Interaction Models: A Scalable Approach to Human-AI Collaboration

Summary

A research preview introducing interaction models that natively handle real-time communication across audio, video, and text rather than relying on external scaffolding. The authors argue that current AI systems create a "collaboration bottleneck" by forcing users into turn-based interfaces, when humans work most effectively through continuous, simultaneous exchange.

The approach uses time-aligned micro-turns (200ms chunks) to enable seamless interruptions, visual interjections, simultaneous speech, and concurrent tool use — achieving strong performance on both intelligence and interactivity benchmarks.

Why it caught my eye

(notes to add)