Since the start of the pandemic, Covid-19 has been the subject of thousands of scientific publications … per week. How, when you are a researcher, find your way without drowning in this flood of expert literature, the volume of which has exploded in recent months? By letting artificial intelligence (AI) read them for you. It has been possible for a few months at the Institut Pasteur, thanks to a technology called Deep Search (“Advanced research”, in French).
→ READ. Artificial intelligence, between promises and fears
Thanks to the computing power of its AI (developed by IBM), the program is able to read and search scientific databases, in less time than “Would not put any human brain”, notes Xavier Vasques, director of IBM’s technology centers. This program is a bit like the intern who writes the reading sheets for the presenter of a literary program. With one more advantage. “Deep Search does not save reading time so much as it aggregates and links data, and extracts the information it deems most relevant,” underlines Professor Michael Nilges, director of technology and head of the structural bioinformatics unit at the Institut Pasteur.
In this case, which drugs were used, for what result? Have any co-morbidities been detected? More than a “super” search engine, advanced research is capable “To deduce a certain number of hypotheses, adds Michael Nilges. As well as issuing a critical analysis of aggregated data ”.
A tool driven by researchers
And the man in all of that ? He remains master on board, specify the teams. “The tool does not invent a solution on its own. If he reads and ties things together, it is because a researcher, behind, has trained the tool “, explains Xavier Vasques, at IBM, himself a former researcher in mathematics. The machine is fed by “The right keywords to detect, the right databases to search, all configured by Pasteur researchers, engineers, biomathematicians… ”, confirms Michael Nilges.
In the laboratory, the teams “Will then carry out verification and experimentation work to confirm or refute the hypotheses” thus issued. In the long term, the goal is to regularly update the AI, in order to increase performance and improve it for generations of researchers to come.