The feeling of being classified: raising empathy and awareness for AI bias through perspective-taking in VR

Abstract

In a world increasingly driven by AI systems, controversial use cases for AI that significantly affect people’s lives become more likely scenarios. Hence, increasing awareness of AI bias that might affect underprivileged groups becomes an increasing challenge. As Virtual Reality has previously been shown to increase empathy through immersive perspective-taking, we conducted a laboratory study in which participants were confronted with a biased Wizard of Oz AI while embodying personas that varied widely in their ability to achieve high financial credit scores due to their age and gender. We found that participants embodying personas in VR felt significantly more empathy toward the characters they embodied and rated the AI as significantly less fair compared to a baseline condition in which they imagined to be these characters. Furthermore, we investigate differences between embodied personas and discuss qualitative results to gain insight into the participant’s mental model creation.