An intrinsically interpretable model is one whose structure makes its reasoning transparent by design. Interpretability is not added afterward through a separate explanation method, it is rather built directly into the model. In such models, predictions arise from mechanisms that can be inspected and understood, for example:
- Clear coefficients in linear models,
- Additive contributions in generalized additive models,
- Explicit decision paths in trees,
- Human-readable rules in rule-based systems.
Thus, intrinsically interpretable models reflect a design choice: prioritizing transparency in the model architecture itself, rather than explaining complexity after the fact.
