Actions, Speech, and Looks: What Shapes How We Feel About In-Vehicle AI Assistants?

Paper Title

Actions, Speech, and Looks: What Shapes How We Feel About In-Vehicle AI Assistants?

Publication Info

• Topic area: Human-centered design of intelligent in-vehicle assistants (IVAs).
• Keywords: In-vehicle assistants, autonomy, embodiment, conversational style, human-computer interaction, trust, user experience, anthropomorphism, proactive systems, automotive AI.

Background and Problem

• Problem / challenge: Current IVA designs often emphasize high autonomy and anthropomorphic features, but user preferences regarding these design choices remain unclear, especially in physically invasive contexts.
• Significance: Understanding user preferences for IVA design can improve trust, comfort, and acceptance, which are critical for safe and effective human-vehicle interaction.
• Motivation and related work: Previous studies have explored IVA functionality, embodiment, and conversational styles, but gaps remain in understanding how autonomy, anthropomorphism, and speech style interact to shape user perceptions. This study builds on prior findings to address these gaps using a large-scale experimental design.

Solution

• Proposed approach: A video-based online experiment systematically varying IVA autonomy, embodiment, and conversational style across two scenarios (temperature adjustment and seat adjustment).
• Novelty:
  1. Empirical evaluation of user preferences for IVA autonomy levels, including autonomous actions with and without explanations.
  2. Comparison of virtual and physical IVA embodiments, ranging from abstract to humanlike designs.
  3. Investigation of formal vs. informal conversational speech styles in IVAs.
  4. Large-scale study (n = 1238) with robust statistical analysis of user perceptions and preferences.
• Procedure and key techniques:
  • Experimental design: 2 (scenarios) × 4 (autonomy levels) × 4 (embodiments) × 2 (speech styles).
  • Participants viewed videos simulating IVA interactions and rated warmth, competence, discomfort, social presence, trust, and user experience using validated scales.
  • Statistical analysis included General Linear Models (GLMs) and linear mixed-effects models to assess main effects and interactions.

Results

• Concrete findings:
  • Autonomy: Systems acting autonomously without explanation were rated lowest across trust, competence, and user experience. User-initiated and system-initiated dialogue conditions were preferred.
  • Embodiment: Abstract virtual agents were most favored, while humanoid robots elicited high discomfort and low trust.
  • Conversational style: Informal speech was consistently rated higher in warmth, competence, and social presence compared to formal speech.
• Advantage over baselines:
  • System-initiated dialogue balanced proactivity and transparency, outperforming autonomous actions without explanation.
  • Abstract virtual agents avoided spatial and attentional challenges posed by physical embodiments.
  • Informal speech enhanced user perceptions without compromising functionality.
• Experiments / evaluation:
  • Sample size: n = 1238 participants.
  • Measures: Warmth, competence, discomfort, social presence, trust, user experience, usage intentions.
  • Scenarios: Temperature adjustment and seat adjustment.
  • Statistical robustness: GLMs with heteroscedasticity-robust standard errors; linear mixed-effects models for repeated measures.
• Limitations and future work:
  • Lack of real-world immersion; findings based on video simulations.
  • Cultural bias due to German participant sample.
  • Limited exploration of physical invasiveness and alternative humanoid robot designs.
  • Future research should investigate dynamic adaptation of IVA speech styles and explore user thresholds for physical invasiveness.

Summary

This study provides empirical insights into how IVA autonomy, embodiment, and conversational style influence user perceptions, trust, and acceptance. Users preferred system-initiated dialogue over fully autonomous actions, abstract virtual agents over physical robots, and informal conversational speech over formal styles. These findings challenge assumptions favoring high autonomy and anthropomorphism in IVA design, suggesting that transparency, minimal embodiment, and conversational tone are key to enhancing user experience. Future IVAs should prioritize proactive engagement with clear communication, avoid intrusive physical designs, and adapt speech styles to context.