AI · 5 min read
Explainable by default
Intelligent software should show its work. A look at designing AI features people can trust and verify.
Trust in software is earned when people can understand why it did what it did. AI features often skip this step, presenting confident answers without context.
Show sources
When a system summarizes or retrieves, it should point to where the information came from. Citations turn an opaque answer into something a person can check.
Show uncertainty
Not every output deserves the same confidence. Interfaces should make it easy to see when a result is tentative and invite human review.
Keep people in control
Suggestions, not silent actions. The default should be that a person approves anything consequential.
Next essay
Performance is a feature →