New Scientist looks at the AI that’s been assigned the unenviable task of reading every scientific paper online and finding the important ones:
Semantic Scholar, which launches today from the Seattle-based Allen Institute for Artificial Intelligence (AI2), can automatically read, digest and categorise findings from the estimated 2 million scientific papers published each year. Up to half of these papers are never read by more than three people. The system aims to identify previously overlooked connections and information.
“Our vision is of a scientist’s apprentice, giving researchers a very powerful way to analyse what’s going on in their field,” says Oren Etzioni, director of AI2. “If you’re a medical researcher, you could ask ‘what’s the latest on these drug interactions?’ Or even a query in natural language like, ‘what are papers saying about middle-aged women with diabetes and this particular drug?’”
The system works by crawling the web for publicly available scientific papers, then scanning the text and images within them. By identifying citations and references in the text, Semantic Scholar can work out which are the most influential or controversial papers. It also highlights key phrases found in similar papers, extracting and indexing the datasets and methods each researcher used.
“With millions of papers coming out each year, there are no Renaissance men or women anymore,” says Etzioni. “People’s eyes glaze over and they miss that key paper or technique that they could use, in a medical case, to save somebody’s life.”