Data for: Who Defines the Risks of Artificial Intelligence?

Description

This dataset contains raw data from a 2 × 6 between‑subjects online experiment conducted in April 2026. Participants were 635 Chinese adults recruited from the Wenjuanxing online panel. The experimental design manipulated two factors: AI risk visibility (visible vs. invisible) and information source (AI, scientists, governments, AI companies, traditional media, or science key opinion leaders). The data include pre‑test and post‑test measures of trust in AI and trust in science, as well as measures of perceived information credibility, fear of AI, AI usage frequency, social media use, perceived relevance, and demographic variables. The dataset is provided in CSV format and includes a codebook with variable labels and value descriptions. It supports replication of the main analyses reported in the manuscript, including ANOVA, mediation, moderated mediation, and three‑way interaction models.

Steps to reproduce

We conducted a 2 × 6 between‑subjects online experiment in April 2026. A total of 638 Chinese adults were recruited from the Wenjuanxing online panel. After data quality screening (attention checks, response time filtering, and straight‑lining detection), 635 valid responses remained. Participants were randomly assigned to one of 12 experimental conditions defined by two factors: AI risk type (visible vs. invisible) and information source (AI, scientists, governments, AI companies, traditional media, science key opinion leaders). Each condition presented a mock Weibo post describing an AI risk, which participants viewed for 40 seconds under a forced‑viewing procedure. All measurements used 7‑point Likert scales except trust in science (0–100 single item). Key variables included pre‑test and post‑test trust in AI, pre‑test and post‑test trust in science, perceived information credibility, fear of AI, AI usage frequency, social media use, perceived relevance, and demographics. The questionnaire was administered online via the Wenjuanxing platform. All analyses were performed using SPSS version 26.0 and the PROCESS macro version 4.1 for mediation and moderation models.

Institutions

• Jinan University

  Guangdong, Guangzhou
• Tsinghua University

  Beijing, Beijing

Categories

Licence