Calibrate
Design patterns for human–machine trust, for vehicles, robots and drones that move, lift, cut and fly. Each pattern starts with a real incident and ends with a design you can try.
Before
Before the machine acts, does the human agree?
- 04Appropriate FrictionBefore a high-stakes action, the interface makes the operator slow down and confirm.
- 06Intent PreviewBefore the machine acts it displays a proposed plan so a wrong plan is caught before it becomes a wrong action.
- 07Capability DisclosureThe machine says what it cannot do and where it will not run, then refuses to engage outside those limits.
- 11Readiness CheckThe system confirms the human can actually supervise, and escalates when they cannot.
During
While it runs, can the human see and predict it?
- 02The Glass BoxThe machine shows its reasoning, not just its output, so operators can trust or override it with context.
- 05Shared Mental ModelThe operator’s picture of what the machine is doing matches what it’s actually doing.
- 14Alert TriageAlerts are ranked by how urgently a human must act, so the critical one isn't buried.
- 15Stop AnywhereA stop works from any state, and stays stopped.
Transition
When control changes hands, is the human ready?
- 01The HandoffThe machine yields control on purpose and gives the human time to orient before they are accountable.
- 12Return of ControlThe machine takes control back visibly, only when conditions are right, and doesn't fight the human.
- 13Operator-to-Operator HandoverWhen responsibility passes between operators, the system carries its true state across.
Failure
When it can't do the job, what happens?
After and over time
What happened, and what is it doing to us over time?
