Research Artifacts

Poster: Interaction-Time Arbitration in Multi-Tool AI Tasks: Quantifying Delegation Instability Under Concurrent AI Availability

UW iSchool Research Fair · March 2026

As AI systems become embedded across everyday platforms, users increasingly work with multiple AI tools within the same task. Prior research has examined trust calibration, automation levels, and output evaluation, but has paid less attention to how users reroute execution across concurrently available systems while pursuing a stable objective.

This project formalizes interaction-time arbitration as a measurable construct capturing within-task instability in execution commitment across concurrently available AI systems, holding task goals constant. I define arbitration episodes as routing transitions between systems during goal-stable drafting tasks and propose behavioral metrics including episode duration, switch frequency, and reopening rate. A planned within-subjects study compares single-system and multi-system conditions to test whether arbitration cost predicts subjective workload, perceived control, and task performance under matched task goals and comparable interface conditions.

By introducing a behavioral framework for quantifying delegation instability, this work provides an empirical foundation for studying how concurrent AI availability reshapes control, authorship, and agency in multi-tool environments.