Reproduce “SciForge: An AI-Native, Multimodal Workbench for Scientific Discovery” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
SciForge Team, Zhangyang Gao, Minghao Fang, Yifei Liu, Hanhui Yang, Xinyu Gu, Shixiang Tang, Siqi Sun and 5 more
SciForge provides a research workbench that keeps agent actions, scientific objects, claims, evidence, provenance, and human review connected in an inspectable record. Its Evidence and Project DAGs organize reasoning and review. In eight prototype use cases, a BGC workflow prioritized 23 of 430 regions for follow-up; an EGFR docking result missed its preregistered threshold, and neither case establishes scientific validity.
Key insights
- Establishing a 'fail-closed' mechanism for unsupported scientific modalities prevents general-purpose AI models from misinterpreting data they cannot natively handle, ensuring domain-appropriate processing.
- The integration of asynchronous, interconnected Evidence and Project DAGs provides comprehensive, real-time provenance tracking for every agent action and human decision, enhancing scientific rigor and reproducibility.
- Human judgment is strategically incorporated via a thin GUI and explicit review gates, allowing for effective human-AI collaboration and governance in complex scientific workflows.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.