It Takes Two to Trust: Mediating Human-AI Trust for Resilience and Reliability

New technologies are destined to disrupt, but artificial intelligence (AI) has achieved unusual cultural impact, inspiring visceral fear in some, yet rapidly proliferating through pervasive adoption. But by its nature, adoption of AI necessitates more than mere acceptance: it requires trust. Trust surpasses cooperation; cooperation elicits predictability, but cannot enable the vulnerability described by human trust theorist Niklas Luhmann. The bond of trust must be developed through interaction, and humans inherit social constructs and societal norms that regulate verbal and behavioral communication, indicating internal state. AI must learn to read these cues to engender trust and recover when its human partner becomes distrustful. Our experimental platform, Hapti Bird, creates an environment for testing scenarios and observing interactions from which to better understand how humans trust each other, and how they trust AI. For this paper, we performed a variation of the Iterated Prisoner's Dilemma with haptic devices and sensory capture involving two human subjects in an embodied joint action paradigm, taking the form of a video game. From video-derived heartrate (HR) and heart rate variability (HRV) we predict in-game cooperativity between subjects up to 7 seconds into the future (71% F1 score). Facial expressions tell of significantly different experiences of subjects depending on the amount of time given to establish trust, and when that trust is broken. Our accumulated findings educated an AI, named Hapti Bot, to embody the formulation of trust between humans. Hapti Bot was trained using genetic algorithms, which continuously generated AI candidates within given parameters. This process mimicked biological evolution and produced a Bot optimized to thrive in the Hapti Bird environment.