Inspiration
Existential risks have always fascinated me. That sounds a bit morbid, but it's quite an interesting issue philosophically! It also has tremendous implications for our actions today. However, the concepts behind existential risk are very difficult to understand. What does exponential growth really feel like? How much does 0.1% or even 1% matter to our lives? These issues are important to understand so that we can think clearly about our challenges ahead.
What it does
The first part of the hack is a wiki page that can contain user-uploaded articles. I utilized an open source framework, django-wiki, to build the page with Python. The second part of the hack is a Flask app that shows how far artificial intelligence has come. It can have conversations about the weather and colors, but also comes up with random philosophical thoughts, long-winded (not original) arguments (from Toby Ord), and things that Elon Musk may or may not have tweeted at some point!
How we built it
I learned what Django was and how to build a website on it. I relied heavily on the open source library django-wiki, and didn't know how best to alter the internal settings of the code (especially the CSS). For the AI chatbot, I relied on the open source framework for Chatterbot, using those frameworks to build a model and train it on podcasts related to existential risk. After that did not go as planned, I trained the model a bit more on Elon Musk's tweets.
Challenges we ran into
For the wiki, I didn't know what I was doing at first! I learned what Django and Flask were, and downloading the software I needed was difficult. I had difficulty installing the necessary frameworks to put together a successful project. Part of the challenges came from lack of knowledge or understanding when creating the hack; part came from not even knowing how to get started! I also couldn't find or explore enough conversational data to make Tim a formidable discussion partner. The model was trained on either podcasts, tweets, or a library of basic conversation. What would have been best would have been a data set of text or slack messages that it could train on. Instead, its functionality became its major weakness. I also didn't have time to filter the Twitter data, making that aspect of conversation even worse.
Accomplishments that we're proud of
I'm proud that I turned something in! After showing up late and not finding a team, I was scared that I'd have nothing to present today. I also learned a tremendous amount that I will utilize in future projects that I'm planning, starting with a personal website.
What we learned
I learned many things: how to operate on the command line, how to use Django (the basics), how to install many different packages (and figure out when installations go wrong), what kinds of things matter when under a time crunch (and what doesn't), as well as how to learn from those around you. The collaborative energy of the hackathon was the best part of the event.
What's next for learning about existential risk?
I would like to create more data visualizations of these tricky concepts: they're important and hard to understand! On the front end, I'm hoping to customize some of the CSS on both the chatbot and the wiki to be a bit neater and more colorful! Additionally, Wikipedia pages have "discussion" pages associated with each wiki, so I may ask the open source developers if they've thought of implementing more functionality in this area (and maybe helping)! Also, the chatbot Tim has some clear deficiencies, so I'm hoping to build a better bot on my own or with other code. He was built to be trained on conversational data; unfortunately, I don't have access to many conversations and that data would have to remain private. This makes his conversations pretty poor, as he can't anticipate what someone is about to say. That plus the minimal training makes him pretty deficient.
Log in or sign up for Devpost to join the conversation.