Fail-safe

Fail-safe means building in a safeguard ahead of time so that when something goes wrong, the outcome is the safest one possible instead of the most dangerous one. Think of an elevator: if the cable snaps, it doesn't just plunge to the ground. It's designed so the brakes automatically clamp down and stop it in place. You can't prevent the failure itself, but you can control what state the system falls into when it fails.

Apply this to traffic lights and it looks like this: if a power outage knocks out a traffic light entirely, it's safer for all directions to flash and signal drivers to slow down than for one direction to freeze on green while another freezes on red. This principle matters even more for machines, vehicles, and software that make their own decisions. There will eventually be a moment when a sensor suddenly goes dead or the decision logic spits out a strange answer, and at that moment the system needs to be designed to stop or hand control to a human rather than keep pushing forward.

So fail-safe isn't a technique for building a perfect system — it's a technique for preparing for the moment an imperfect system fails. A classic example is a self-driving car: when its sensor signals conflict or its judgment becomes uncertain, instead of rushing toward its destination, it should be designed to pull over and come to a gradual stop.

If you want to get a feel for this yourself, try asking an AI chatbot:

"A self-driving car is in a situation where two of its sensors are giving conflicting readings while driving. What options should a fail-safe design offer in this case? Suggest three specific ones."

See if the answers converge toward risk-reducing choices like "stop," "slow down," or "hand control back to the human." That's a good way to get a sense of what fail-safe thinking actually looks like.