My original goal was to use a corpus of fictional stories with machine learning that the user could interact or “chat" with. As I started researching more about machine learning, this project turned into an exploration of “creativity."

I made multiple prototypes,

one with Rita.js

and

one with just p5.js.

Both were helpful in driving what I wanted the final interface to look like, but I decided that I had to use machine learning in order to actually function in the way I wanted.

To conceptualize my underlying questions regarding creativity and machine learning, I used Firebase to store user inputs to be visualized at the end of each user experience.

I decided on using charRNN from ml5.js since it was beginner friendly and I was able to find a lot of easy-to-understand examples. A big thank you to Julie Lizardo for help in training the model!

Following ml5 documentation, I was able to use the model with my past ideas and created: a “chatbot" that came up with new text based on user input and stored that input. With each new user, previous users' inputs would be recalled and compared against.

Essentially, users would co-create a story with the chatbot and see how similar their story was to previous stories also made with the chatbot.