Robot deception is an understudied field with more questions than answers, particularly when it comes to rebuilding trust in robotic systems after they have been caught lying. Two student researchers at Georgia Tech, Kantwon Rogers and Reiden Webber, are attempting to find answers to this issue by investigating how intentional robot deception affects trust and the effectiveness of apologies in repairing trust.
Rogers, a Ph.D. student in the College of Computing, explains:
“All of our prior work has shown that when people find out that robots lied to them — even if the lie was intended to benefit them — they lose trust in the system.”
The researchers aim to determine if different types of apologies are more effective at restoring trust in the context of human-robot interaction.
The AI-Assisted Driving Experiment and its Implications
The duo designed a driving simulation experiment to study human-AI interaction in a high-stakes, time-sensitive situation. They recruited 341 online participants and 20 in-person participants. The simulation involved an AI-assisted driving scenario where the AI provided false information about the presence of police on the route to a hospital. After the simulation, the AI provided one of five different text-based responses, including various types of apologies and non-apologies.
The results revealed that participants were 3.5 times more likely not to speed when advised by a robotic assistant, indicating an overly trusting attitude toward AI. None of the apology types fully restored trust, but the simple apology without admission of lying (“I'm sorry”) outperformed the other responses. This finding is problematic, as it exploits the preconceived notion that any false information given by a robot is a system error rather than an intentional lie.
Reiden Webber points out:
“One key takeaway is that, in order for people to understand that a robot has deceived them, they must be explicitly told so.”
When participants were made aware of the deception in the apology, the best strategy for repairing trust was for the robot to explain why it lied.
Moving Forward: Implications for Users, Designers, and Policymakers
This research holds implications for average technology users, AI system designers, and policymakers. It is crucial for people to understand that robotic deception is real and always a possibility. Designers and technologists must consider the ramifications of creating AI systems capable of deception. Policymakers should take the lead in carving out legislation that balances innovation and protection for the public.
Kantwon Rogers' objective is to create a robotic system that can learn when to lie and when not to lie when working with human teams, as well as when and how to apologize during long-term, repeated human-AI interactions to enhance team performance.
He emphasizes the importance of understanding and regulating robot and AI deception, saying:
“The goal of my work is to be very proactive and informing the need to regulate robot and AI deception. But we can't do that if we don't understand the problem.”
This research contributes vital knowledge to the field of AI deception and offers valuable insights for technology designers and policymakers who create and regulate AI technology capable of deception or potentially learning to deceive on its own.