The discussion highlights the complexities of ensuring non-discriminatory models, particularly when certain features may be correlated with sensitive attributes like race or gender. Stripping away these features can become a daunting task, as rich datasets often reflect these details in unexpected ways. The conversation also touches on the nuances of explainability, emphasizing that correlation alone does not render a feature illegitimate or illegal.
