Designing for Responsible Trust in AI Systems: A Communication Perspective

Trust in AI requires intentional communication design, not just trustworthy principles. The paper argues that current focus on what makes AI trustworthy misses the critical question of how people actually develop trust. This means AI developers must actively think about what signals they send to users, rather than assuming trustworthiness will be self-evident.

Transparency and user interaction are powerful but underutilized trust signals. Beyond just explaining AI decisions, the paper identifies that allowing users to interact with AI systems and observe how they work creates important opportunities to build appropriate trust—suggesting that one-way explanations alone are insufficient.

Different users process trust differently, creating equity challenges. The MATCH model highlights that people have varying cognitive processes and limitations when judging trustworthiness. This means a single communication approach may work for some users while misleading or confusing others, requiring designers to consider their target audience carefully.

The industry needs guardrails against deceptive trust-building tactics. The paper emphasizes that without regulation, companies could exploit how people form trust judgments through manipulative design choices. The authors call for industry-wide standards to prevent malicious use of trustworthiness cues.

Not all trust signals are equally reliable or appropriate. The paper introduces concepts like "warranted trustworthiness cues" (signals that genuinely reflect actual trustworthiness) versus potentially misleading ones, and provides a checklist to help creators choose honest signals suited to their specific users.