Objective:

To compare consensus outcomes between manual revisions and AI-assisted revisions of subthreshold statements using retrieval-augmented generation (RAG) and multi-agent modules.

Approach:

• Delphi Method: Utilized a structured technique for eliciting expert judgment and achieving consensus in the context of RNA-based therapeutics.
• AI-Assisted Pipeline: Implemented an AI-assisted Delphi pipeline to mirror human reviewer workflows in the post-round 1 phase.

Key Findings:

• AI assistance is expected to rescue more statements by providing comprehensive evidence summaries and citation-anchored revisions.
• Final consensus rates are anticipated to be similar to those achieved by human reviewers.

Interpretation:

Explorations of AI-facilitated Delphi exercises indicate potential for process acceleration while underscoring the importance of human oversight.

Limitations:

• Delphi rounds are time-consuming, often taking 2 to 6 months.
• Panelist engagement tends to decline with each round, complicating the formation of consensus.

Conclusion:

The study aims to evaluate the effectiveness of AI in enhancing the Delphi process for RNA-based therapeutics.

Sources:

• Delphi Method Review
• AI Delphi Exploration