Earlier this year, Microsoft unveiled Murphy, the latest eerily smart chat bot out of the machine learning labs. Last year, it produced one that guessed how old users were based on a picture. Next came Tay, a “teen girl” who quickly learned, via the magic power of Twitter, to be a Nazi.
The team’s latest bot, Murphy, has an altogether more noble mission: mashing weird shit up. Ostensibly created to answer users’ most outlandish “what if” questions, what the bot actually does is a weird, programmatic photoshop. Microsoft’s official post on the project, which began earlier this year, features answers to the questions, “What if Charlie Chaplin had a really long beard?” and “What if Marilyn Monroe was the Mona Lisa?” Innocent fun!
You can add the Murphster to Facebook Messenger, Skype, Telegram, and, in limited form, Kik and Slack. As with any bot, your first order of business is trying to trick it, to tease out the limits of its artificial intelligence. That doesn’t take long for Murphy, who, more often than not, tells you to ask a different question, because he’s still learning. Over time, the idea is that this won’t be the case—that enough queries and enough results will create a bot that has heard most of our likely questions and brewed up a handful of satisfactory responses. That it’s still throwing “try again” messages out, six months after its release, is not promising.
Right now, though, you’re pretty safe asking it to do shit having to do with political figures or celebrities.
But its best responses otherwise come from trying to break it.
Boy, we’ve all been there.