Human-AI Trust Updating: A Dynamic Trust Model Based on the Extended Kalman Filter

In addition to the behavioral indicators mentioned above, the recorded content also includes: AI ratings the trustworthiness of believers Expected return on each round of investment A written explanation of the reasons for investment decisions Prediction of the next round of refund amount 3. Publicly available datasets Using the trust behavior data published by Rosenberger et al. (2020), which includes two independent experiments: Experiment 1: N=54 Experiment 2: N=55 Both experiments used classical trust games, with participants as investors engaging in 52 rounds of repeated interactions with two types of pre-set believers (generous vs. selfish).

Highly credible believers return 100% -200% of their investment amount with increasing probability, while low credibility believers return 50% -100% with increasing probability. In addition to the investment amount per round (1-10 units), this dataset also includes subjective evaluations collected every 10 rounds: trust attitude, expected return, and judgment certainty score. Data file description study1.xlsx： subj_idx, Reliable, fair, trustworthy, observation, return, all, cond, feedback, round, trust study2_Dataset.xlsx： subj_idx, observation, cond, feedback, round, trust study2_AI.xlsx： subj_idx, observation, cond, feedback, round, trust, Decision reasons, next round of return prediction