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

TL;DR

Surprisingly, evidence shows that the embedded visualization led to greater inappropriate reliance on AI, primarily as over-reliance, due to factors like perceptual challenges, visual proximity illusions, and highly realistic visual representations, Nonetheless, the embedded visualization demonstrated benefits in spatial mapping.

Abstract

Artificial Intelligence (AI) and indoor sensing increasingly support decision-making in spatial environments. However, traditional visualization methods impose a substantial mental workload when viewers translate this digital information into real-world spaces, leading to inappropriate reliance on AI. Embedded visualizations in Augmented Reality (AR), by integrating information into physical environments, may reduce this workload and foster more appropriate reliance on AI. To assess this, we conducted an empirical study (N = 32) comparing an AR embedded visualization (X-ray) and 2D Minimap in AI-assisted, time-critical spatial target selection tasks. Surprisingly, evidence shows that the embedded visualization led to greater inappropriate reliance on AI, primarily as over-reliance, due to factors like perceptual challenges, visual proximity illusions, and highly realistic visual representations. Nonetheless, the embedded visualization demonstrated benefits in spatial mapping. We conclude by discussing empirical insights, design implications, and directions for future research on human-AI collaborative decision in AR.

Figures (9)

• Figure 1: An example spatial arrangement of the coffee machine targets locations for a single participant location, illustrating four distinct difficulty levels: same+close, same+far, cross+close, and cross+far. For each participant location, there are 4 trials, one difficulty level per trial. The participant's location is marked in red, and colored dots indicate target locations in each difficulty level. The white dashed lines indicate that the coffee machine targets are placed at equal line-of-sight distances from the participant.
• Figure 2: The 8 participant locations in the study, distributed across two floors of the building. Participants were instructed to complete 4 trials at each location, following the order indicated by the numbered labels and route. The letter "L" is an abbreviation for location.
• Figure 3: A stimulus workflow from an actual trial. Each trial began with a 20-second countdown. Participants viewed either the X-ray or Minimap visualization while searching for the optimal target. After identifying their choice, participants verbally announced the selected target and rated their confidence on a 1–7 scale. Finally, they physically pointed to the perceived location of the chosen target. For illustration, the figure shows only one trial at location L, but participants completed four trials at each location.
• Figure 4: One example of what a participant experienced during the experiment. Between participants, the order of the 4 conditions (Minimap+NoAI, Minimap+AI, X-ray+NoAI, X-ray+AI) was varied and counterbalanced using a Latin square design. Factors that remained fixed across all participants included the sequence of standing locations (L1-L8) and the order of the four spatial arrangements (Same+Close, Same+Far, Cross+Close, Cross+Far) at each standing location. Within each AI condition, the presentation order of the AI suggestion optimality levels (opt, sub, worst) was randomized, while ensuring a balanced distribution across participants. For illustration, only two examples of AI suggestion presentation orders are shown.
• Figure 5: Classification of reliance as appropriate, over-, and under-reliance based on the alignment between AI suggestions and human decisions. Appropriate reliance on AI was defined as participants accepting optimal suggestions or overriding suboptimal and worst suggestions. Inappropriate reliance included: (1) over-reliance: accepting suboptimal or worst suggestions; and (2) under-reliance: rejecting AI suggestions in favor of a worse option.