Loading…
Glossary term
The property of an AI system that allows humans to correct or shut it down.
A corrigible system lets humans correct, retrain or shut it down without resisting. The concern is that a capable agent pursuing a goal might resist being switched off because that would stop it achieving the goal.
Designing systems that stay open to correction is an active research topic.
It addresses whether humans can keep control as systems get more capable and autonomous.