Table of Contents

Notes

1. Introduction

• Blackbox nature is problematic
  • Model interpretation is seemed to claimed as a solution

YET! Interpretability is poorly defined

• What interpretability means??
• Why is it important

Despite this ambiguity, papers claim interpretability 😅

So there are only few options about reality :

1. the definition of interpretability is universally agreed upon, but it’s not set in writing
2. the term interpretability is ill-defined, and thus claims regarding interpretability of various models may exhibit a quasi-scientific character

Our investigation of the literature suggests the latter to be the case

Both the motives for interpretability and the technical descriptions of interpretable models are diverse and occasionally discordant, suggesting that interpretability refers to more than one concept

• Many papers propose interpretability as a means to engender trust
• Other papers suggest a connection between an interpretable model and one which uncovers causal structure in data
• The legal notion of a right to explanation offers yet another lens on interpretability.

Often, our machine learning problem formulations are imperfect matches for the real-life tasks they are meant to solve Eg:

• even though our goal is to discover causal relations to a particular disease, simply minimizing only error could leave to uncovering spurious correlations
• Distribution shift