New AI assistant CoLabScience learns to proactively intervene in scientific discussions, moving beyond reactive chatbots to enable true human-AI collaboration in biomedical research.

CoLabScience introduces PULI (Positive-Unlabeled Learning-to-Intervene), a reinforcement learning framework that determines optimal timing and approach for AI to contribute to ongoing scientific conversations.

The system leverages project proposals and both long and short-term conversational memory to provide timely, context-aware interventions rather than waiting passively for user prompts.

BSDD (Biomedical Streaming Dialogue Dataset), a new benchmark of simulated research discussions, was created to support development and evaluation of proactive AI collaboration in scientific settings.