CoRead: Using Implicit Behavioural Signals for Mixed-Initiative AI Support in Collaborative Academic Reading

Abstract

Collaborative academic reading depends on coordination that existing tools do not provide. Readers need to know when peers are confused, who can help, and when AI assistance would add value rather than interrupt. Two interaction problems remain unresolved: when AI should intervene (grounded in sustained behavioural signals rather than isolated events or explicit requests) and how assistance should be orchestrated across individuals, peers, and the group. We present CoRead, a mixed-initiative collaborative reading interface addressing both. A formative study (N = 12) grounded five design requirements informing a coordination architecture that routes help across pre-reading alignment, in-session support, and post-session synthesis, abstaining when intervention would disrupt more than it helps. A Wizard-of-Oz feasibility evaluation (N = 12) showed 81.1% acceptance of implicit AI interventions and 53.8% acceptance of peer-routing offers, SUS = 73.3 (95% CI [69.7, 76.9]), NASA-TLX = 47.0, and automated cosine-similarity grounding checks passing 19 of 21 AI-generated explanations.