In the glorious, old, unhinged days, before Trump was the president and the greatest existential threat facing humanity, you could tell whether or not he had actually written a tweet by what device it was sent from. The crazy late-night stuff came from an Android, whereas the sober, on-message stuff came from an iPhone. Apparently Trump is not swayed by Apple’s design excellence and walled garden of apps.
However, as the new, purportedly more “presidential” Donald Trump is successfully muzzled by his staff, tweeting less and less frequently from his Android, it has become harder to differentiate the real @realDonaldTrump from the fake @realDonaldTrump. (Do those tweets count as fake news?) Fortunately, a journalist for The Atlantic has constructed a bot that uses machine learning to determine from the tweet’s very language choices who wrote it.
It’s not exactly foolproof yet, as his handlers are getting better at emulating his style and his personal fixations remain a moving target. But then, that’s why the machine is learning as it goes. Interestingly, the greatest tells for whether or not Trump wrote something are whether or not the tweet mentions his own handle, the media, @foxnews, or includes extra spaces. This, in case it isn’t obvious enough, is because he is a senile, poorly educated narcissist with a crude grasp of the English language. The best way to tell if a staffer wrote something, on the other hand, is if the tweet includes hashtags, which in a rare showing of good taste Trump himself does not care for.
The bot doesn’t seem to computer much lower than 44 percent for some reason, but when it’s certain, it’s certain:
Which was about this tweet:
While the bot judged thusly:
About this one:
Inhale deeply: That’s the smell of an original Trump tweet right there. You feel it in your bones, but it’s good that the data backs it up, as well. In the meantime, if Trump’s handlers really wanted to keep him off the tweets, they know one foolproof way to keep him occupied.