Effect of AI Performance, Risk Perception, and Trust on Human Dependence in Deepfake Detection AI system

AI performance directly shapes how much people rely on it: The study of 400 participants found that people adjust their trust in deepfake detection AI based on how well it actually performs, suggesting that transparent communication about an AI system's accuracy is critical for responsible deployment.

People's perception of risk matters as much as the AI's actual performance: Participants calibrated their dependence on AI not only based on detection results, but also on how risky they perceived the situation to be, indicating that user education about real threats is essential alongside AI tool development.

Current deepfake detection tools create confusion rather than clarity: The research highlights that limitations in existing AI detection tools lead to user mistrust and difficulty distinguishing real from fake content, underscoring the need for more reliable systems before widespread public deployment.

Explainability and transparency are essential for responsible AI: The authors emphasize that deepfake detection systems must be designed to be "transparent and explainable" to support everyday users, rather than operating as black boxes that users cannot understand or verify.

There's a gap between AI capability and public readiness: The findings suggest that while synthetic media detection AI exists, it's not yet reliable or understandable enough for the general public to use safely, raising questions about whether such tools should be deployed at scale without further development.