The New Lab Assistant: Utilizing Multi-Agent AI Architectures in Medical Research

Abstract

The article concerns navigating complex literature in specialized fields. The aim of this study is to investigate the use of Multi-Agent Systems (MAS) to optimize literature review, data extraction, and statistical validation in medical research, as well as to support in silico approaches. Researchers employ a "triangulation" approach, orchestrating specific models for their unique strengths. The comparison included ChatGPT, Claude, Perplexity, and NotebookLM. The results showed that ChatGPT and Claude were stronger in synthesis and structured interpretation, while Perplexity and NotebookLM were especially useful for source-based retrieval and transparent evidence tracking. Broad synthesizers such as ChatGPT are effective for rapid data mapping and identifying cross-disciplinary trends, while deep readers such as Claude are well suited for critical audits of multiple full-text papers simultaneously, identifying methodological nuances that metadata searches often miss. Evidence-oriented tools such as Perplexity and NotebookLM strengthen this process by grounding responses in retrieved or uploaded sources, helping filter out clinically irrelevant data or animal-only trials. Such triangulation improves both the speed and the reliability of literature-based medical research.