Interpretability
Interpretability is the degree to which a person can understand how an AI model works internally and how it turns inputs into outputs.
What is Interpretability?
NIST distinguishes interpretability, the meaning of a model's output in context, from explainability, the mechanism behind it, though the terms are often used interchangeably. Large neural networks are generally hard to interpret.
How it works
Researchers inspect model internals, test responses to controlled inputs or build simpler surrogate models.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions