With a single tweet, Andy Herd raised a question about the job security of every TV writer:
"i fed a recurrent neural network with the scripts for every episode of friends and it learned to generate new scenes"
The comic artist behind Pandyland shared scenes from an episode of "Friends" Monday that he trained a computer program to auto-generate.
"I was motivated by childlike fascination at the possibilities of machine learning and how it could be applied to humor," Herd said over email. "I think the scripts are actually better than some current TV sitcoms. I have received a lot of positive feedback."
To train the computer Herd used a neural network, a type of artificial intelligence in which the computer was fed every episode from the popular '90s comedy. The computer then learned patterns to generate new scenes. This technique is used in other instances, such as a popular Microsoft computer program in China that was trained to carry on conversations with users by studying existing conversations.
So, is the computer's work as good as Herd says? We asked a "Friends" superfan, John Kardasis, who estimated he's watched all 10 seasons more than 20 times and basically has all episodes memorized.
"There's no way this series could've gone 10 years with a computer writing it, " Kardasis said.
"The actual script wasn't funny. It was funny in that the computer was generating sort of gibberish."
Here's one such exchange:
"Phoebe: Wow lady! You're just gonna come over to him jumpy! (They start to cry.)
"Chandler: So, Phoebe likes my pants.
"Monica: Chicken Bob!
"Chandler (in a muffin) (Runs to the girls to cry) Can I get some presents."
"In all honesty, I don't think it captured any of the characters well," Kardasis said. "It was all over the place."
While Herd hasn't seen anyone else use such an approach to automatically generate a TV script, he imagines somewhere in the blossoming field of artificial intelligence it was attempted.
But Herd seems to have a ways to go before producing a usable TV episode that fans will watch.
