COURSE 26L1100% FREE
Verified 2026-08-14

Uncertainty in the UI

Showing users when the system is unsure—sources, hedges, low-confidence states, and “verify this” cues—without fake precision meters. HAX guidelines stress making clear **what** the system can do and **how well**, and explaining why it d...

What it is

Showing users when the system is unsure—sources, hedges, low-confidence states, and “verify this” cues—without fake precision meters. HAX guidelines stress making clear what the system can do and how well, and explaining why it did what it did when wrong. PAIR patterns include setting expectations and deciding how (or whether) to show model confidence.

<!-- IMAGE: answer with sources + “may be wrong” vs fake 97% badge -->

HIGH PRIORITYUI WIREFRAME COMPARISON
◷ IN PRODUCTION

Visual Spec & Architecture Diagram

Good vs bad uncertainty UI: Bad = fake precise 97. snarky certainty on ambiguous question; Good = confidence band, 'I'm unsure', cite sources, invite confirm. Side-by-side phone mockups.

Educational Focus: Chapter thesis is UI for uncertainty—must be shown.

Why it matters

Overconfident UI copy causes overtrust (Course 29 medical overtrust parallels). PAIR notes that admitting a prediction could be wrong may reduce trust in that instance but improve long-term calibrated reliance.

How it works (plain)

Prefer: citations, alternatives, explicit unknowns, confirmations for irreversible acts. Avoid: decorative 97% badges without calibration. Explain benefits in user language—not model architecture lectures (PAIR: explain the benefit, not the technology).

Everyday example

Maps saying “uncertain location” vs a pin that looks exact.

Try it

Redesign one AI screen to add a source link and a verify checklist. Remove any uncalibrated percentage.

Myths

⚠️ Myth: Any confidence number helps.
✓ Reality: Uncalibrated scores mislead—sometimes worse than none.
⚠️ Myth: Hiding uncertainty increases retention ethically.
✓ Reality: It’s a dark pattern with real-world harm.

Sources