Humans Vs. AI: Cards Against Humanity Writers Face Off With AI to Keep Their Jobs
Sun, April 18, 2021

Humans Vs. AI: Cards Against Humanity Writers Face Off With AI to Keep Their Jobs

CAH writers competed with AI on Black Friday / Photo Credit: Penty Photography (via Shutterstock)

 

The creators of Cards Against Humanity (CAH) were up for their annual Black Friday stunt as they competed with AI to see who could create the most popular news set of cards, which would be based on how many individuals pay for more $5 packs, wrote Nick Statt of The Verge, an American technology news network. Each person could upvote or downvote their favorite cards for each side on CAH’s official website before making a purchase.

If the human employees reigned victorious, they would receive $5,000 bonuses or get terminated if the AI won. The Verge said that CAH doesn’t have any plans to fire their writers if they lose, but the competition was still a clever stunt to emphasize on the human vs. machine narrative at a time when automation may pose a threat to millions of jobs, including writing, in the next decade. 

This year, CAH live-streamed the human writers' room while updating a live list of the most popular human and AI-generated cards that would have the potential to “make it into the eventual physical card packs.” Statt’s favorite card included: “Sitting in the back of the plane, smoking a cigar and reading the Flickr privacy policy,” which signifies the age-old debate of whether AI is for or against the Oxford comma. 

But CAH did not go easy. The AI the company used isn’t just your average random text generator, but a legitimate neural network, borrowed from the open-source GPT-2 model developed by OpenAI. It was already trained on nearly 40,000 books worth of internet text to enable it to predict and fill out the next word or punctuation realistically. 

CAH took it to the next level as it trained the model to write CAH cards using tens of thousands of its own cards. “We stopped the training once it could “consistently” produce cards matching the grammar and tone of the game,” as stated in CAH’s website. CAH then ran a filtering algorithm to ensure that the model’s AI-generated cards would match CAH’s standard format while avoiding ones “that were similar too existing cards.” The result? The human team narrowly beat AI by about $700 or so.