One of the critical requirements of the AI Act is Human Oversight. This section of legislation specific to High Risk AI Systems centres around the ability to interpret the AI model’s output. Let’s dive into some key AI Act articles concerning this topic.

In this article, we’ll walk you through some key concepts in explainable AI using a narrative example. We’ll apply the two most commonly used methods for interpretability, LIME and SHAP, and highlight some of their properties.