Intrinsic Honesty and Trust in AI
Description
This dataset is used in the paper "From Intrinsic Honesty to Trust in Artificial Intelligence (AI): How Moral Integrity Influences Trust in AI Systems". This paper uses a repeated trust game to explore trust dynamics in artificial intelligence (AI) systems. The experiment assessed how trust is formed and evolves with AI systems exhibiting different personalities. Participants engaged in repeated tasks, allowing the distinction between initial trust and trust over time. Findings indicate that socially oriented AI, with consistent and rewarding behavior, elicited the highest trust levels but was least reciprocated with trust compared to competitive and exploratory AI systems. Trust was also sensitive to perceived fairness, since participants were less trusting when AI had the same initial endowment. Trust recovery after a breach proved challenging regardless of the initial level of trust. The findings offer practical insights for designing trustworthy AI systems and informing AI governance policies. The study advances our understanding of trust in AI, offering actionable guidance for AI deployment in trust-sensitive environments.
Files
Steps to reproduce
The data were collected by asking subjects to participate in an online experiment, which was deployed on an Internet website for the first part and on oTree for the second part. The first part asks subjects to complete a dice task measuring their honesty. Then, subjects complete various questionnaires assessing their individual and socioeconomic characteristics. The second part asks subjects to answer three Trust Game tasks with an AI partner. Subjects plays as the trustor in the first task and the trustee in the second task. We study whether subjects trust AI in the first task and reciprocate the trust of the AI in the second task. The third task changes initial endowments to study whether subjects trust the AI when she benefits more from the interaction.