The Latent Feature Made Me Do It: Adventures in Ethical AI

When building predictive models, a subset of variables or features typically combine to produce latent features that drive the majority of the predictive power. To ensure variables beyond simple protected classes (such as age or gender) are not driving biased outcomes in combination, a set of novel approaches and interpretable architectures will be presented to ensure your AI models go beyond superficial explainability and empower your teams to deeply understand your data and machine learning models to deploy AI responsibly.