Issue Behavioural overtrust becomes irreversible once AI applies physical force

Summary

In a sidebar interview, Ohio State engineering dean Ayanna Howard explains that what people say about trusting AI often differs sharply from how they behave. In her studies, participants worked with robots deliberately programmed to make mistakes and afterwards said they did not trust them. Their observed actions, however, showed that they relied on the robots anyway. Howard warns that this behavioural overtrust turns dangerous once AI has a physical body, because robots exert force on their surroundings and the effects of their actions may not be reversible. She also points to real-time processing as a problem, since a delay of a second or two that is acceptable in a chat can make a walking robot drop something, crash or hurt someone. Her remedy is to keep a human in the loop somewhere at all times, which she concedes does not guarantee safety but helps counter overtrust.