ai-trust--transparencylisted

# AI Trust & Transparency Design interfaces where users can see into the AI's reasoning, calibrate their trust appropriately, and verify claims independently. The GLASS framework makes AI decision-making visible without overwhelming users. ## Core Principle Trust is not a boolean. Users should not "trust AI" or "distrust AI" - they should develop **calibrated trust**: high confidence when the AI is reliable, healthy skepticism when it's uncertain. Your job is to give them the signals to calibrate correctly. --- ## The GLASS Framework | Letter | Principle | Design Question | |---|---|---| | **G** | Ground in Sources | Can the user trace every AI claim back to a verifiable source? | | **L** | Layer Explanations | Can the user get a 5-second answer AND a 5-minute deep dive? | | **A** | Advertise Limitations | Does the interface proactively tell users what the AI is NOT good at? | | **S** | Show Confidence | Can the user see how certain the AI is about each output? | | **S** | Support Override | Can the user correct, override, or reject AI outputs without friction? | --- ## The Trust Calibration Spectrum Design for the right trust level - not maximum trust. | Trust Level | User Behavior | Design Goal | When Appropriate | |---|---|---|---| | **Over-trust (Automation Bias)** | Accepts all AI outputs without checking | Introduce friction to encourage verification | High-stakes decisions (medical, financial, legal) | | **Calibrated Trust** | Verifies selectively based on co