AutoSynthesis: An agentic system for automated meta-analysis
AutoSynthesis is an end-to-end multi-agent system for automated meta-analysis. It formulates search strategy, retrieves literature, screens studies, extracts statistics, computes effect sizes, and performs random-effects meta-analysis. It screened over 28 studies and extracted more than 20 quantitative claims. Pooled effect estimates are similar to Hedges' g of expert-conducted meta-analyses.
Development
- First ReportAutoSynthesis: An agentic system for automated meta-analysisarXiv cs.AI
- Current AssessmentThis work signals a shift toward AI-driven systematic review automation, which could reduce the time and cost of evidence synthesis in medicine, education, and policy. It also highlights the growing role of agentic systems in scientific research.AIGC.NEWS · analysis
Meta introduces AutoSynthesis, a multi-agent system that automates the entire meta-analysis pipeline, from literature search to effect size computation, achieving results comparable to human experts.
AutoSynthesis demonstrates that multi-agent orchestration can replicate complex, multi-step scientific workflows with high fidelity. The system's ability to produce PRISMA-compliant reports suggests a path toward automating evidence synthesis at scale.
This work signals a shift toward AI-driven systematic review automation, which could reduce the time and cost of evidence synthesis in medicine, education, and policy. It also highlights the growing role of agentic systems in scientific research.
For Meta, AutoSynthesis showcases its AI research capabilities and could be productized as a service for academic publishers, pharmaceutical companies, or government agencies needing rapid evidence synthesis.
Next signal: validation on larger, more diverse meta-analysis datasets and integration with live literature databases. If successful, AutoSynthesis could become a standard tool for researchers and policymakers.