Skip to content
DIFAN JIA
Back to Publications

CHI 2026 · Accepted

Can AR Embedded Visualizations Foster Appropriate Reliance on AI in Spatial Decision-Making? A Comparative Study of AR X-Ray vs. 2D Minimap

Xianhao Carton Liu, Difan Jia, Tongyu Nie, Evan Suma Rosenberg, Victoria Interrante, Chen Zhu-Tian

PDFDOI
Can AR Embedded Visualizations Foster Appropriate Reliance on AI in Spatial Decision-Making? A Comparative Study of AR X-Ray vs. 2D Minimap

Abstract

This paper studies whether AR-embedded visualizations can improve how people rely on AI during spatial decision-making. We compare an embedded AR X-ray view against a traditional 2D minimap in AI-assisted, time-critical target-selection tasks. In a user study with 32 participants, the AR-embedded condition unexpectedly produced more inappropriate reliance (mainly over-reliance), while still showing advantages for spatial mapping.

Why This Question Matters

AI and indoor sensing systems can guide users in complex spaces, but people still need to judge when AI is reliable. If a visualization format increases blind trust, decision quality can drop even when the interface looks more "immersive."

Study Setup

  • Task type: AI-assisted spatial target selection under time pressure
  • Comparison: AR embedded X-ray vs. 2D minimap
  • Participants: 32
  • Goal: Measure whether visualization style affects appropriate vs. inappropriate reliance

Key Findings

  • Embedded AR views did not automatically produce better reliance calibration.
  • Over-reliance increased in the AR embedded condition.
  • Contributing factors included perceptual challenges and visual proximity illusions.
  • Despite that, embedded visualization helped with spatial mapping and environmental understanding.

Key Contributions

  • Empirical evidence on how visualization modality shapes reliance behavior in AR+AI workflows.
  • A direct comparison of embedded AR visualization and 2D minimap in the same decision context.
  • Design implications for building AR interfaces that support calibrated human-AI collaboration.

Design Implications

  • Do not assume higher realism leads to more appropriate trust.
  • Pair embedded visual cues with uncertainty communication and cross-check support.
  • Evaluate reliance outcomes directly (not only speed/accuracy) when designing AR decision aids.

Video / Demo

Demo video and supplemental materials will be added here.