ai-error-resiliencelisted

# AI Error Resilience Design AI products that fail gracefully, communicate uncertainty honestly, and help users recover without losing trust. The RECOVER framework treats AI errors as a design material, not a bug to hide. ## Core Principle Traditional software has bugs. AI has **probabilistic outputs on a spectrum of correctness.** You cannot design AI UX using binary error/success patterns. Instead, design for a continuum: right, mostly right, partially right, uncertain, wrong, and dangerously wrong. --- ## The RECOVER Framework | Letter | Phase | Design Question | |---|---|---| | **R** | Recognize | Can the system detect when its output may be unreliable? | | **E** | Express Uncertainty | Does the interface clearly communicate degrees of confidence to the user? | | **C** | Contain Blast Radius | If the AI is wrong, what's the worst that can happen? How is damage limited? | | **O** | Offer Alternatives | Does the user get a Plan B when Plan A might be wrong? | | **V** | Verify Collaboratively | Can the user easily check, correct, or confirm the AI's output? | | **E** | Evolve from Errors | Does the system learn from this error type to prevent future occurrences? | | **R** | Restore Confidence | After a failure, how does the product rebuild the user's willingness to try again? | --- ## AI Error Taxonomy Not all AI errors are created equal. Each type requires a different UX response. | Error Type | Description | Severity | UX Response Pattern | |---|---|---|---| | **