Bias in the Black-Box

Worked with a teammate to create a website and interactive example to showcase the dangers of bias in new machine learning AI models.
Created for CS3043 Social Implications of Information Processing.
Screenshot of page header, "How to avoid bias in your data driven AI solution"

Website header and top of home page

Screenshot of website, contains inputs for a location name, a monthly crime count, a population size, and a diversity percentage, also contains a button to get an AI prediction

User input for AI prediction

Centering on a primary interactive example, the user can engage with a dangerously biased AI tool, making clear how impactful problematic development practices can be.
Beginning with an AI machine learning algorithm written in Python, all scripts were ported to PyScript for web hosting, alongside a custom HTML and CSS front end.
Screenshot of PyScript code, the code integrates front end with python back end

PyScript code written to integrate front end with PyScript backing

Screenshot of a code snippet and analysis of that code. The code chooses an AI training model and the caption describes the issues related to the choice of a problematic model
The page also includes a section dedicated to analyzing the errors in the example code. By going section-by-section the page describes each individual issue and how they might be approached or solved in new projects.
↑Back to Top