tweets, photos, and videos encapsulating the biggest events of the day. You’ll be presented with the stories on a special screen right when you open the app, and you’ll even be able to temporarily subscribe to updates for events like the Grammys, building them into your timeline for just a few hours. But all of this teased curation is generalized, like TV channels, rather than personalized to your unique tastes. And it’s also, ultimately, the promise of more content from people that you don’t even follow.
But at the end of the day, deep machine learning can still do something human curators cannot, making these news events and missed tweets more personalized to your specific digital footprint.
And so for Twitter, the design problem at play–simplifying the original feed–may be more difficult than cutting noise, presenting relevant content, helping us discover new content, or being seamlessly integrated into the context of our lives. In the age of Silicon Valley’s new penchant for AI-curated feeds, these features have rapidly become table stakes.
Twitter’s new machine learning company may be able to fix Twitter’s design, but if the result is the perfectly pruned feed, that is both an incredible proposition, and a completely typical one. Which is why I won’t be surprised if, instead of leveraging AI to show us less, Twitter will simply lengthen the buffet. Don’t sweat what you can’t finish. There’s always more to try.