Interpretability
Domain: Safety & Ethics
🎯 Motivation
How easy it is for a human to understand the AI’s internal decision-making process. More interpretable = more trustworthy.
📋 Concrete Example
Simple decision tree: ‘If age > 18 and income > 30K, approve loan.’ Very interpretable. Deep neural network with 175B parameters: hard to interpret. Trade-off between performance and interpretability.
🔗 Analogy
A clear recipe vs a magic black box - the recipe (interpretable) shows each step. The black box (uninterpretable) gives results but not reasons. Interpretability is the recipe.