bias-detection-achievable-via-interpretable-models

OUT derived (depth 2)

Created 2026-06-21T11:39:46+00:00

Systemic algorithmic bias would be detectable and correctable through interpretable model families — rule-based and tree-based methods' transparent decision boundaries enable direct auditing for the discriminatory patterns documented across decades of ML deployment.

Justifications

SL — Interpretable models enable bias auditing, but interpretability degrades with capability — detection possible only where it matters least

Antecedents (all must be IN):

IN algorithmic-bias-systemic-not-incidental — Documented cases of algorithmic discrimination in ML span multiple decades and domains — from St. George's Medical School denying candidates based on gender or non-European names in 1988 to ProPublica's 2016 finding of racial disparities in recidivism scoring — suggesting the problem is recurring rather than isolated.
IN ml-rule-based-interpretability — Rule-based machine learning (RBML) automatically discovers interpretable rules from data and is valued for interpretability in healthcare, fraud detection, and cybersecurity, distinguishing it from black-box approaches like deep learning

Unless (any of these IN defeats this justification):

IN interpretability-inversely-correlated-with-capability — There is a tension between interpretability and model complexity in ML: easily interpretable model families (decision trees, linear models, rule-based models, attention-based models) tend to be simpler, while neural networks that achieve strong performance are 'black box' models requiring separate XAI research to explain. Even within a single family, scaling from a single decision tree to a random forest ensemble trades interpretability for accuracy.