Latest in Culture

Image credit: Daniel Cooper

An AI taught me to be a better tweeter

This is what happens when you let a bot critique your social feed.
332 Shares
Share
Tweet
Share
Save
Daniel Cooper

My name is Daniel Cooper, and I tweet ... a lot.

Twitter is an extension of my subconscious, a pressure valve that lets half-baked thoughts escape my mind. In the last seven years, I've tweeted 73,811 times, and yet none of those 140-character messages has made me internet-famous. For all my efforts, I've accrued just 5,635 followers, most of whom are in tech and were probably made to follow me by their boss. It seems that no matter how much I try, I'm never going to become a celebrity tweeter.

That gnawing neediness in my soul explains why I was intrigued by Post Intelligence, a startup with a deep learning algorithm that can supposedly make you better at social media. For now, Post Intelligence works with Twitter and Facebook, but there's potential for it to engage with several other platforms. The algorithm is the brainchild of Bindu Reddy and Arvind Sundararajan, two former Google product managers. Reddy can count Blogger, Google Video and Google+ on her record, while Sundararajan helped develop Gmail and AdSense. Together, they previously founded media agency MyLikes and Candid, a Secret-esque anonymous network.

Pi, as Post Intelligence is nicknamed, studies your record to learn how you tweet and then looks for patterns to help improve your strategy. It slurps down your most recent 3,000 or so posts, determines your most commonly used words, calculates the scope of your influence and analyzes the sentiment of said tweets. Alas, Pi managed to overestimate my legendary levels of miserableness by marking these jokey tweets as having a negative tone.

The platform also examines what times of the day your messages are most likely to get good traction. Given that the bulk of my audience is in the UK and in the eastern US, it's little surprise that my posting windows are 11AM, 1PM and 4PM GMT. So, when I write an explosive tweet with the potential to make everyone in the world chuckle, I should schedule them for those times for the best potential results.

But it's the main interface that's the most interesting because it encourages you to look for trending topics to glom onto. The day that I gained access to the platform, a British politician announced that he would take on his sixth concurrent job: editing a daily newspaper. I'd already made several unsuccessful attempts to write a joke on the subject, but this time I could use Pi's rating system as a guide.

You see, rather than issuing a prescriptive system that will tell you what to tweet, Pi's algorithms will give you a rough idea of your tweet's potential. A "Prediction Bar" sits below the compose field, giving you a score out of 10 for the messages that you write. It's almost a game unto itself, as you trial-and-error your way toward a hot tweet in the hopes that the system will bless it.

My next attempt was to try and satirize Cosmopolitan's decision to go all-in on Jamie Dornan. The Fifty Shades of Grey furniture was plastered above the magazine's Twitter page, presumably in celebration of St. Patrick's Day. Unfortunately, a tweet with a measure of subtlety can fly past the algorithm's senses and, no matter what combination I tried, I only ever got a score of two out of 10. Or, and this is the more outlandish theory, my weird sense of humor isn't actually funny and my whole life has been a lie.

There's another issue too: Pi can't be used as a full-blown Twitter client -- only as an adjunct to the main site. You can't see the firehose of tweets as the day unfolds, and you have to stop yourself from reacting as you normally would. Rather than firing off a response, you have to pause, switch tabs and then tweet inside the Pi window, checking for your tweet's virality. That's a blessing because you're forced to tweet more thoughtfully, and a curse because you lose the instant gratification that Twitter provides.

But what's interesting, at least to me, is how having Post Intelligence in my life has redefined my relationship with the site. Whereas before I would use it almost exclusively upon instinct, typing whole tweet threads upon instinct, now, I just ... don't. Instead, I'll try and work out the neatest, highest-scoring phrase for my new master in the hope of getting a higher score out of 10. It's not yet made me internet-famous -- my follower count still remains hopelessly low -- but perhaps I'm not far from my big break.

From around the web

ear iconeye icontext filevr