Artificial intelligence (AI) tools are helping scientists with not only organizing and managing data, but  validating and correlating scientific hypotheses. While most conventional tools serve mainly as citation indexes, AI-based tools offer more in depth information and analysis. For example, AI-based “speed-readers” are useful due to the vast number of papers published, with some estimates that 1 million new papers are published globally each year, or one paper every 30 seconds. These speed-readers utilize algorithms that extract content from papers as well as filtering, ranking and grouping search results. In order to provide more advanced capabilities, algorithms can be customized to create knowledge graphs, which illustrate the relationships between extracted data points, such as an algorithm indicating whether a drug and protein are related if they are mentioned in the same sentence.  Organizations developing these types of AI-based tools also plan to provide supplemental information by identifying the hypotheses in the research papers and checking each paper against other relevant scientific documents in order to validate the hypotheses.

AI-based tools are also helpful in niche applications. Simple to use, many AI-based tools have user interfaces that are similar to popular internet search engines, such as Google; however, the AI-based tools provide much more information, such as popularity metrics, datasets, methods and indirect citations, which is when a method or notion is so well-established and commonplace that its origin is not cited by researchers. For all their benefits, AI-based tools require human involvement, as each hypothesis the tool generates must still be tested. Moreover, most AI-based tools are extremely costly and are usually limited in the scope of scientific literature they can search.

Source: Nature