Consciously UnBiased

Battling unintentional bias!

Comment

Inspiration

We all come from different backgrounds, no matter how similar the access to the resources, because of the content available and the number of interpretations available, it is impossible to not identify a little differently. Having cleared this, let's not forget even without the content availability, this world is a very diverse place with people coming from not just different backgrounds, communities, financial backgrounds, etc With the world going more global with every new initiative, it is very important for communities in different professional spaces to be inclusive to be able to bring out the best in people and make sure they move forward. A lot of people , including us, who are super inclusive and want to be a part of the safe space for diversity to exist, end up saying or doing things that are biased completely unintentionally and unconsciously. We have recently started interacting with the global communities and realize something new in every conversation, especially the things we usually qualify as normal.

What it does

It helps you understand what all you might have said unconsciously that either biased or stereotyped some particular community or person. It shows you all the things you said in a conversation that might be considered as offensive biased against some communities without even realizing it.

How we built it

We have built it using a chrome extension which records your microphone input to get your part in the conversation, an application to display all the processed and analyzed statements where some bias may have occurred to help you identify them and the main block of our project, the NLP model which receives an audio file from the extension, performs analysis on it and classifies the statements as either biased or unbiased. Once classified it passes on these statements to the app, so that the user can view it. We also built a web app that serves the same functionality as the Chrome extension as alternative for someone who doesn't want to download the extension.

Challenges we ran into

  • It took a really long time to figure the NLP Model and get the classification to what it is right now. We had very less experience working with neural networks or the BERT model which we're using by fine-tuning.
  • None of had any experience with chrome extension and it took a lot of research and trial and error to get that to what it is now.
  • The display application was supposed to help people practice their speech to consciously remove bias but had to be scaled down to a display application, given the time constraint.

Accomplishments that we're proud of

All of it! From the idea, design thinking to the actual implementation. Especially the workarounds to get the minimum things working. Plus getting an audio and then cleaning it to get the right text on which the model can be used, was a tough nut to crack! In the end, after being awake for more than 30 hours we're beyond happy to see the entire process come together and work (somewhat) how we imagined it! :D

What we learned

  • There's so many things we learnt this weekend! Starting with actually building a neural network, actually implementing it on a data to add to this, preprocessing of an audio file to make it compatible for our NLP model.
  • Building a chrome extension was a first time for all of us but I guess we pulled it off pretty well :)

What's next for Something Something Bias

  • Right now the NLP model only displays whether bias has occurred or not and we want to add the different types of biases that occur so the user has a very clear picture of what needs to be worked on.
  • We also want to ramp up the app with a VR character bot and give the user a very real experience of having a conversation where the bias can be identified and corrected consciously.
  • Adding a hardware component like a vibration motor to detect bias in real-time i.e. during the meeting so that you're instantly aware and can make sure you correct it!
Share this project:

Updates