SECAI Core · phase 6 of 15
Explainability
An AI principle of the extent to which stakeholders can understand the reasoning behind a system's output
The Explain card
- Plain English
- Explainability is the degree to which stakeholders can understand the reasoning behind a system's output: why this decision, based on which inputs and factors.
- Example
- A loan model declines an applicant. An explainable system reports that income-to-debt ratio and short credit history drove the decision; an opaque one just says no. The first can be checked and challenged, the second cannot.
- Why it matters
- If you cannot explain an output, you cannot tell a correct decision from a manipulated or biased one. Explainability supports debugging, regulatory review, and spotting when an attacker has nudged the model.
- Hook
- Explainability is the model showing its working, not just the answer.
Where it sits in the deck
Phase 6: Model Quality, Ethics, and Responsible AI Principles
Before deploying a model, understand the quality and ethical dimensions — bias, fairness, explainability, and the human oversight needed to govern outputs responsibly.
