XAI techniques can detect blackbox bias

Screenshot 2018-01-20 16.18.28

A good up to date review of how bias is hiding in ML models and what we can do about it, came out this week.  That article leaned heavily on the recent academic paper “Detecting Bias in Black-Box Models Using Transparent Model Distillation“.

This is a good example of how XAI has broader applications beyond just delivering an explanation to an end-user.  In this paper they described using an XAI technique (interpretable tree additive models with pairwise interactions) as part of a clever mechanism to identify systemic bias in a black-box model.

Leave a Reply