Artificial Intelligence (AI), a technology that processes a lot of information from a variety of sources, can help document the spread of the COVID-19 disease and assist in making healthcare decisions. AI is primarily based on how well machines are trained from huge amounts of data so that it can recognize unusual patterns and make decisions accordingly.

“COVID-19 reveals limits of AI-driven tools,” Santosh said. “It is primarily due to the lack of sufficient data. AI tools are required to learn over time in parallel with experts. To collect large amounts of data, we should not wait for days, months and years for such a global threat.”

Santosh’s article addresses the challenges the lack of this amount of data presents when tracking a fast-spreading outbreak of a previously unknown infectious agent. With a new and rapidly developing outbreak, he says AI experts can train systems on the limited data initially available, while continuing to learn and check for anomalies as new data is added. This approach, which he calls "active learning," avoids waiting to collect large amounts of data before employing AI tools.

In addition, he says that systems that learn data from Wuhan, China, should be expected to be tested in Rome, Italy, which he calls a cross-population train/test model.

To improve tracking and treatment of COVID-19, Santosh also recommends using more than one data type to support decision making. Since COVID-19 has its genetic material stored in the form of RNA, AI tools can employ RNA sequences as a data set. CT scans are also used to detect COVID-19 infection and can be another source of data.

“AI-driven tools must be able to harness all possible types of data, not just be limited to one type,” Santosh said.

Press Contact
Hanna DeLange
Contact Email
Contact Website website