Interpretability is how well people can understand why an AI system produced an output or decision. Deep-learning models can make this difficult because they do not provide clear insight into their decision-making. section 5.8
The stakes rise when AI decisions affect people. Lending and criminal justice are examples where opaque decisions challenge transparency, accountability, and trust. section 11.17
For systems that directly affect people, safeguards should include:
- Validation against technical and ethical standards, including reliability, fairness, and privacy. section 10.2
- Clear logic and explanations in plain language for those affected. section 10.2
- An understanding of how decisions were reached, not simply an assertion that the computer made them. section 11.17
