Reimagining ‘The Office’: An Experiment In A.I.

Original article can be found here (source): Artificial Intelligence on Medium

Reimagining ‘The Office’: An Experiment In A.I.

Hello fellow humans!

In the Spring of 2019, Elon Musk’s OpenAI research lab made quite the splash, announcing its GPT-2 artificial intelligence was too dangerous to release to the public. This A.I. was able to generate believable text, staying on topic and using proper grammar. Fears abounded that soon malicious bots would be writing novel and dangerous news articles, social posts, and more.

Well, that didn’t really come to pass, and OpenAI soon made their pretrained models available to the general public.

Naturally, we did what anyone would do and promptly fed the model all the sitcom scripts we could find. The end product? A brand new script for the hit show The Office.

How It Works

Conceptually, the GPT-2 artificial intelligence model isn’t too scary. It takes a sequence of tokens (words) and tries to predict the next token (word) in the sequence. For example:

  • Input: I ran
  • Output: I ran faster

After that, it takes the output and pipes it right back through the machine.

  • Input: I ran fast
  • Output: I ran faster than

It repeats that process until you have your great new sentence!

  • Final Output: I ran faster than a rocket-powered squirrel.

Easy, right?

Where Things Get Interesting

With GPT-2, OpenAI has release a pretrained model that is fully capable of generating organic text. In fact, you can see it in action on line. Right now. Right here.

But that’s boring.

Where things get interesting with GPT-2 is that you can train it unique text sets in such a way that your retrained model will match any desired formatting. For example, innovators around the world have used retrained A.I. models to generate: food recipes, screen plays, Twitter bots, and more!

A Brand New Episode of The Office

Naturally, we jammed as much lore from the hit classic, The Office, to see what we’d get out. Without further ado: The Office — The Lost Chapter!

The opening plays and the scene opens. We’re welcomed by Michael and Dwight in heated debate.

MICHAEL: (whispering) All the stuff that happens in this office. I love it. I love this chair. This is so comfortable. I love this system. I love this computer. I love this phone. I love this workspace.

DWIGHT: This is all mine.

MICHAEL: (jumps onto desk)

DWIGHT: This is my office.

MICHAEL: (to camera) Can you tell who’s ahead of his time?

DWIGHT: (stares)

MICHAEL: (to camera) Everyone, just hope that I can spend as little time here as I can.

DWIGHT: This is my desk.

MICHAEL: Yes. I just need to cross some t’s and dot some i’s. Alright, three people, three paintings, three drawings, three sounds. Why don’t you just walk away we’ll go our separate ways.

Sounds like there’s drama in Scranton! Hijinks ensue, I’m sure. Later in the episode, we’re happen upon Pam. She appears worried.

PAM: (to camera) I’ve been listening to some noise. I think it could be the receptionist.

MICHAEL: (scoffs) Yes. That’s Angela. She’s terrific. She’s so cute. She’s so sweet. You know what? I love her. I think she’s so important to this office. I don’t want her to be the victim of a hate attack. Because I don’t want to see her hurt again. You know what? I don’t want to see her again. I have to step away. I have to go.

DWIGHT: Jim!

JIM: Who is this?

DWIGHT: Meredith.

JIM: What’s she worth?

DWIGHT: $1 billion.

PAM: She’s not only the single greatest granddaughter of Abraham Lincoln, but she’s the only woman in this office who has never worked outside of her office’s hands.

DWIGHT: Great. She’s strong. She’s soft. She’s clean.

End scene.

And there we have it! A (short) episode of The Office generated entirely by artificial intelligence. Who needs screenwriters anyways.

Got Your Own Great Idea?

Want to find out what artificial intelligence can do for you? Let’s talk!