Leveraging artificial intelligence to summarize abstracts in lay language for increasing research accessibility and transparency

Summary

This study demonstrates the feasibility and effectiveness of using ChatGPT-4 to automatically generate accessible lay summaries of scientific abstracts at scale on a national clinical recruitment platform. Researchers and volunteers evaluated 192 summary sentences from 33 abstracts, finding high levels of accuracy and relevance, with volunteers perceiving the AI-generated summaries as significantly more accessible and transparent than original abstracts. The implementation successfully expanded the ResearchMatch platform to display lay summaries for over 750 published studies, providing evidence for a scalable framework to enhance research transparency and public engagement.

UK applicability

The methodology and platform approach could be adapted to UK research infrastructure, though would require integration with UK-based research recruitment platforms and evaluation of ChatGPT's suitability for communicating findings to UK audiences. UK research ethics and transparency governance increasingly emphasises public engagement and accessible research communication, making this scalable approach potentially relevant to UK research institutions and funders.

Key measures

Accuracy (95.9%, 95% CI 92.1–97.9), relevance (96.2%, 95% CI 92.4–98.1), volunteer perception of accessibility (85.3%, 95% CI 69.9–93.6), volunteer perception of transparency (73.5%, 95% CI 56.9–85.4), harmfulness assessment (0 summaries deemed harmful), and number of studies expanded to (750+)

Outcomes reported

The study evaluated ChatGPT-4's ability to generate lay-language summaries of scientific abstracts and measured their accuracy, relevance, accessibility, transparency, and potential harmfulness as assessed by researchers and volunteers. The research also assessed the feasibility of scaling this approach across a national research platform.

Topic tags