“These computers can sometimes pick out signals that the human ears can’t quite pick out,” says Dr. Wilson.
“If you do have a cough, don’t trust an app; get a real test. You don’t want to be accidentally exposing people,” he says.
On the flip side he does see an advantage in the use of artificial intelligence.
“If the app says ‘oh yeah we think you have COVID,’ well then maybe that encourages you to go and get that definitive test so you can be sure. I think there is some space for this, I just think we just have to be careful on how we use it,” says Dr. Wilson.
The bigger picture for the benefits of smartphone apps is their value in being part of the diagnosis process.
Doctor Wilson points out that COVID testing is not readily available all the time.
He explains in basic terms how artificial intelligence works.
“The computer makes its best guess for each cough. Then it learns, ‘was I right or was I wrong?
Did this patient have COVID or not?’ When it’s wrong, it modifies its strategy. That’s the sort of learning part of machine learning.”
Dr. Wilson says the computer then modifies its strategy trying to get a little more accurate. He says the process repeats and gets more accurate in its outcomes over time.
MIT’s algorithm was trained with 200,000 forced cough samples. It accurately identified almost 99% of coughs from people positive for COVID-19 and 100% of coughs from asymptomatic patients.
The app has not been approved for use at this time.