Study: AI fails to diagnose COVID-19 from coughs

Machine learning algorithms do a poor job of diagnosing whether people have COVID-19 by the sound of their coughs, according to researchers.

A Massachusetts Institute of Technology (MIT) team had found that AI-based systems correctly identified 98.5 percent of coughs from people who were confirmed to have COVID-19, including 100 percent of coughs from people with no symptoms.

However, Alan Turing Institute and Royal Statistical Society researchers found that even the most accurate machine learning-powered cough-detecting models performed worse than a model based on user-reported systems and demographic data, such as age and gender. 

The researchers collected and examined a dataset of audio recordings from over 67,000 people recruited from the National Health Service's Test and Trace and REACT-1 programs, which asked a random portion of the population to perform and send back nose and throat swabs to test for COVID-19. 

Commentators pointed out that the UK Department of Health and Social Care had awarded contracts worth over GBP 100,000 in 2021 to Fujitsu to develop the government's so-called "Cough in a Box" initiative, an AI-powered app funded by to collect and analyse audio recordings of COVID-19 symptoms.