ai-ux-patternslisted

# AI UX patterns — designing for non-determinism Deterministic software gives the same output for the same input, every time. UX assumes the system is right. **AI features are probabilistic:** they are sometimes wrong, sometimes unsure, and sometimes slow — and the UX must be designed *around that uncertainty*, not in spite of it. Google's People + AI Research (PAIR) guidebook and Apple's Human Interface Guidelines for machine learning converge on the same core: the job of AI UX is **earning and calibrating trust under uncertainty.** Four patterns do most of the work — **confidence signals, graceful failure, reversibility, human-in-the-loop** — plus **progressive disclosure** to keep control from becoming overwhelm. This skill is the *trust & uncertainty* layer for probabilistic features. It complements `hooked-ux` (the habit loop for any product); it does not overlap it. ## The mental model: calibrated trust The failure mode at both extremes: - **Over-trust** — the UI projects false confidence, the user stops checking, the model is wrong, and the cost lands on the user. Erodes trust permanently the first time it bites. - **Under-trust** — the UI hedges on everything, asks for approval constantly, surfaces uncertainty the user can't act on. The user concludes the feature is useless and turns it off. Good AI UX **calibrates**: confident where the system has earned it, honest where it hasn't, and always recoverable. Trust is built *progressively* — start cautious