Trump Guessr

Inspiration

Inspired by Jordan Peele's Obama impersonation, I sought to create an accessible trivia game which would highlight the ever-increasing humanity of neural networks. Using AI, Peele was able to create a computer-generated video of Obama saying things that he never said. While the might initially seem superficial, this project highlights how exploitable this kind of technology can be. Instead of impersonating certain figures through video, I opted to use Trump's thousands of tweets, since they are easier to work with.

What it does

My project leverages the TensorFlow library in order to synthesize brand-new tweets using a recursive neural network trained on over all 33,000 of Trump's tweets. The robust algorithm provides the support for an attention-grabbing trivia game that shows one fake (auto-generated) tweet, and three real ones, giving the user a score depending on fast they choose the correct answer. Aside from the technologically-intriguing backend, the UI provides for a tasteful experience that keeps the users wanting more.

Potential Impact/Significance

The rationale for having the user garner more points the faster they answer is as follows: Even instantaneous gaps in critical thinking can have large consequences in society. Additionally, since users tend to have short attention span while reading news headlines, quickly identifying misinformation is an important skill. It encourages critical thinking in a time where misinformation and fake news is spreading rapidly in our media. Being able to differentiate between fake and real Trump tweets will encourage users to be more aware of the information that they consume, helping people deal with the influx of information that they will receive.

How I built it

I built this website from the ground-up using pure HTML, CSS, and Javascript. I delved into the intricacies of the TensorFlow's open source machine learning library in order to realize the success of my project. In order to save my work, I used GitHub as a Version Control Protocol.

Challenges I ran into

One large issue I ran into when managing the Natural Language Processing was setting up the structure and parameters of the Recursive Neural Network in order to effectively generate believable tweets. Additionally, when developing my website, I experienced trouble with creating faulty cross-origin requests while fetching the real tweets. I was asynchronously fetching a public repository of tweets from another url, which violated the protocol of Cross-Origin Resource Sharing (CORS), a security mechanism which prevents websites from downloading resources hosted on a different server. I spent a lot of time trying to fix this issue and ended up having to store a large 2-dimensional array locally, meaning I had to parse this efficiently in order to quickly serve the real tweets.

Accomplishments that I'm proud of

I'm proud of my success in my first venture into machine learning with TensorFlow. As a beginner, I was overwhelmed with the complexities of back-propagation and recursive neural networks for the natural language processing of our 45th POTUS Donald Trump's 33,000 tweets. I was also happy that I created something engaging and informative yet refined in UX/UI design.

What I learned

In the age of information, machine learning is becoming one of the requisite skills for engineers. Tensorflow is an immensely powerful tool, one which I hope to utilize in the future. Thankfully, this project allowed me to obtain a base knowledge of machine learning. I hope to increase my knowledge base in the coming years, potentially by making improvements to this project. Aside from the technical knowledge that I gleaned through this project, I also learned non-technical skills that are necessary to achieve a real-world impact in society. This project was initially done by one person. However, once I reached out to peers with similar interests and formed a group, incorporating the ideas of the entire team resulted not only in a creative website but the “buy-in” of every member. Not only did we enjoy the camaraderie engendered through our combined efforts, we came to understand that the energy developed in working together allowed us to transcend what could have been accomplished by one individual. Freed from the burden of thinking I have to make an impact on society on my own, I look forward to working with others by collectively making contributions because it is much more motivating and realistic. The old adage was proven to be true once again: teaching others is the best way to learn. My project attempted to increase information literacy and awareness, and through creating this project, I learned a lot about these topics myself. I have learned that political awareness and being able to filter out incorrect information is a valuable skill, and one which I want to bestow on others to make a real-world impact.

What's next for Trump Guessr

One step I am looking forward to implementing is tweets from figures besides Donald Trump. While he is well-known for his use of Twitter, there are lots of other high-profile figures and accounts whose tweets users receive information from. Extending to Twitter accounts beyond Trump will demonstrate the broad potential impact of the exploitation of AI. One implementation could be allowing users to search for an account they’re interested. Rather than listing specific accounts for them to choose from, giving users the ability to search ANY account further emphasizes how easily AI can be abused to spread misinformation. Another design change would be making the options look like actual tweets.