What are the differences between intrinsic explainability and post hoc explainability in the context of Explainable AI (XAI) in finance?

Intrinsic explainability refers to models designed to be inherently interpretable, allowing users to understand their decision-making processes without additional tools. In contrast, post hoc explainability involves techniques applied after a model is trained to interpret its predictions, which can be used with various complex models but may not fully represent their behavior.

Intrinsic explainability is characterized by models that are simple and transparent by design, such as linear models and decision trees. These models allow users to easily understand how decisions are made from the outset. On the other hand, post hoc explainability applies techniques after a model has been trained to provide insights into its predictions, making it versatile for complex models like deep neural networks. While post hoc methods can help in understanding these models, they may not accurately reflect the model's behavior due to approximation involved.