Reproduce “Chronology of Multi-Agent Interactions for Provenance of Evolving Information” 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

Ching-Chun Chang, Isao Echizen

A system for tracking the chronological provenance of evolving information in multi-agent generative AI is presented, embedding symbolic chronicles directly into generated text. It achieved near-perfect accuracy (99-100%) in reconstructing agent interaction sequences at sufficient bias strengths, offering a self-contained method for post hoc attribution even under significant lexical substitution.

Key insights

  • Provenance information can be intrinsically linked to evolving content through lexical biasing, offering a self-contained 'chain of custody' without relying on fragile external metadata.
  • The strength of the lexical bias (δ) provides a critical trade-off between the accuracy of chronological attribution and the quality (perplexity) of the generated text.
  • Reconstructing the temporal order of multi-agent contributions is achievable from the generated content itself, distinguishing it from traditional watermarking limited to single-source attribution.

Summary by alphaXiv.

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