Currently there is a prevalent problem in our justice system, as it is systemically oppressive towards certain groups, and these problems manifest themselves through mass incarceration and biased criminal justice verdicts. These injustices prompted us to use AI and technology to provide justice for all.
One tool used to investigate suspects is a lie detector. However, many scientists have explained these are not effective and can easily be exploited and eluded by suspects. We also wanted to provide a solution to help people spend time with their families and return home for the holidays. Moreover, sentencing crimes is an inherently biased task, as they are given by judges or juries, which is a subjective process. This can cause implicit or explicit biases to influence decisions, furthering systemic oppression.
Thus, we built an application, CourtsAI, to address all three of these problems, allowing families to be reunited and avoid the biases in our criminal justice system. Our application provides AI solutions to these issues, and therefore is a great regulator and assistive tool for our current criminal justice system.
What it does
Our web app uses Machine Learning to make the court system more fair. Our app is meant for government officials, however, anyone can use the deception detector. The app has 3 main sections, a deception detector, a bail decider, and a sentencing predictor. To access any of the tabs, a user must be logged in through our Firebase Auth.
The deception detector uses Tensorflow RNN Text Classifier to decide whether or not a passage of text is telling the truth and can be used by anyone. Our model was trained on web scraped data to find patterns from truths and deception. A user can also upload an image and allow the model to run on that, as we have OCR powered by Google Cloud’s Vision API. This is super useful for witness testimonies, or any words spoken by anyone in a court to detect hidden deception.
Our next section is a bail decider using a Tensorflow Linear Model and is meant for government officials. Because there is no way to regulate whether or not someone deserves bail, our AI model is able to predict whether or not they deserve bail based on their record. Every defendant has a record and it is stored as a csv file which all courtrooms have access to, making our app very accessible. The model is unbiased because it does not take into account variables such as race, age, name, sexuality, etc.
The last section is a sentence decider and can also be used by government officials. It is similar to our bail decider and uses a Tensorflow Linear Model and inputs the person’s csv record. Although there are many factors that we cannot always predict when sentencing, our model is a great regulator and can help make more fair sentences no matter the race, gender, name, or any factors that can cause bias.
How we built it
Overall we used HTML, CSS, VueJS, Tensorflow, TensorflowJS, Firebase Auth, and the Google Vision API. We created all of our models in Tensorflow and took out any factors that could cause bias in order to create the most fair system and something that could regulate the court system. The models were programmed in Python and exported to TensorflowJS. The web app was created using VueJS, HTML, and CSS. For our OCR on the deception detector we used the Google Cloud Vision API and Firebase Auth to make sure users are logged in to access the features.
Challenges we ran into
We ran into challenges while creating the models as this was our first RNN Text Classification Model. We also ran into issues with the OCR as it was also our first time using the Vision API, but after a good amount of debugging we were able to make it work.
Accomplishments that we're proud of
We are proud of creating an application that is able to have a real world impact and help create a more just and fair system for all. We are proud of building the entire application in the short allotted time frame, and we are hopeful for the potential impact our app may have.
What we learned
We learned how to use the Vision API as well as create a Text Classification model. We also learned how to use TensorflowJS, which was new since most often we use Tensorflow with Python.
What's next for CourtsAI
We plan on finding more data to improve our Deception Detector as well as improve on our other two models. From there we hope to host our application so everyone can use it, as currently we do not have access to powerful enough servers. We would also love to see our application be piloted in a court setting to see how it would perform.