Swarm UIs: Impact of Assistance on Users’ Sense of Agency

Paper Title

Swarm UIs: Impact of Assistance on Users’ Sense of Agency

Publication Info

• Topic area: Human-Computer Interaction (HCI), focusing on swarm user interfaces and their impact on users' sense of agency.
• Keywords: Swarm UIs, sense of agency, system assistance, autonomy, predictability, task difficulty, human-robot interaction, user responsibility, interface adoption, cooperative tasks.

Background and Problem

• Problem / challenge: Excessive assistance in swarm UIs can reduce users' sense of agency (SoA), leading to diminished responsibility and decreased adoption of the interface.
• Significance: Maintaining users' sense of agency is critical for high-stakes applications like surgery or search-and-rescue coordination, where users must retain control to make informed decisions.
• Motivation and related work: Previous studies explored assistance levels in teleoperated systems but lacked focus on direct interaction with swarm UIs and their impact on SoA. The paper addresses gaps in understanding how autonomy, task difficulty, and predictability influence SoA in swarm UIs.

Solution

• Proposed approach: The study investigates nine types of assistance in swarm UIs, defined by autonomy levels and proxy usage, and their impact on users' SoA.
• Novelty:
  1. First empirical evidence of decreasing SoA with increasing assistance in swarm UIs.
  2. Introduction of three novel types of SoA specific to swarm UIs: proxy module, satellite modules, and the swarm as a whole.
  3. Demonstration that task difficulty does not affect SoA in swarm UIs, while predictability impacts SoA differently across modules.
• Procedure and key techniques:
  • Conducted three experiments (N=105 participants) using drag-and-drop tasks with Toio robots.
  • Experiment 1: Studied nine assistance types combining autonomy and proxy factors.
  • Experiment 2: Examined the combined effect of task difficulty and assistance type on SoA.
  • Experiment 3: Investigated the combined effect of system predictability and assistance type on SoA.
  • Explicit SoA measurement via an 8-point Likert scale.

Results

• Concrete findings:
  • Higher autonomy levels reduce SoA significantly (e.g., Ainert autonomy scored 7, Acmplt autonomy scored 1.58).
  • Proxy usage decreases SoA compared to direct control of all modules.
  • Predictability impacts SoA, with noisy trajectories reducing SoA (Straight: M=4.79, Noisy: M=4.44).
• Advantage over baselines:
  • Interaction with modules restores SoA compared to fully autonomous systems.
  • Semi-autonomous levels (e.g., Asnap) provide better SoA than highly autonomous levels (e.g., Atrg).
• Experiments / evaluation:
  • Experiment 1: Significant effects of autonomy and proxy on SoA (F=74.5, p<.001).
  • Experiment 2: Task difficulty had inconclusive effects on SoA.
  • Experiment 3: Predictability significantly impacted SoA (F=10.35, p=.003).
• Limitations and future work:
  • Limited task complexity and experimental surface area.
  • Future work could explore implicit SoA measures, larger task areas, and real-world applications like search-and-rescue scenarios.

Summary

This paper provides empirical evidence that increasing assistance in swarm UIs decreases users' sense of agency, with autonomy levels exerting a stronger effect than proxy usage. It introduces three distinct types of SoA specific to swarm UIs: for the proxy module, satellite modules, and the swarm as a whole. While task difficulty showed no significant impact, system predictability influenced SoA, particularly for the proxy module. These findings offer actionable insights for designing swarm UIs that balance autonomy and user control, ensuring responsibility and adoption in critical applications.