interpretable

Interpretable is an adjective describing something that can be interpreted or understood. In everyday use it often refers to explanations that are clear and comprehensible. In the field of data science and artificial intelligence, interpretability denotes the extent to which a human can understand the cause of a model’s predictions or decisions.

Two broad notions are common. Inherently interpretable models are designed so their behavior is understandable without additional tools. Examples include linear models, decision trees, rule-based systems, and generalized additive models, where each component has a clear meaning and the overall decision process can be followed step by step. Post-hoc interpretability involves generating explanations for complex or opaque models after the fact, using methods that aim to approximate or rationalize the model’s behavior.

Key concepts used to assess interpretability include transparency (visibility into the model’s structure and parameters), decomposability (understanding individual parts such as features and rules), and simulability (the ability to reason through the model’s logic on a given input). Additionally, explanations may be local (pertaining to a single prediction) or global (describing overall model behavior).

In practice, there is often a trade-off between interpretability and predictive performance. Simpler, interpretable models may sacrifice accuracy on complex tasks, while highly accurate but opaque models require surrogate explanations or post-hoc analyses. Interpretability is particularly valued in high-stakes domains such as healthcare, finance, and law, where transparency supports trust, accountability, debugging, and compliance with ethical or regulatory standards.