A summariser that can write like a human, is no longer the realm of fiction
“Sliced, diced and digested: AI-generated science ready in minutes” reads the headline of the latest coverage, in Nature Index, of the way Artificial Intelligence (AI) is changing the way scholarly publishers make their content available to scientists.
Algorithms are now available to help us digest complex information from scientific and clinical research at scale. These solutions may alter the way we access this information and how we use and benefit from it.
These developments coincide with an era in which time is an ever more scarce commodity; busy schedules leave us with little time to invest in digesting complex information. Most professionals now consume bite-size chunks of information and only occasionally do a deep dive into some of the original studies from which these nuggets of information are extracted. However, these studies—evidenced-based content published by scholarly publishers — remain among the most trusted sources of information.
Extraction algorithms, used in summarisation, are therefore more and more sought after. The goal is to take complex information from a long, detailed study, for example, and summarise this into the type of bite-size content audiences need. Since extraction algorithms belong to the field of natural language processing—and not machine learning—they do not need a large corpus of source material to be trained on. This context, therefore, heralds the advent of AI-assisted content consumption.
The latest summarisation solutions, like the AI Summaries developed by SciencePOD in collaboration with the LIA lab at Avignon Université in France, are sufficiently advanced to give a good understanding of what each study is about, its key findings, and its implications for future research. The solution delivers texts written in full sentences and gives a sense of the core meaning of the original study within 300 words. Summaries like these differ from abstracts, which are based on the subjective interpretation of an author regarding what their paper is about. AI summaries, are a more faithful reflection of the most meaningful words in the original study.
Ultimately, the goal of AI summaries is to make research findings available to a wider audience, outside the field of study. The automated extraction of contextual information, such as the main key words, spelt-out acronyms and the inclusion of a lay definition of technical terms, helps any non-expert understand each study. The appetite for such solutions among publishers is ripe. And ultimately, they will pass on these solutions to the authors of research and clinical studies to help them communicate their work and achievements more widely. This solution will stimulate cross-fertilisation between research disciplines.
This does not mean that science journalists will suddenly become redundant. The ability of algorithms to perform storytelling is not fully refined yet. Current state-of-the-art summarisation solutions still require the eye and touch of a professional science and medical editor to polish the text, interview experts and put the story of the latest progress into its wider context. Yet, summarisation is advanced enough to deliver insights into the key findings of a research study or an academic book chapter within seconds.
AI summarisation is an essential step in the digital transformation of science and medical publishers, making research accessible and understandable at an unprecedented scale.