Interpretability in Intelligent Systems – A New Concept?

Abstract

The very active community for interpretable machine learning can learn from the rich 50+ year history of explainable AI. We here give two specific examples from this legacy that could enrich current interpretability work: First, Explanation desiderata were we point to the rich set of ideas developed in the ‘explainable expert systems’ field and, second, tools for quantification of uncertainty of high-dimensional feature importance maps which have been developed in the field of computational neuroimaging.