Back to Research Timeline

Group Paper Reading · August 5, 2026

AI Transparency, Autonomy & Trust Calibration

A paper discussion examining whether greater AI transparency always improves human decision-making and trust.

AI TransparencyHuman–AI InteractionTrust Calibration

Overview

The paper contrasted computational accounts of transparency, where explanations help align trust with system reliability, with HCI evidence that excessive explanation may reduce autonomy or increase cognitive load.

It combined an agent-based simulation with a survey of organisational decision-makers and then incorporated survey-derived relationships back into the simulation.

Key Themes

  • Trust calibration as alignment between user trust and actual system reliability.
  • The relationship between explanation depth, cognitive load, and perceived autonomy.
  • Individual differences in how users respond to transparent AI systems.
  • Adaptive transparency rather than a single explanation policy for every user.

Main Takeaways

  • More transparency is not automatically better for every person or decision context.
  • Explanation design should consider autonomy and cognitive burden alongside accuracy and trust.
  • Dynamic or preference-sensitive explanations may offer a better design target than blanket transparency requirements.

Discussion & Limitations

  • The short personality instrument provides only a limited representation of complex individual differences.
  • The simulated agents did not interact with one another, limiting what the model can say about social dynamics.
  • Reintroducing survey findings into the simulation does not provide fully independent validation of those findings.