Advanced Concepts
Black-box model
AI whose way of reaching an answer is difficult for people to understand.
Example
A system predicts risk without an easily interpretable chain of rules.
Why people use it
It describes AI whose reasoning is hard to inspect directly.
What you'll hear
“We can see the answer, but not easily explain how it arrived there.”
What this means for you
Ask for suitable explanations, testing and controls for the decision being supported.
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 black-box mean the AI system is inaccurate?
- No. Transparency and predictive accuracy are different properties.
- Can we still test whether it works?
- Yes. Its answers can be compared with known results even when the internal process is difficult to explain.
- Does an explanation beside the answer reveal the real process?
- Not necessarily. An explanation may be a simplified description or a separately generated account.