House of Minds

THE ACTUAL DATASET IS IN THE GITHUB UNDER participants.csv

Inspiration

The idea behind this project came from us wanting to find out if humans can actually be random. People often believe that they can make random decisions, but we might have some bias we don't realize. We wanted to see whether human generated randomness could be told apart from computer generated randomness.

What it does

House of Minds is an interactive web application with 2 main parts. The first part is where you can try to generate your own randomness. You can select 6 coins to be heads/tails and even choose a number between 1-10 and 1-100. Afterwards, you can see how you compare against the people I surveyed for my dataset.

The other part is where you can play a game. You play a 10 round guessing game where you try to determine whether a set of 6 coins was generated by a human or by a computer.

How we built it

I collected data from 36 people at the TAMU Datathon. I went up to people and asked if I could I a few questions. Once they said yes, I asked them to choose randomly between heads and tails, I asked them to choose a number between 1-10 and 1-100, and asked them to try to get to 10 seconds in their head without a timer.

To build the actual application I used Python and streamlit. It uses the collected dataset to get the participant statistics to generate the comparisons. I used databricks Genie to help design the frontend for my application.

Challenges we ran into

A big challenge I ran into was that I was working with a small dataset. Although the 36 people I got were very useful, 6 coinflips a person did not provide enough information to build a highly accurate human vs computer classifier. I initially looked at more complicated approaches to modeling human behavior, but they did not consistently outperform simpler baselines. This led me to focus on a model that could demonstrate a real pattern with the information we had.

Accomplishments that we're proud of

I am proud that we created an original dataset from scratch. Since this is the first time we have done something like that, it was really fun and interesting to see the process of collecting our own data. Also, instead of only doing visualizations, We built an interactive game that users can use to compare themselves against real people which I believe is extremely amazing.

What we learned

The project taught us that collecting data and building a useful model can be 2 different challenges. Even when the dataset contains interesting patterns, those patterns might not have reliable predictions. What was really interesting is just how much more heads was chosen first instead of tails, but that doesn't mean that every sequence will start with heads. Most importantly, I learned the value of designing an application around what the collected data can support.