LUCID: Exposing Algorithmic Bias through Inverse Design

AI systems can create, propagate, support, and automate bias in decision-making processes. To mitigate biased decisions, we both need to understand the origin of define what it means for an algorithm make fair decisions. Most group fairness notions assess a model's equality outcome by computing statistical metrics on outputs. We argue that these output encounter intrinsic obstacles present complementary approach aligns with increasing focus treatment. By Locating Unfairness through Canonical Inverse Design (LUCID), generate canonical set shows desired inputs model given preferred output. The reveals internal logic exposes potential unethical biases repeatedly interrogating process. evaluate LUCID UCI Adult COMPAS data sets find some detected differ from those metrics. results show shifting towards treatment looking into algorithm's workings, are valuable addition toolbox algorithmic evaluation.