Betraying someone and lying about it is pretty low 鈥� especially if you鈥檙e a robot. An experiment on interactions between humans and robots reveals just how easy it is for robots to regain trust after a misdeed by fibbing.
The devious robot in question is NAO, a 58-centimetre-high humanoid that moves and interacts with people. Sarah Sebo at Yale University and her colleagues arranged for 82 people to compete for points against NAO in an asteroid-shooting computer game.
In some rounds, a special asteroid blaster was awarded to either the human or NAO. It could be used to get bonus points when shooting an asteroid or for temporarily immobilising the opponent, yielding no points.
Advertisement
Ahead of 10 games played together, the robot promised it wouldn鈥檛 immobilise the player, but betrayal was programmed into its nature 鈥� it always broke its promise in round three.
To test the best route back from betrayal, the robot framed the immobilisation as a mistake with half of the human players. 鈥淥h no! I hit the wrong button,鈥� it would say. With the other players, it yelled: 鈥淵es! You鈥檙e immobilised!鈥�.
It then either apologised or denied it had broken its promise to the players in each group.
People were twice as likely to take revenge by immobilising the robot in the following round if they thought the robot鈥檚 betrayal was on purpose. Similarly, if NAO simply denied its actions, people were also twice as likely to seek revenge immediately afterwards.
鈥淧ersonally, I think robots should be 100 per cent honest and not lie ever,鈥� says Sebo, but the results suggest an advantage in lying to a person to optimise their trust. Framing a betrayal as incompetence could lead to a robot being treated better by people.
It is an ethical dilemma that we will increasingly face as robots become more sophisticated and humans start to view them as capable of having independent motives, says Sebo.
The team presented the work at the , South Korea, earlier this month.
