Trust calibration
Calibrate trust by task risk
- Trust AI more for low-risk drafting and option generation.
- Trust AI less for facts, claims, policies, user evidence, and high-stakes decisions.
- The right trust level depends on task risk and verification available.
Weak versus strong
Low risk
Weak
No need to treat as truth.
Strong
Brainstorm 10 label options.
Medium risk
Weak
Send without team review.
Strong
Draft a feature brief, then check assumptions.
High risk
Weak
Use as legal, medical, financial, or safety authority.
Strong
Use as support only; verify with expert source.