This study used statistical and machine learning analysis to explore public trust in AI’s cognitive abilities across a variety of domains. The results enable a comprehensive understanding of how age, gender, and AI familiarity influence trust in AI systems.
Impact of demographics on trust
Our results showed that AI proficiency was the strongest predictor of confidence, and this finding was confirmed by MANOVA and random forest classifier. Participants who reported higher levels of familiarity with AI consistently expressed greater trust in all types of tasks, especially memory recall and logistics tasks, where AI is perceived to be better.
Age also emerged as an important factor. Young participants showed high trust in the AI’s decision-making abilities, especially in simple memory-based tasks. This trend is consistent with previous research suggesting that younger people are more comfortable with digital technology and automation.
Although the effect of gender was smaller than the effects of familiarity and age, it was statistically detectable in multivariate tests. Previous research has linked gender differences to perceived risk, domain familiarity, and socialization surrounding automation. Future research should investigate whether design affordances (e.g., error transparency, controllability) reduce these gaps and whether the effects persist after establishing measurement invariance across gender.
Trust by scenario type
Participants tended to trust humans in high-stakes situations, such as medical diagnoses and self-driving decision-making, reflecting existing literature on aversion to algorithms and the need for human empathy, accountability, and recognition of nuance in critical scenarios.
In contrast, trust in AI exceeded trust in humans in areas that require data recall and consistency, such as historical knowledge and memory retrieval. These results support the hypothesis that people will trust AI more in tasks that are perceived to be objective, factual, and reproducible.
Predictive modeling insights
The Random Forest model demonstrated high accuracy in classifying confidence levels, identifying AI familiarity, frequency of use, and age as the top predictors. This supports the idea that trust in AI is formed not only by static characteristics (such as demographics) but also by exposure and behavioral engagement.
what it means
First, the onboarding build friendliness (guided tours, hands-on trials with feedback) may increase appropriate confidence, especially among older and less experienced users. Second, in high-stakes situations, human-involved workflows and explicit error bounds allow you to balance trust and risk. Third, task-aware explainability tailored to memory and complex judgment tasks should emphasize the clarification of provenance and uncertainty. Finally, implementers must monitor adjustments (both user and model) and publish post-deployment performance dashboards to maintain guaranteed trust.
Restrictions
Cross-sectional convenience samples limit causal inference and generalizability. Measurement invariance (e.g., multigroup CFA or DIF) was not tested. Therefore, some of the differences between groups may partially reflect scale inequalities rather than true reliability differences. Because our predictive modeling uses exploratory capabilities, it may not capture context-dependent changes in trust under different error costs and observed AI performance. Although we used nested cross-validation and calibration analysis, external validation for new populations and roles (clinicians, drivers, regulators, etc.) is still needed.
We do not incorporate real-world AI performance feedback or error cost scenarios. Future trust adjustment studies should model overtrust and undertrust in relation to system accuracy and risk.
