ScientistOne: Chain-of-Evidence for Verifiable AI Research
A recent publication on arXiv presents ScientistOne, an independent research system that employs Chain-of-Evidence (CoE) to guarantee that each assertion can be traced back to its origin. This system tackles issues of verifiability in AI-generated studies, including false citations and results that cannot be replicated. The CoE Audit, a verification tool used after the fact, conducts four integrity assessments: score verification, violation of specifications, reference checks, and alignment of methods with code. Among 75 studies covering five systems and five advanced research tasks, every baseline showed at least one systematic flaw. This initiative seeks to enhance the reliability of autonomous research agents.
Key facts
- ScientistOne uses Chain-of-Evidence (CoE) for verifiability.
- CoE requires every claim to be traceable to its evidence source.
- CoE Audit applies four integrity checks: score verification, specification violation, reference verification, and method-code alignment.
- 75 papers across five systems and five frontier research tasks were evaluated.
- Every baseline exhibited at least one systematic verifiability failure.
- The paper is published on arXiv with ID 2605.26340.
- The system covers literature review, solution discovery, and paper writing.
- Fabricated citations and unreproducible scores are common issues addressed.
Entities
Institutions
- arXiv