Human Writers Take on AI in Crafting Twisted Game Cards

You might call the contest “AI against Humanity.”

On the Friday after Thanksgiving, when families are likely to be shopping or playing games, the makers of one of the most twisted card games sold in stores held a game of their own, pitting their writers against AI.

In what has become a Black Friday tradition, the makers of the game Cards Against Humanity challenged their human writers to create new cards at the same time an AI algorithm was charged with the same task. If the AI created better cards, meaning they were more popular in online sales to viewers watching both teams create the cards, they threatened to fire the writers. If the writers won, they got a $5,000 bonus.

“Black Friday probably represents the worst things about our culture,” Cards Against Humanity co-creator Max Temkin said in a statement last year when the challenge debuted. “It’s this really repulsive consumerist frenzy right after a day about being thankful for what you have. So it’s always seemed like a really good subject for parody to us.”

The gamemakers created the gimmick to see if AI could write as strange and creative as human writers. Spoiler alert: AI lost, the second year in a row.

The humans and AI battled for 16 hours in creating as many cards as possible. The cynical, satirical phrases crafted by humans won by selling 82,860 packs of cards, compared to AI’s 81,135 packs.

Human cards had phrases like “Sure, sex is good, but have you tried ______?” and “Losing a loved one to Fox News.” The top two AI cards read, “These amazing testicles I’m about to share with you” and “Sitting on my son’s bed, thinking ‘I could kill him.’ ”

Mashable.com pointed out a few choice cards, too:

“So far, the humans have come up with some pretty great cards, including ‘Unwelcome sexual attention from grown men,’ ‘Getting sad for no reason,’ ‘Three hours of pimple-popping videos,’ and, fittingly, ‘A pointless job at a soulless corporation.’

According to a story on theverge.com, the stunt isn’t just a gag for laughs.

“It’s a legitimate neural network, borrowed from the open source GPT-2 model created by AI research company OpenAI and trained specifically to write CAH cards,” The Verge reported. “The GPT-2 model is already trained on roughly 40,000 books worth of internet text to ensure it can reliably predict and fill out the next word or punctuation mark in a sentence with realistic effect. But then CAH went further and trained it using tens of thousands of its own cards.”

The company trained the AI on its cards, unpublished cards and suggestions from online. It stopped training the AI as soon as it could “consistently” create cards similar to the grammar and tone of the game.

“The writers sold 2 percent more packs, so their jobs will be replaced by automation later instead of right now. Happy Holidays,” the CAH management wrote.