Nearly a year ago, the team here at Broadly was blessed with a slew of bot-generated headlines almost too perfect for our site. To name a few:
- Nice Day for a Sexual Relationship
- The Juggalo Rapper Who Sang on a Man for Farting
- The Feminist Terrier Pawing Her Way to Map and Find Girls at Risk of Homelessness
- Learning to Skateboard When You're on Your Vibrator?
- Before the Paleo Diet, Walt Whitman Took a Dick Pic
After we finished laughing at them, I must admit we were slightly concerned: Who was this so-called bot and why was it capable of doing our jobs better than us? Skateboarding on a vibrator? A feminist dog saving girls from homelessness?? Click, click, click! Would we soon be replaced by a genius algorithm with its finger on the pulse of our ethos?
Luckily, almost a year later, I’m happy to report that Broadly’s employees are still indeed humans. However, after Paul Glenn, the mastermind behind the Broadly headline bot (and VICE’s Senior Data Analyst on our global business analytics team), shared the algorithm’s latest creations, we were feeling uneasy all over again.
- Men in Hollywood Ready to Take the Perfect Nude
- How Doctors Gaslight Women into Haunting Sculptures
- Lazy and Foolish Men Still Fail to Realize How Much the Super Blue Blood Moon Will Ruin Your Life, Astrologically Speaking
- The Creative, Subversive Ways People Are Insufferable Babies Who Need to Know About Brett Kavanaugh's Anti-Choice Record
- Statue of Liberty Protester Could Face Up to Extremely Intense Orgasms
- Training My First Crush
- Bigots Are Donating Thousands to Keep Stalking Victims Safe?
- The Infuriating Hubris of Men Who Are Heavily Tattooed Body Pillows
- Broadly Call-In: We Want to Get Over Your Ex
- Winona Ryder Is the Unbearably Stubborn Hedonist of the Viking Reenactment Scene, in Photos
Concerned after realizing I wanted to read every single one of these non-existent articles, I asked Glenn how his algorithm works, hoping we could channel its methodology in our own mortal brains. “It works using a probabilistic modeling technique called Markov chains, which we implement using an open-source package called Markovify,” he told me. “A Markov chain works by looking at sequences and calculating the probability that one item follows another. I trained the model by giving it all of the Broadly headlines from the U.S. site in 2018.”
For those who aren’t up to date on Markov chains (me), Glenn offers a clear example. “Sometimes the result of that is it just mashes together a few headlines,” he said. “Like ‘Images Exploring What It's Like Being a Phone Sex Operator Taught Me About How Men Think,’ which is a combination of Images Exploring What It's Like to Be Part of India's Queer Community, What Being a Phone Sex Operator Taught Me About Lonely Men and What Being a Cam Girl Taught Me About How Men Think.” That seemed like a lot of work, and as I got deeper and deeper into the 200 (!) bot headlines Glenn sent over, it became clear that maybe this algorithm wasn’t as spot-on as I thought:
- The Point in Your Butt
- How to Eat Out a Non-Op Trans Woman Running for US Office Than Ever Before
- The Candidate Who Plans to Be a Space to Let Loose Online
- A Safe, Effective Birth Control With Your Partner Smells
So it doesn’t look like the bot is totally coherent enough to take over our jobs…yet! And it has definitely inspired us. If any of you lovely readers happen to know a puppy saving girls from homelessness who’s available for interview, we’d very much appreciate being put in touch.
And lastly, Glenn let me know that his team is looking for some more data analysts (especially women and gender non-conforming people)! If you’re interested in algorithms, you can apply here and here—just don’t create any bots that could actually take over our jobs please and thank you.