Unbiasify

Eliminating gender biases- one step at a time!

Comment 2

Inspiration

As a woman, ever found yourself marginalized at workplaces? Ever been a victim to systemic gender imbalance? Ever found yourself victimized by technology -for instance, you're trying to search salaries for Women in Tech on job searching platforms only to realize the results generated by AI are discriminating against you? Or are you a person of color who gets marginalized by AI which gives you incorrect outputs because it has been trained on biased datasets with white majority?

For some context, the symptoms of heart attack are different for men and women. Now if ML models are trained on biased datasets with largely men, a woman's symptoms would easily be misdiagnosed leading to life-threatening complication! Or consider you're trying to search for the average salary of a woman tech engineer and you realize that AI shows you significantly lesser salary for women when compared to men, leading to marginalization of women and discriminating against us!

This is what Unbiasify is trying to solve- Eliminating bias one step at a time!

The world we live in is growing at an exponential rate. We're living in the information age where decisions are based on data. In the flourishing age of Machine learning, while attempts are made to develop AI as neutral as possible, eliminating gender bias has become an increasingly crucial step to work on because our modern day decisions depend on data, and are also impacting the wellbeing of many minorities. As per SSIR, unfair allocation of resources, information, and opportunities for women was manifested in 61.5 percent of the systems (source:https://ssir.org/articles/entry/when_good_algorithms_go_sexist_why_and_how_to_advance_ai_gender_equity) "By 2022, 85% of AI projects will deliver erroneous outcomes due to bias in data, algorithms or the teams responsible for managing them. This is not just a problem for gender inequality – it also undermines the usefulness of AI” according to Gartner, Inc (source: https://www.forbes.com/sites/carmenniethammer/2020/03/02/ai-bias-could-put-womens-lives-at-riska-challenge-for-regulators/?sh=2697ab).

We need to take action to ensure our personal human biased do not find a way into ML and AI algorithms!

We need a solution that: -Irrespective of the majority or biases existing in the data set, trains the ML prediction in such a way that when you query, the output is completely unbiased! -Consequently, seek to empower women and other minorities as their concerns are now addressed are no longer being victimized by AI!

What it does

In simple words, Unbiasify offers a ML training platform, that seeks to eliminate bias from the dataset uploaded so that your queries are now bias-free! For the users, Unbiasify is a webpage that:

  • Allows you to upload a dataset, observe the existent bias in your dataset with the help of data visualization. -Shows you a visualization of how your unbiased dataset would look like after the backend compares various algorithms.
  • Once your data has been "unbiasified", users on query both biased and unbiased data and see for themselves how bias has been eliminated! The results of their queries now reflect reality and do not favor the majority!

How we built it

Building this sensitive yet a needed project required a lot of steps: -We began by reading research papers and articles to clearly identify the flaws in existing AI and the implication it has on the wellbeing of minority communities. Following this, we studied statistical models and algorithms on how we can turn our idea into reality.

  • Once we had a solid picture in mind, we started with the backend. We studied a dataset, classified it by comparing various algorithms and found the random forest classifier to work best, hence we leveraged it!

-Next step involved, un-biasifying the data using statistical parity (with and without sensitive attribute) with our sensitive attribute being Gender.

-With our backend all set up, we started with UI designing and front end development using Canva, HTML and CSS, and lastly integrated the front end and the backend together! This included parsing and fetching data both from frontend and backend which was a great challenge!

-Lastly, we registered for a domain : thetravelhacker.tech

Challenges we ran into

-A lot of our team members left the project halfway, so we had a lot of pressure building up towards the end of the hackathon.

-Integrating the front end with the backend was something new to us and we had to learn how to go about implementing that.

-A team member had a midterm coming up due to which she couldnt fully commit to the hackathon.

Accomplishments that we're proud of

-Successfully managing to integrate the front end and the backend. -Finding the most accurate algorithm to work with our dataset.

  • Being able to execute this project in under 24 hours!

What we learned

When researching for the project idea, we were shocked to see how prevalent disparities in the datasets used to train ML models. We were grateful to have learnt how to integrate the backend with front end and UI styling.

What's next for Unbiasify.

Unbiasify now is a platform that can extensively leveraged in any sector where systemic discrimination, marginalization, or stereotyping is prevalent. In the future, we hope to see Unbiasify collaborated with AI ethics researchers and equity advocates to make sure its as accommodating and inclusive as possible. Additionally, we see it in the future as a proper software with interactive and dynamic features that can be downloaded to use by any ML researcher, any application developer, teacher, students or anyone genuinely interested in the field of ML/AI and wants to be an advocate for eliminating gender bias!

Built With

Share this project:

Updates