Computational Trust Model

We developed a computational model capable of predicting—above human accuracy—the degree of trust a person has toward their novel partner by observing the trust-related nonverbal cues expressed in their social interaction.

Work began with a study to demonstrate that when people have access to the nonverbal cues of their partner, they are more accurate in their assessment of their partner’s trustworthiness. Confident that there is trust-related information conveyed through nonverbal cues, we furthered our investigation to identify and confirm a set of four nonverbal cues -- face-touching, arms-crossed, leaning backward, and hand-touching—indicative of lower levels of trust and demonstrated people’s readiness to interpret these same cues to infer the trustworthiness of a social humanoid robot. We continued our investigation into trust-related nonverbal signals and built temporal models to investigate the sequential interplay among the trust-related nonverbal cues. By interpreting the resulting learned model structure, we discovered that the sequence of nonverbal cues a person emits provides further indications of their trust orientation toward another person.

From this work, we gained a deeper understanding of trust-related nonverbal behaviors. We utilized this domain knowledge in the feature engineering process in designing a trust prediction model. We developed a model capable of accurately predicting the degree of trust a person has toward another individual by observing the trust-related nonverbal cues and sequences expressed in their social interaction. Our computational trust model achieved a prediction performance that is significantly better than our baseline models (including the a-priori baseline) and more accurate than human judgment.

The multi-step research process combined the strength of experimental manipulation and machine learning to not only design a computational trust model but also to deepen our understanding about the dynamics of interpersonal trust. This intersection of methodologies from social psychology and artificial intelligence research provides evidence of the usefulness of interdisciplinary techniques that push and pull each other to advance our scientific understanding of interpersonal trust.