I love finding new ways to create and animate characters. Recently I dipped my toes into AI to see if a robot could help me create fun content. One of the first things I tried was feeding it a few Dumbest Scientist scripts. Here's an example of what it came up with:
I’m a big fan of the laws of physics. They’re so simple, yet so elegant. For example, the law of gravity states that all objects fall at the same rate regardless of their mass. This is a very useful law. It’s how I know that my cat weighs the same as a bowling ball.
The law of gravity is also why I’m not allowed to play with my cat anymore.
It did a pretty good job of figuring out the general structure I used for the monologs as well as pretty much nailing the dry absurdity.
But sometimes it would go off the rails a bit:
I’ve always been fascinated by the way that living things work. I remember when my mother was pregnant with me, she would often say “I’m so sick of being pregnant!” and I would reply “You should try being a fetus! It’s much worse!”
So I started thinking about the kind of character that could use that weirdness to its advantage. I thought about Eugene Goostman, the chatbot who passed the Turing test, and how he was intentionally designed as a 13 year old to account for grammatical errors and an immature understanding of how the world works. It seemed to me that if the character was a robot themselves, that would help contextualize the oddities. Here's the result:
The Process
The AI I used for this is GPT-3. It is trained on approximately a bajillion gazillabytes of text, I just have to give it some examples of what I want it to do. I started by writing three original prompts. Here's one:
Topic: Driving
As a robot, I have a very difficult time identifying traffic lights, which makes driving extremely difficult. Also, it’s very difficult to adjust the air conditioning with my claws.
Each example had a topic. I then added another topic (in the case of the comic above it was "children") and let the AI figure out what comes next. The results were delightful and I decided to create a little webcomic out of it.
About 80% of the results were useable. Very few were incoherent and a few were offensive, but most of the time the reason for rejection was because they just weren't funny or interesting. I did not alter any of the text generated by the AI. If I didn't like something, I rejected the whole thing.
I then took the text generated by the AI, divided into panels myself and illustrated based on the text. I tried a few where I gave the AI examples with the text already divided into panels, and it broke its results into panels as well, but I felt like it was too restraining and forced the AI into more formulaic responses.
The final result is a sweet little back and forth collaboration between me and a real life robot.
Comments