Reproduce “AI-Supervisor: Autonomous AI Research Supervision via a Persistent Research World Model” 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

Yunbo Long

Yunbo Long developed AI-Supervisor, an autonomous AI research supervision system that uses a persistent, uncertainty-annotated Research World Model and a multi-agent consensus protocol to identify empirically grounded research gaps and develop methods. The system achieved a 4.44 alignment score with ground-truth contributions for gap discovery and produced methods with 32% higher novelty through cross-domain search.

Key insights

  • A persistent Research World Model, structured as a knowledge graph, enables effective knowledge accumulation, transfer, and structured gap reasoning across multiple research projects.
  • Multi-agent consensus, particularly with shared visibility among agents, significantly enhances the precision and reliability of identifying empirically verifiable research gaps.
  • Combining root-cause analysis (5-WHY) with cross-domain technique search within a quality-gated development loop leads to the creation of more novel and effective research methods.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.