Feb 03, 2023 21:00:00

Research results that there is a limit to the number of times a robot that repeats mistakes can regain human trust

As the saying goes, 'The

Buddha's face is three times ,' it is said that no matter how gentle a person is, he or she loses trust in someone who repeatedly fails. A research team led by Connor Esterwood of the University of Michigan reported research results that it is difficult for robots that make repeated mistakes to regain human trust.

Three Strikes and you are out!: The impacts of multiple human–robot trust violations and repairs on robot trustworthiness - ScienceDirect
https://doi.org/10.1016/j.chb.2023.107658

How robots can regain the trust of humans after making mistakes
https://techxplore.com/news/2023-02-robots-regain-humans.html

In a study by Estherwood et al., 240 experiment participants were asked to work collaboratively with a robot in a virtual space, each placed behind a table with two displays and three buttons. I was. The display showed the current score of the robot and the participant's team, the processing speed of the box, and the serial number for evaluating the box presented by the robot.

Each team scored 1 point for each correct box placed on the conveyor belt and 1 point for each incorrect box placed on the conveyor belt. If the robot chooses the wrong box and the participant points out the mistake, the box will be shown as incorrect on the display, but the score will not increase or decrease.

In the experiment, the research team programmed the robot to occasionally make mistakes and lose the trust of the participants, and after making a mistake, the robot apologized. One of four trust-restoring behaviors was performed to participants: denial, explanation, and promise to prevent recurrence.

Experimental results showed that after the robot's third mistake, none of the trust-repairing behaviors were able to fully restore participants' trust or loyalty to the robot.

Robotic apologies, explanations, and promises to prevent recurrence showed some effectiveness, while denial was consistently ineffective in restoring trust.

``Robot researchers need to develop more effective trust restoration methods so that robots can better restore trust after making these mistakes,'' Estherwood said. He also said, 'The robot must be pre-trained in basic tasks before trust repair. If not, the robot may not be able to restore human trust.'

``There is a limit to the number of times trust can be completely restored,'' Estherwood said, pointing out that the results of this study apply not only to robots and humans, but also to humans.