Algorithm Transparency and User Trust: A Review of Transparency Design in Data Science Systems

Abstract

This review investigates how algorithm transparency influences user responses to data science systems. It distinguishes information disclosure from users’ perception that a system is transparent. It also compares explanations, performance information, uncertainty communication, and user control. The review covers twelve peer-reviewed studies published between 2004 and 2023. Relevant and task-appropriate transparency can improve perceived understanding and self-reported trust. Clear explanations may also support acceptance when users can connect the information to a decision. However, these positive attitudes do not necessarily improve behavioral reliance or task performance. Detailed information may cause overload and make a system harder to understand. Persuasive explanations may also increase acceptance of incorrect advice. The effects depend on system accuracy, task type, user experience, and opportunities to influence or challenge the output. Transparency should therefore be evaluated through separate attitudinal, behavioral, and performance measures. Its main goal should be calibrated trust rather than maximum trust.