Examining the Effect of Explanations of AI Privacy Redaction in AI-mediated Interactions

Abstract

AI-mediated communication is increasingly being utilized to help facilitate interactions; however, in privacy sensitive domains, an AI mediator has the additional challenge of considering how to preserve privacy. In these contexts, a mediator may redact or withhold information, raising questions about how users perceive these interventions and whether explanations of system behavior can improve trust. In this work, we investigate how explanations of redaction operations can affect user trust in AI-mediated communication. We devise a scenario where a validated system removes sensitive content from messages and generates explanations of varying detail to communicate its decisions to recipients. We then conduct a user study with $180$ participants that studies how user trust and preferences vary for cases with different amounts of redacted content and different levels of explanation detail. Our results show that participants believed our system was more effective at preserving privacy when explanations were provided ($p<0.05$, Cohen's $d \approx 0.3$). We also found that contextual factors had an impact; participants relied more on explanations and found them more helpful when the system performed extensive redactions ($p<0.05$, Cohen's $f \approx 0.2$). We also found that explanation preferences depended on individual differences as well, and factors such as age and baseline familiarity with AI affected user trust in our system. These findings highlight the importance and challenge of balancing transparency and privacy in AI-mediated communications and suggest that adaptive, context-aware explanations are essential for designing privacy-aware, trustworthy AI systems.

Figures (10)

• Figure 1: Conceptual overview of AI-mediated communication. The AI coordinator mediates between a researcher and collaborator by redacting sensitive information and providing an explanation to the recipient.
• Figure 2: Diagram of the interactions between the researcher, collaborator, and coordinator, with tasks performed by each that include and do not include private information indicated.
• Figure 3: Screenshot of the question, redacted answer, and explanation (if provided) that are included for each per-interaction evaluation group
• Figure 4: Responses of participants to Survey Section \ref{['subsec:survey2']} about their overall experiences with the system, normalized to $0-1$. The last set of bars illustrate the average across all questions. The value corresponding to the most positive response is shown in red. Significant differences are shown in red with one asterisk for $p<0.05$.
• Figure 5: Responses of participants to Survey Section \ref{['subsec:survey1']} about their overall experiences with the system, normalized to $0-1$. Responses are separated by the explanation received by participants and by the amount of redaction needed for the answer. The value corresponding to the most positive response is shown by a dotted line. Significant differences are shown in red with one, two, and three asterices for $p<0.05$, $p<0.01$, and $p<0.001$, respectively.