Multi-AI Discovery Protocol Pilot
Research question
Why it matters
Hypotheses
A structured multi-AI workflow can produce research outputs that are more auditable and less vulnerable to hallucination, confirmation bias and false consensus than an undocumented single-model workflow.
A structured multi-AI workflow does not materially improve research auditability or reduce hallucination, confirmation bias or false consensus compared with an undocumented single-model workflow.
Claim ledger
| Claim | Status | Confidence |
|---|---|---|
| C-001 AI agreement alone does not constitute independent verification. | verified | high |
Evidence
NIST identifies generative-AI confabulation as a risk, supporting the need to verify AI-generated claims against external evidence rather than treating model agreement as proof.
Source: Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
Predictions & tests
Predictions
When the same research question is investigated using MADP and an undocumented single-model workflow, the MADP process will produce a more complete audit trail and identify more unsupported or contradictory claims before publication.
Tests
None recorded.
AI use and review
AI research roles
No AI-use records published yet.
Red-team / replication / external review
No review records published yet.
Version history
| Version | Change | IDL level | Published |
|---|---|---|---|
| v0.1 | Initial pilot investigation created with H0, H1, first claim, verified source, linked evidence and preregistered prediction. | IDL-0 |