"Excuse me, may I say something..." CoLabScience, A Proactive AI Assistant for Biomedical Discovery and LLM-Expert Collaborations
New 'PULI' framework lets AI assistants proactively join scientific conversations with 90% precision.
A team of researchers led by Yang Wu has published a paper introducing CoLabScience, a novel AI assistant designed to transform how scientists collaborate with large language models (LLMs). Unlike standard chatbots that wait for user prompts, CoLabScience is built to be proactive, using a new framework called PULI (Positive-Unlabeled Learning-to-Intervene) to determine the optimal moments to offer insights during live scientific discussions. This system analyzes a team's project proposal and maintains both long- and short-term conversational memory to make context-aware interventions, aiming to accelerate biomedical discovery by making AI a more engaged partner.
To train and evaluate this proactive capability, the team also created the Biomedical Streaming Dialogue Dataset (BSDD), a benchmark built from simulated research discussions derived from PubMed articles. Experimental results show that the PULI framework significantly outperforms existing reactive baselines in both intervention precision and overall collaborative task utility. This work, submitted to ACL 2026, represents a shift from passive AI tools to active collaborators, potentially unlocking new workflows where AI systems can anticipate needs and contribute ideas autonomously within expert teams.
- Introduces CoLabScience, an AI assistant that proactively intervenes in scientific dialogues instead of waiting for prompts.
- Core PULI framework uses reinforcement learning trained on the new BSDD benchmark, outperforming reactive models.
- Designed to enhance human-AI collaboration in biomedical research by providing timely, context-aware suggestions.
Why It Matters
Transforms AI from a reactive tool into an active research partner, potentially accelerating scientific discovery cycles.