Advanced Concepts
Interpretable model
AI whose way of reaching predictions can be understood from its design or a manageable explanation.
Example
A small decision tree shows the conditions leading to an outcome.
Why people use it
It makes the path to a prediction easier for people to inspect.
What you'll hear
“Can we follow the steps behind this decision?”
What this means for you
Assess performance and fairness separately from ease of interpretation.
Can you control it?
Developer-only
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Does interpretable mean fair or accurate?
- No. Understandable logic can still be wrong or discriminatory.
- Can an understandable system still be too complicated in practice?
- Yes. A very large set of understandable parts can still be difficult to follow as a whole.
- Does explaining the system prove why something happened in the world?
- No. Its internal reasoning and real-world cause and effect are different questions.