AI system accelerates scientific evidence review

Researchers at Queen Mary University of London have developed an AI-powered framework designed to help scientists rapidly analyse large volumes of academic literature while maintaining transparency and human oversight.

The system, MetaBeeAI, combines large language models with expert validation tools to support systematic reviews - a core method used to synthesise scientific evidence. These reviews are often considered the gold standard in research, but can take months or years to complete due to the rapidly increasing volume of published studies.

As a demonstration, the team applied the system to nearly 1,000 research papers examining the effects of pesticides on bees, a key area of concern for biodiversity and ecosystem health. The platform was able to extract, organise and analyse key information from across the dataset, with researchers focusing on verifying and refining the outputs.

 Unlike fully automated approaches, MetaBeeAI is designed to keep scientists directly involved in the review process. The system highlights relevant text within original papers and allows users to trace each output back to its source, enabling correction and validation throughout.

Dr Rachel Parkinson, who led the research [1], said the aim is to reduce repetitive workloads rather than replace scientific judgement.

“The system is not designed to replace scientists. Instead, MetaBeeAI supports researchers by reducing repetitive workloads and helping them focus on interpretation, decision-making and scientific insight.”

The study found the system performed well in identifying factual information such as species names and chemical compounds, while iterative human feedback further improved accuracy and consistency.

The researchers say the framework could be adapted for use across a range of disciplines, including medicine, climate science and public health, where evidence synthesis is increasingly constrained by the scale of available data.

The approach also highlights the importance of transparency in AI-assisted research, particularly in addressing concerns around accuracy, reproducibility and automated errors in generative systems.

By combining machine-scale processing with expert oversight, the team suggests that tools like MetaBeeAI could help accelerate evidence synthesis and support faster responses to complex global challenges.

More information online

1. MetaBeeAI: An AI pipeline for structured evidence extraction from biological literature, published in ScienceDirect. DOI: https://doi.org/10.1016/j.ecoinf.2026.103813